WO2021226731A1 - Method for imitating human memory to realize universal machine intelligence - Google Patents

Method for imitating human memory to realize universal machine intelligence Download PDF

Info

Publication number
WO2021226731A1
WO2021226731A1 PCT/CN2020/000109 CN2020000109W WO2021226731A1 WO 2021226731 A1 WO2021226731 A1 WO 2021226731A1 CN 2020000109 W CN2020000109 W CN 2020000109W WO 2021226731 A1 WO2021226731 A1 WO 2021226731A1
Authority
WO
WIPO (PCT)
Prior art keywords
memory
information
machine
activation
features
Prior art date
Application number
PCT/CN2020/000109
Other languages
French (fr)
Chinese (zh)
Inventor
陈永聪
曾婷
陈星月
Original Assignee
陈永聪
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by 陈永聪 filed Critical 陈永聪
Publication of WO2021226731A1 publication Critical patent/WO2021226731A1/en

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/049—Temporal neural networks, e.g. delay elements, oscillating neurons or pulsed inputs
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods

Definitions

  • the application of the present invention relates to the field of artificial intelligence, in particular to the field of establishing general artificial intelligence similar to human intelligence.
  • the current artificial intelligence is usually designed for specific tasks, and there is no general artificial intelligence that can complete a variety of uncertain tasks.
  • Current artificial intelligence usually finds the mapping relationship from a large amount of labeled data. They cannot infer the cause, predict the result, and make choices and responses from the input information. Therefore, the current machine intelligence and human intelligence are very different.
  • the application of the present invention further proposes methods and steps for establishing artificial intelligence similar to human thinking, emotion and personality.
  • the brain has the ability to extract features of input information at multiple resolutions, and its purpose is to establish connections between input information at different resolutions.
  • the brain has the ability to associate. It can predict the causes and possible results of input information based on past experience.
  • the brain has the ability to generalize, which is to apply past experience to different objects.
  • there is a need and emotional system in the brain This is when the brain creates a variety of possible responses under the stimulation of input information, and selects the response that meets the expectations of the brain.
  • the brain has the ability to imitate. The brain imitates multiple past experiences and combines these experiences with real-life information through generalization capabilities to perform imitation, and constantly adjusts according to the actual situation during the output process.
  • Fig. 1 is the main part of the implementation of general artificial intelligence proposed in the application of the present invention.
  • S1 is the part that realizes the feature extraction of multi-resolution data. Through the S1 part, the machine turns the input information into multiple inputs, each of which is a corresponding data feature at a different resolution.
  • S2 is the part that realizes the ability of Lenovo to activate. The S2 part is mainly based on three methods: when the machine searches for relevant experiences in memory, the methods used are the “proximity activation” principle, the “similar activation” principle and the “strong memory activation” principle. Among them, “proximity activation” means that after the specific information in the memory is activated, it can only activate the information near it.
  • Similar activation refers to specific features in memory. When receiving activation signals from other features, the receiving ability is positively correlated with the similarity between them. “Strong memory activation” refers to the higher the memory value, the stronger the ability to receive activation signals from other characteristics. On the basis of these three principles, the machine can realize the association ability similar to the human brain. S3 realizes the part of generalization capability. In the application of the present invention, the empirical generalization method we propose is to find the common parts of these similar processes by reducing the resolution in multiple similar processes, and use these common parts as process features. When reconstructing the specific process, as long as the object in reality meets all the features required by the object in the process feature at a specific resolution, then it can be replaced.
  • S4 is the needs and assessment part.
  • the most critical foundation in this part is that when the machine builds a network of relationships, it stores the changes in needs and emotions with the information that caused these changes. Through repeated repetition of similar situations, specific information will be more closely related to specific needs and emotions.
  • the information stored by the machine includes external input information, internal input information, the machine's own needs/emotions and the relationship between these information, and the strength of these connections is quantified through the memory and forgetting mechanism. With these relationships, the machine uses the memory network to find and input information-related experience using the ability of association. The machine uses generalization ability to generalize these experiences to the input information.
  • the machine combines these generalized fragments to imitate similar organizational methods in the past experience and joins them together as a planned response, and then evaluates whether the results of the planned response meet their expectations under the principle of "seeking advantages and avoiding disadvantages". If it meets, output the planned response. If it does not match, select the response again.
  • step S5 the machine expands the concept (intermediate target link) in the plan response through segmented imitation (more detailed intermediate target link), and expands it to the underlying experience that can be specifically executed in memory, and according to time and Space division to perform these responses.
  • the thinking process of the entire machine is iterative.
  • the way the machine processes new information each time is to turn the current goal into an "inherited goal.”
  • the machine performs multi-resolution feature extraction on the new information.
  • In the relationship network through the ability of association, find the experience related to the "inheritance goal” and "new information”.
  • these experiences are generalized to the input information.
  • the generalized experience fragments are combined into a possible response plan according to the time and space relationship. Then based on past experience, under the principle of "seeking advantages and avoiding disadvantages", evaluate the possible impact of the response plan on oneself.
  • the response plan is re-established; if it is passed, the response plan is taken as the output plan, and each link in the plan is expanded to more specific intermediate links by means of segmented imitation.
  • This process is also iterative until it reaches the bottom experience that the machine can execute immediately.
  • This process is a process of thinking while doing. In this process, once new information is input, the machine returns to "turn the current target into the inherited target, and perform multi-resolution feature extraction of the new information". Go in the process. Therefore, in the application of the present invention, the machine only needs to use very simple steps to iterate repeatedly to achieve a human-like thinking process and human needs and emotional response.
  • Figure 1 is the main components of the present application.
  • Figure 2 is a method to achieve multi-resolution feature extraction.
  • Figure 3 is a schematic diagram of a functional module organization.
  • Figure 4 is a schematic diagram of another functional module organization.
  • the brain already has the ability to use multi-resolution to extract features of input information (assuming this is the ability brought by genes).
  • the preprocessing part of the information first extracts the information features at different resolutions from 1 to K1 (natural number) layers, and each layer of resolution may have multiple corresponding features. Then the brain starts searching for similar features in the memory. It is assumed that the basic search method of the brain is to activate the features that need to be searched, so that it emits a specific pattern of activation electrical signals. This electrical signal can propagate in the memory space and will attenuate with the propagation distance.
  • the brain when a set of "dining table" features are input to our brain, the brain first transforms them into information features at multi-resolution (for example, at the coarsest resolution, it is an overall three-dimensional image; then it is at a finer resolution.
  • the shape of the desktop and the outline of the table legs under the lower resolution; then the desktop texture at a finer resolution, other details of the edge contour, etc.) and then the features at each resolution emit the activation electrical signals corresponding to these features in turn (assuming There is a kind of nerve tissue that can send out its activation electrical signals to the corresponding mode under the excitation of different input information characteristics).
  • these activating electrical signals can only activate our nearest memory (because the distance in the memory space is short, the attenuation is small, we call it proximity activation) and memories similar to the input information (because the pattern matches, they have good receiving ability, we Call it similarity activation), and memories that impress us (because these memories have more neurons or synapses, and their ability to receive them is better).
  • the "dining table” we may recall a pack of snacks that we put on the dining table yesterday, or recall the scene of making handicraft with our mother at the dining table after an hour.
  • it implies the assumption that our memory is stored in chronological order, and it implies the assumption that the information stored in the memory of each time period is stored in a manner similar to reality.
  • the "dining table” signal activates the scene of "I and my mother are doing handwork at the table” after our hours, it may also activate the memory of "the window glass was suddenly smashed by a leather ball” (although this message is related to the dining table).
  • the distance is far, but it has a high memory value.
  • the time sequence of information storage is in the order of attention, because they are the time sequence of information input), because this scene has more neurons or synapses. Storing them also has a stronger ability to receive electrical signals (we call it strong memory activation). Since our brain stores our emotions while storing memories, those strong emotions are also deeply memorable, and they also have more neurons or synapses to store them, so we can still recall our emotions from "happy”.
  • the brain has the ability to predict on the basis of association and "seeking advantages and avoiding disadvantages". Prediction is the ability that the brain uses all the time. Its foundation is the network of relationships between things and concepts established by the brain through the mechanism of memory and forgetting.
  • the relationship network mainly accomplishes 2 things: 1. Increase memory for information that needs to be increased memory (use more neurons or synapses to remember). 2. Establish the strength of the connection of each information to needs and emotions. Once the relationship network has established these two relationships, when new information is input, we can find similar memories through the Lenovo activation system. And by sequentially activating similar memories in the two directions of the time dimension and the similarity dimension, the causes and results of these similar memories can be found.
  • the brain can use an analogy method to generalize past experience (cause or result) to input information, thereby generating predictions for input information, so that it can make its own choices based on needs and emotional systems.
  • Segmented imitation is the concrete implementation of the brain’s response plan: the main concepts (intermediate links) in the planned response are expanded layer by layer (more detailed intermediate links) through segmented imitation, to the underlying experience that can be specifically implemented in memory , And execute these responses according to time and space division.
  • the input data is divided into multiple channels through a filter.
  • these channels include specific filtering for the contour, texture, tone, and dynamic mode of the graphic.
  • these channels include filtering for speech recognition such as audio composition and pitch change (a dynamic mode).
  • These pre-processing methods can be the same as the existing image and voice pre-processing methods in the industry (for example, wavelet transform is a better method to achieve multi-resolution feature extraction), and will not be repeated here.
  • the machine can also directly input the pre-processed data as multi-resolution data, and find the local similarity among them as features.
  • the windows of different resolutions can be time windows or spatial windows.
  • windows of different sizes are used to find local similarity.
  • the data is selected using windows of different sizes, which mimics the human attention interval.
  • the data in the large window corresponds to the use of low resolution
  • the data in the small window corresponds to the use of high resolution.
  • the specific steps can be as follows: the machine can successively use the partial windows W1, W2, W3,..., Wn, where W1 ⁇ ... is a natural number) to compare all window data under all inputs and find Local similarities are repeated as features.
  • the similarity comparison algorithm can be used. Because the similarity comparison algorithm is a very mature algorithm, professionals in the industry can implement it based on public knowledge, so I won't repeat it here.
  • the machine puts the found local similar features into a temporary memory bank. Every time a new local feature is added, its initial memory value is assigned. Every time an existing local feature is found, the memory value of the underlying feature in the temporary memory bank is increased according to the memory curve.
  • the information in the temporary memory bank complies with the memory and forgetting mechanism of the temporary memory bank. Those low-level features that survived in the temporary memory bank, after reaching the threshold of entering the long-term memory bank, can be put into the feature library as long-term memory features. There can be multiple long-term memory banks, and they also follow their own memory and forgetting mechanisms.
  • the machine not only needs to build the underlying feature map database, but also needs to build a model that can extract these underlying features.
  • the underlying feature extraction algorithm established by the machine One of the possible algorithms is to find the similarity comparison algorithm A in the local similarity.
  • the machine uses a method of preprocessing the information (for example, after various coordinate base transformations, removing or compressing some of the base coefficients), the machine uses large windows (low resolution) and small windows (high resolution) To extract the data features in the window.
  • Using a window is equivalent to imitating human attention, so that we can obtain multi-resolution features while extracting the position of the data feature in the input. These positions will be used to reconstruct the mirror space of the input information using features. And this storage method of mirror space is the basis for us to realize the principle of proximity activation.
  • neural network algorithm B Another algorithm for extracting underlying features is neural network algorithm B. It is an algorithm model based on a multilayer neural network. After this model is trained, it is more efficient than the similarity algorithm.
  • the machine uses the selected information features as possible outputs to train a multilayer neural network. We can use a layer-by-layer training method.
  • S204 the machine successively uses local windows W1, W2,..., Wn, where W1 ⁇ ... is a natural number) to train the algorithm model.
  • W1, W2,..., Wn, where W1 ⁇ ... is a natural number In the optimization, one is to increase the neural network layer from zero to L (L is a natural number) layer on the corresponding previous network model every time the window size is increased.
  • S205 when optimizing the neural network with the added layer, there are two options: 1.
  • Each neural network model corresponds to a resolution.
  • the machine needs to select one or more neural networks according to the purpose of extracting information this time. Therefore, the machine may obtain two kinds of neural networks for extracting information features.
  • One is a single algorithm network with multiple output layers. Its advantage is that it requires less computing resources, but its ability to extract features is not as good as the latter.
  • the other is multiple single-output neural networks.
  • This method requires a large amount of calculation, but the feature extraction is better. It should be pointed out that the above method can process images and voices, and can also process information from any other sensors in a similar way. It should also be pointed out that choosing different resolutions means choosing different windows and different feature extraction algorithms. So the extracted feature size is also different. Some underlying features may be as large as the entire image.
  • a training method of a multilayer neural network is proposed: a multi-resolution training method.
  • the multi-resolution training method refers to decomposing the input information into different resolution layers. Then use the partial resolution layer to train the neural network. For example, first use low-resolution information data to train the neural network, and then gradually increase the resolution to train the neural network. When the required accuracy is reached, the information of other resolution layers can be discarded. Of course, the order of using the resolution layers can also be adjusted according to the purpose of identification.
  • the machine can also train multi-layer neural networks separately according to the input information characteristics at different resolutions, group them according to the resolution, and then weight the outputs of multiple neural networks as the total output.
  • the static feature map is established based on the resolution, which represents the machine's self-built classification of things based on similarity. For example, two tables may belong to the same category at a rough resolution, but they may belong to different categories at a fine resolution.
  • the machine only needs to extract the features of the input information according to different resolutions, and take these features as a whole to represent the input information.
  • the input information is compared with the information in the memory, the machine uses different resolutions to make the comparison. For example, the same two things are composed of multiple resolution feature maps. When comparing their similarity, you only need to compare at different resolutions to quantify the similarity between the two.
  • the present application proposes a dynamic local similarity comparison method.
  • windows of different sizes are used to track different parts of things. For example, if a person runs over, walks over or slides over, we can use different windows to represent different resolutions. For example, when we use a large window to treat the whole person as a whole, we track the movement pattern of this window, and we find that the movement patterns are the same in these three cases. But when we use a smaller window to extract the human hands, legs, head, waist, buttocks and other parts of the movement mode separately, we distinguish the difference of these three movement modes. Furthermore, if we use more windows to focus on the movement pattern of the hand, we can get a finer resolution movement pattern.
  • the machine In addition to the spatial resolution, the machine also needs to establish different temporal resolutions. For example, we describe the constant flow of people on the street, which is a mode of crowd movement. But from a more subtle time resolution, we can find morning and evening shifts. Crowd flow peaks in time. We compare the changes of the motion trajectory at different time resolutions to get the rate of change. The rate of change is an important dynamic feature of movement in time. Therefore, the extraction of motion patterns is based on a certain time resolution and a certain spatial resolution. The machine processes a large amount of dynamic data to find common dynamic features.
  • the machine Every time the machine finds a similar movement pattern, the machine puts the data representing this movement pattern into the temporary memory bank as a candidate for the dynamic feature map, and assigns a memory value to the candidate for the dynamic feature map.
  • the machine uses windows of different sizes and iteratively uses the above process on the data, so that the machine can obtain a large number of dynamic feature map candidates in the temporary memory bank.
  • the machine Like the static feature map, the machine also uses the memory and forgetting mechanism to survive the fittest on the extracted dynamic feature map. Those movement patterns that are widely present in various moving objects will be discovered again and again, thereby increasing the memory value again and again, and finally entering the long-term memory bank and becoming our long-term memory.
  • the memory value of this feature map candidate increases its memory value according to the memory curve.
  • all memory values in the temporary memory bank follow the forgetting curve and gradually decrease over time. If the memory value decreases to zero, then the feature map candidate is deleted from the temporary memory bank. If the memory value of a feature map increases to the preset standard, then this feature map is moved to the long-term memory bank and becomes a long-term memory.
  • the memory value represents the time that the corresponding feature map can exist in the database. The larger the memory value, the longer the existence time. When the memory value is zero, the corresponding feature map is deleted from the memory bank.
  • the increase or decrease of the memory value is carried out in accordance with the memory curve and the forgetting curve. And different databases can have different memory and forgetting curves.
  • the above process is constantly used, and finally a large number of feature maps are obtained.
  • These feature maps can be put into the fast search memory bank.
  • the key to the realization of association lies in the establishment of a memory network and the use of three activation principles: the proximity activation principle, the similar activation principle, and the strong memory activation principle.
  • the machine remembers the information according to the time sequence of input. So if the machine's attention interval switches back and forth between two spatial positions, then the adjacent space in the memory is not the actual adjacent position in space, but the two constantly switching spatial positions in the memory, because they are in the order of adjacent time. Placed.
  • the "strong memory activation" of the machine is accomplished through the memory and forgetting mechanism.
  • These feature maps are stored in memory over time. At the same time, these multi-resolution feature maps will in turn send out their own activation signals in the memory space. It activates not only nearby memories, but also memories similar to it. It needs to be emphasized that each extracted feature will emit its own activation signal.
  • the same thing, scene, and process may emit multiple activation signals corresponding to the resolution characteristics at different resolutions.
  • Those features that receive the activation signal and are activated because they are activated once, their memory value increases according to the memory curve.
  • the memory value of all the features in the memory interval decreases according to the forgetting curve of the respective memory bank. In this way, the memory value of those features that are repeatedly activated will increase with the number of activations, so that the activation signal will be stronger, which will form a positive feedback and increase your own memory. Therefore, those memories with high memory value are either repeatedly activated and gradually increase the memory value, or they are given enough memory value at one time so that the information is memorized.
  • the first category is the information characteristics of external input, including the characteristics of all external sensor input information. They include visual, auditory, smell, touch, taste, temperature, humidity, air pressure and other information. These information are closely related to the specific environment. They are based on the original The organization method of data storage can reconstruct the three-dimensional mirror space; they maintain their memory value according to the memory and forgetting mechanism.
  • the second category is internal self-information, including power, gravity direction, body posture, operation of various functional modules, etc. These information have nothing to do with the environment, and their memory values are set according to a preset program.
  • the third category is data on the state of machine needs and needs, including data such as safety value, dangerous value, profit value, loss value, goal achievement value, dominance value, and own body state evaluation value; it also includes data related to these needs and needs. Status data.
  • the machine also generates various emotions based on the satisfaction of its own needs. The relationship between these emotions and the situation where one's own needs are met is set through a preset program.
  • the machine can also reversely use the relationship between internal conditions, external conditions and the state in which its own needs are met to adjust the preset program parameters of emotion generation, thereby using its own emotions to influence the outside world.
  • the method we adopted is to establish different symbolic representations of the machine's own demand type and emotional type.
  • the machine When an event occurs in the mirror space of the machine, the machine needs to store the current mirror space in the memory.
  • the machine stores all feature maps (including feature maps, demand symbols, and emotional symbols) and their initial memory values (positively correlated with the activation value when the storage occurs, but not necessarily linear) in memory.
  • the requirements of the machine can be varied, and each type of requirement can be represented by a symbol. Such as safety and danger, gains and losses, dominance and dominance, respect and neglect, etc. The difference and amount of the demand types do not affect the claims of the present application. Because in the present application, all requirements are handled in the same way.
  • the emotions of the machine can be varied, and each type of emotion can be represented by a symbol. For example, such as excitement, anger, sadness, tension, anxiety, embarrassment, boredom, calmness, confusion, disgust, pain, ashamedy, fear, happiness, romance, sadness, sympathy and satisfaction.
  • the difference and amount of emotion types do not affect the claims of the present application. Because in the present application, all emotions are handled in the same way.
  • the initial activation value assigned by the machine to the input information will also be propagated to the machine's needs and emotional data through the relationship network, resulting in the machine's instinctive response to this information.
  • the demand and emotional data of machines are a very important type of "anthropomorphic" data. It is closely related to external input information and one's own internal information. Their relationship is: when external data or internal data is input, the machine will respond, and these responses will get external feedback and change the internal state (for example, the battery becomes less).
  • we give the machine a need type similar to that of a human and a demand gain value that represents the situation in which the demand is satisfied.
  • the specific implementation method can be: in the process of training the machine, humans use preset symbols (such as language, action or eye contact) to tell the machine which environments are safe and those environments are dangerous, or can tell the machine further Different grades of machines. Just like training a child, just tell it "very dangerous”, “more dangerous” and “a little dangerous”. In this way, the machine can gradually increase the connection strength between the dangerous environment or the common features in the process and the built-in demand symbol of danger through training, memory and forgetting (because of the increased number of repetitions).
  • preset symbols such as language, action or eye contact
  • the machine processes the input information next time, after giving the input information the same initial activation value, the activation value of some features is closely connected with the danger symbol, and it transmits a large activation value to the danger symbol.
  • the machine is immediately aware of the danger and will immediately process this dangerous information based on its own experience (which can be preset experience or self-summed experience).
  • its own experience which can be preset experience or self-summed experience.
  • humans already have a lot of experience to pass on, during training we can also directly tell the machine how dangerous those specific things or processes are. This is a way to preset experience for the machine.
  • the preset experience can use language to allow the machine to establish a memory frame to connect the dangerous factors with the danger, or it can be realized by directly modifying the existing relationship network of the machine (modifying the memory value of the danger symbol in the corresponding memory frame).
  • the two values of safety and danger tell the machine how to identify safety and danger factors, so as to learn how to protect itself.
  • the gain value and loss value tell the machine which behaviors we encourage and which behaviors will be punished. This is a reward and punishment system. Just like training children, we only need to reward or punish them after they perform certain behaviors. Or when rewards and punishments happen, just tell them why.
  • we can also preset experience (such as telling it in advance that those behaviors will be rewarded and those will be punished, or directly modify its brain neural connections to achieve the goal.
  • the loss of profit system focuses on the result of the behavior, and the domination and dominance focus on the scope of the behavior. It is used with the loss of profit system.
  • the same training method. We can also associate the machine’s own body state evaluation and needs with emotions and external input information, the purpose is to let the machine understand its own body state evaluation value and the relationship between them. For example, on a rainy day, if the machine finds If its own power or other performance is declining rapidly, it stores these memories. If the same situation is repeated many times, the machine will establish a closer connection between the performance degradation and the rain.
  • the emotion of the machine is an important way for the machine to communicate with human beings. Therefore, in the application of the present invention, we also take the emotion of the machine into consideration.
  • Human emotional response is an innate response to whether one's own needs are met, but through acquired learning, we have gradually learned to adjust this response, control this response, and even hide this response.
  • preset programs to link the emotions of the machine with whether the needs of the machine are met. For example, when a danger is identified, the emotions of the machine are "worry", “fear” and “fear”, depending on the degree of danger. For example, the various internal operating parameters of the machine are in the correct range, which gives the machine emotions such as "comfort” and "relaxation".
  • the machine's expression may be "uncomfortable” and "worry". Therefore, using this method, we can assign all the emotions that humans have to the machine.
  • the emotion itself is expressed through the facial expressions and body language of the machine.
  • these instinctive emotions of the machine will be adjusted by the reward and punishment mechanism.
  • the trainer will continue to tell the machine its emotional performance, which ones are rewarded, and which ones are punished. You can also directly tell it what the appropriate emotion is in a particular or process. Of course, you can directly modify its neural network connection to adjust its emotional response.
  • the machine can adjust emotions to a degree similar to that of humans, and further, because emotions and other memories are stored together, in the same memory.
  • a machine needs a certain result, it will imitate the memory that brought that result.
  • a certain type of behavior brings a certain result that can be repeated, then the machine will imitate the memory that contains this type of behavior, and of course it will also imitate the emotions in these memories, so it will adjust its emotions for a certain purpose. This is a way of using emotions.
  • the machine uses a memory screening mechanism for the storage of the mirror space: an event-driven mechanism and a temporary memory library mechanism.
  • the machine takes a snapshot of the mirror space and saves it.
  • the saved content includes the features in the mirror space (including information, machine states, needs, and emotions) and their memory values. Their memory value is positively related to the activation value when the storage occurs, but not necessarily linear.
  • a snapshot of the mirror space stores data, which we call a memory frame. They are like movie frames. Through continuous playback of multiple frames, we can reproduce the dynamic scene when the memory occurs. The difference is that the information in the memory frame may be forgotten over time.
  • An event in the mirror space means that the feature combination in the mirror space is compared with the previous mirror space, and the similarity changes beyond the preset value, or the memory value in the mirror space changes beyond the preset value.
  • Memory bank refers to the database that stores these memory frames.
  • the temporary memory bank is a kind of memory bank, and its purpose is to filter the information stored in the memory frame. In the temporary memory bank, if a memory frame contains features whose memory value reaches the preset standard, then this memory frame can be moved to the long-term memory bank for storage.
  • we use a limited-capacity stack to limit the size of the temporary memory bank, and use the fast memory and fast forgetting methods in the temporary memory bank to screen the materials to be put into the long-term memory bank.
  • the machine When the machine is faced with a large amount of input information, those things, scenes and processes that are already accustomed to, or things, scenes and processes far away from the focus of attention, the machine lacks the motivation for in-depth analysis of them, so the machine may not recognize these data, or The activation value assigned to them is very low.
  • the memory value assigned by the machine to each information feature is positively correlated with the activation value when the storage occurs.
  • Those memories with low memory value may soon be forgotten from the temporary memory bank and will not enter the long-term memory bank. In this way, we only need to put the information that we care about into the long-term memory, instead of memorizing the trivial things that do not need to extract the connection relationship every day.
  • the capacity of the temporary memory bank is limited, the temporary memory bank will passively accelerate the forgetting speed because the stack capacity is close to saturation.
  • the tightest local connection relationship constitutes the basic concept (including static feature map and its language, dynamic feature map and its language); a bit looser than the basic concept is the static expansion concept and the dynamic concept expansion concept (including the representative relationship Concept and process characteristic diagram), looser than concept is memory.
  • those static feature maps (or concepts) are usually small parts
  • those dynamic feature maps (including concepts that represent relationships) are connectors
  • those process features are large frames, which are multiple small parts (static Objects) and connectors (dynamic features) are organized according to a certain time and space sequence. Process characteristics are a large framework that we can learn from.
  • the dynamic feature map (including the concept that represents the relationship) is a tool that can implement empirical generalization, and the static feature map (or concept) is the object to be replaced in the generalization.
  • the generalization process is a process of reorganizing and implementing imitation after replacing the real objects and the objects in memory through the information flow with high activation value.
  • connection value is a function of the memory value of the feature maps at both ends of each connection line. Then normalize the connection value sent by each feature map. This will cause the connection values between the two feature maps to be non-symmetrical. Then the similar feature maps between the memory frames are connected according to the degree of similarity, and the connection value is the similarity. After passing the above steps, the obtained network is the cognitive network extracted from the memory bank.
  • the cognitive network alone in a quick search library (a kind of memory library) for some instinctive responses that require fast, such as in autonomous driving applications, or in some simple smart applications (such as production lines) .
  • the memory and forgetting in this kind of relationship network adopts the mechanism of remembering and forgetting the connection value: each time the relationship is used, the connection value increases according to the memory curve. And all the connected values decrease with time according to the forgetting curve.
  • Another method is to put the memory containing frequently called concepts and process features, maintaining the organizational form of the memory bank, into a separate quick search library. In these memories, specific details may have been forgotten, and what remains are memories with strong memory values. These strong memory value memories are activated by association, and related concepts and process features can be quickly recalled. This can speed up the memory search efficiency of the machine.
  • This method can also be used in applications that require rapid response, such as autonomous driving applications, or in some simple smart applications (such as production lines).
  • Method 1 Use the memory value (real number) to represent the number of neurons or synapses; use the activation value to represent the strength of the activation electrical signal emitted by the feature; use a specific code to represent the different mode activation signals emitted by different features; use the bus Instead of the entire memory space to propagate the activation value; use the three-dimensional coordinate point position to represent the position of different feature information in the memory space, and use the spatial distance (the spatial distance between the activation source and the receiving feature) to calculate the attenuation.
  • the input feature releases its own corresponding coded activation electrical signal to the bus through the universal excitation module, and uses the number in the code to represent its initial strength
  • the feature in memory can be read by periodically reading the bus information.
  • the bus To receive the information on the bus, and calculate the amount of attenuation that should be. If there is activation information similar to yourself, for example, it may belong to a large category or a subcategory, etc., then there are different receiving capabilities. If the activation value obtained after the received activation signal passes through its own receiving channel exceeds its preset activation threshold, then this feature uses the received activation value as the initial value and activates itself. Usually, there may be multiple input features simultaneously activating a small memory interval. For example, a "dining table" has multiple features with different resolutions, and they pass through the bus of the memory interval in turn, possibly activating multiple small intervals. Each section may have multiple features related to the "dining table" activated.
  • the feature maps concentrated in these cells are activated again, they are given activation values to each other through adjacent activations. Therefore, their activation values may be "highlighted” in the memory space. And under their common proximity activation, a certain neighborhood may activate the memory of a "delicious" cake on the table at that time. This is because the cake gives a high activation value to the food-related "positive demand” symbol through the preset program related to the taste sensor.
  • the activation value of the food-related "positive demand” symbol is converted into a memory value according to a positive correlation (not necessarily a linear relationship). So, here, the symbol of "positive demand” related to food (such as the demand for delicious food) is a strong memory. It exists near the memory of the "dining table".
  • This set of preset programs will send out higher directional activation values to emotional symbols such as “pleasure” and “satisfaction” under the stimulation of the input of "demand for deliciousness is satisfied”.
  • emotional symbols such as “pleasure” and “satisfaction” of the machine have obtained higher activation values.
  • these activation values are also converted into memory values (not necessarily linear) in a positive correlation manner, so in these memories, emotions are also memorized.
  • these emotional symbols may also be activated, so that the machine can experience emotions such as “pleasure” and "satisfaction”.
  • the machine When the machine needs to seek emotions such as “pleasure” and “satisfaction” (for example, to give the machine such an instinctive need), the machine will look for memories related to “pleasure” and “satisfaction”, and it may activate the "cake", “Dining table” and other memories. These memories may become a response target, and the machine may obtain the experience of "cake” and “dining table” through these target associations, and then generalize these experiences through generalization ability, and under existing conditions, by imitating past experience, After the generalization, various process characteristics are organized, and through segmented imitation, the organized process is subdivided into a large number of intermediate link goals, and then these intermediate link goals are achieved step by step. For example, to complete the process of ordering "cakes", finding "tables” and satisfying their own needs.
  • the above process is a process of distributed computing.
  • This method can also be turned into a two-layer structure.
  • a computing module connected to the bus is placed as a gateway for information exchange with the bus.
  • This computing module is responsible for identifying the activation signal outside the jurisdiction and deciding whether to transfer it to the jurisdiction. It is also responsible for transferring the activation in the jurisdiction to the bus again. The purpose of this is to reduce the number of computing modules.
  • this structure can also iterate itself, using a similar multi-layer structure to further reduce the calculation module.
  • Method 2 is a centralized calculation method. It uses a special calculation module to search memory (memory search module). Whenever a feature of input information under multi-resolution is found, the machine directly activates the most recent memory in the current time and assigns its corresponding activation value according to their memory value. This completes the proximity activation and strong memory activation. It also directly searches for related similar features in the memory, and after finding, directly assigns activation values to these features according to the similarity. The degree of similarity can either use the method of on-site comparison or the method of precoding layer by layer classification.
  • the memory search module can use the same method when the activated feature map sends out the activation signal again.
  • searching for nearby memories initiates proximity activation, searching for those further away with high memory value initiates strong memory activation, and searches for similar features in other memories to initiate similarity activation.
  • the activation signal sent out has its own code and intensity information. This process can be iterated over and over again.
  • Method 3 is a hybrid mode. After the machine completes the similarity activation search through the memory search module, further activation can be carried out in the local network of each memory. Proximity activation and strong memory activation are realized through the connection network established between the features in the memory.
  • Proximity activation and strong memory activation are realized through the connection network established between the features in the memory.
  • One way to realize this kind of local network is: each feature in the memory space establishes a connection nerve with the neighboring feature, and when it is activated, the activation value can be transmitted through these connection lines, which is adjacent activation.
  • the transfer coefficient between the two features is positively correlated with the memory value of the two features, which is the strong memory activation.
  • One processing method includes: (1) The machine memorizes feature maps of various angles.
  • the feature map in memory is a simplified map created by extracting the underlying features of each input information. They are the common features of similar things retained under the relationship extraction mechanism. Although they are similar to each other, they may have different viewing angles.
  • the machine memorizes the feature maps of the same thing in life but from different angles to form different feature maps, but they can belong to the same concept through learning.
  • (2) The machine uses views from all angles, overlaps the common parts of these feature maps, imitates their original data, and combines them to form a three-dimensional feature map.
  • the machine searches for similar underlying features in the memory, it includes searching for a feature map that can be matched after spatial rotation in the memory. At the same time, the machine saves the feature map of the current angle in memory, keeping the original angle of view. When the underlying features with similar perspectives are input again later, they can be quickly searched. Therefore, in this method, the machine uses a combination of different perspective memory and spatial angle rotation to find similar feature maps, which will bring us to the phenomenon of faster recognition of familiar perspectives. Of course, the machine can also only use the method of comparing the similarity after rotating the space angle.
  • the generalization ability is based on the concept of multi-resolution, so the machine needs to establish the concept of different resolutions first.
  • the concrete method of the machine to establish the concept adopts the same way as the human. For example, when a certain image feature map is input into the machine, we give it a language that represents the image feature map simultaneously. The two concerned information is memorized as adjacent information, and they have a close relationship. Then the machine can establish a very close relationship between this image feature map and the corresponding language feature map in the relationship network after repeated repetitions. They can activate each other, generate associations, and can exist in the relationship through similarity activation. Language or other forms of information in different memory intervals are conceptual associations.
  • the expansion of the dynamic concept also establishes a new dynamic feature classification by increasing or reducing the resolution. For example, “running” and “dancing” are collectively referred to as “sports”, and “running” is divided into “fast running”, “jogging” and “long-distance running”. These are also new categories established through different attributes linked by dynamic features, and new language symbols are created to represent these categories.
  • the concept of expansion is a type of language symbols in our memory. They exist by establishing a closer connection with other language symbols, and their content is based on the common characteristics of the content represented by other language symbols. Reflected.
  • a large number of extended concepts have been summarized in the development of human history. In our lives today, most of the concepts are obtained through learning (directly obtained from the summary results of the predecessors), and a small number of concepts are established through the close connection of similar things in memory with a certain language symbol.
  • the same method can also be used to learn. For example, many human concepts are learned through interpretation. For example, when a scene or feeling of "happiness" appears, we are told that this is "happiness". For another example, we may learn the explanation of the concept of "solar system” through a dictionary. The same can be said for machine learning.
  • Method 1 Let the machine learn directly. For example, let the machine learn the meaning of a concept through text and voice. Put this information into memory, and the machine can establish similar activation, proximity activation, and strong memory activation connections between these concepts and their interpretation through repeated learning.
  • Method 2 Directly establish a "fake” machine memory, in which artificial connections between similar activation, proximity activation, and strong memory activation are assigned to the machine (such as assigning high memory value to related information, and memory locations are put together, according to the information Encoding to help the machine find similar features faster, etc.).
  • the machine when inputting the language symbols of these extended concepts, the machine can associate the basic concepts it contains, and then associate it with other forms of information (such as images, sounds, smells, touch, emotions, feelings, etc.). And using these forms of information to join the information flow formed by the language symbols, the machine can find similar information flows through association, and use similar information flows in the past to speculate on the causes and possible results of the information flow. And put these causal relationships into your own needs and emotion evaluation system to determine your own response. This is intelligence.
  • other forms of information such as images, sounds, smells, touch, emotions, feelings, etc.
  • the tools of generalization are mainly the concept of expressing action characteristics. Because the dynamic feature is a dynamic way of movement, its subject is a generalized subject. The machine can use mass points or three-dimensional graphics to represent abstract moving subjects. It is precisely because the subject of motion is the subject of generalization, so that the machine can bring similar concepts into the characteristics of motion, thereby realizing the generalization ability of experience. And this kind of similar concepts can be viewed directly and not similar, but they can be summarized into the same kind of things by reducing the resolution.
  • the concept of expressing the relationship between things is also a dynamic feature. It considers the objects at both ends of the relationship as a virtual whole. Therefore, in the application of the present invention, by assigning a dynamic feature to the concept representing the relationship, the machine can correctly use the concept representing the relationship through this dynamic feature.
  • the relationships represented by languages such as “although...but", “but", “though", “but" can be represented by a dynamic feature of transition.
  • Parallel concepts such as “on one side... on the other side" and "both... and" can be represented by dynamic characteristics of parallel operations.
  • the relational concept of "contained in” can be expressed by the dynamic feature of inclusion.
  • the specific methods for establishing the dynamic characteristics of this relationship are: 1.
  • the machine uses memory and forgetting mechanisms for a large number of languages to find their common points. These common points are usually the concept of dynamic patterns or relationships, because they are related to specific objects. Irrelevance, leading to them can be widely used.
  • the organization of these words has gradually become common words, common sentence patterns, and grammar. This method is similar to the current method of language organization in artificial intelligence, and is a method of mechanical imitation.
  • the method of machine understanding is to memorize the specific static feature maps and dynamic feature maps associated with each use of these concepts, and then save these concepts through the memory and forgetting mechanism.
  • Process feature is an extended dynamic feature. Its features are: 1. Multiple observation objects, they are not necessarily a whole. 2. There is no clear repeating trajectory in the whole movement mode. For example, the processes of going home, going on business, washing hands, cooking, etc., are multiple physical concepts or expanded abstract concepts that constitute a generalized movement mode. It is called a pattern because these concepts can be repeated in our lives. Since it can be repeated, it means that there are common features in the process of representation of these concepts. Otherwise, it is impossible for us to use a concept to represent them.
  • Process characteristics are usually dynamic processes involving large space and long time. The specific details of its implementation are closely related to the environment, so it is difficult to find similarities. But these links are usually represented by language symbols. Therefore, when we look for a process feature, we can first look for the repetitiveness of the language symbols in each link. For example, each time the machine goes to the airport through memory, the language symbols corresponding to each link form a gradually unfolding tower-shaped conceptual relationship.
  • the top level of this concept is "going to the airport", the next level is “ready to go”, “on the way”, “arrival”, and the next level is “preparing luggage”, “finding a car”, “farewell to friends”, “By car”, “On the way”, “Arriving at the airport garage”, “Out of the garage”, “Arriving at the airport entrance”.
  • the next level is "Prepare clothes”, “Prepare toiletries”, “Prepare money”, “Prepare related materials”.... This process can be subdivided continuously.
  • the distinction of each link can be arbitrary. But every time we go to the airport, we get a tower-shaped conceptual organization.
  • This tower-shaped conceptual organization goes through a memory and forgetting mechanism, and finally at each resolution level, only a small amount, indispensable, and frequently appearing concepts can be retained in memory.
  • They are process characteristics at the corresponding resolution. These process characteristics are a series of concepts, organized in a temporal and spatial order. Especially on the ground floor, usually only static feature maps and dynamic feature maps that may be available every time you go to the airport can be left. These feature maps are few in number, but they are indispensable. These are static feature maps or dynamic feature maps that represent key links, such as "security check" or "boarding”. The upper-level concepts connected to the key links are also indispensable (they may be fewer in number). Push upwards one by one, and in the end there is only a top-level concept of "going to the airport". Therefore, the establishment of process characteristics is realized through the mechanism of memory and forgetting from positive selection (the link deliberately memorized by learning from other people's experience) and adverse selection (the upper link corresponding to something every time).
  • the machine can realize generalization in a way that the computer can understand.
  • the following is an example of the machine receiving the instruction "go to the airport” to illustrate the generalization process.
  • the machine receives the instruction of "go to the airport", through association, it activates the relevant information of the language symbol of "go” (may be action images and feelings), and also activates the relevant information of the language symbol of "airport” (may be some Airport image). If the machine has ever been to an airport, then these two activation points will directly activate a large number of other memories through “proximity activation”, “similar activation” and “strong memory activation”.
  • These memories may include static images, moving images, speech and text There are also feelings and emotions. They are all related to time and space in memory.
  • the machine only needs to take this string of information as the intermediate link goal in accordance with the time and space relationship, and gradually imitate and realize it, to achieve the goal of "going to the airport". If the machine has no experience in going to the airport, it may activate related images of "going to the train station", “going to the store", "going to travel”, and also related images of the "airport”. Through the comparison, the machine found that in these images, except for the “airport”, the activation value of the “train station” is the highest. This is because there are multiple activation channels to transfer activation values to the "train station”. For example, the experience of "going to" activates going to the "train station". For example, "airport” and "train station” are similar in rough resolution.
  • the input information of the machine can be connected to a series of underlying static or dynamic non-verbal information (including sound) through the connection relationship between the concepts.
  • Different resolutions can be used to compare similarity to achieve similarity activation, proximity activation and strong memory activation.
  • the activated non-verbal information will once again use the language concept as a bridge to activate other memories that are not similar, or that are nearby, or whose memory value is not very high.
  • the machine selects the information with the highest activation value from 1 to N (natural numbers) and organizes them according to their own time sequence or spatial sequence. They are the generalized experience that the machine needs to imitate. Therefore, the generalization ability does not need to be established deliberately by the machine. The machine only needs to establish the correct concept, and the relationship between the concept, through the association activation, the generalization experience that can be used will automatically emerge.
  • the two keys to achieve this are: 1. Action features and relationship concepts need to be extracted separately to decouple them from specific objects. 2. Establish a reasonable relationship network to realize the correct association ability, and this association ability must be quantifiable.
  • the machine When inputting information, the machine first finds one or more segments of the most relevant memory in the memory. They are a series of information streams related to the input information, which is achieved through association activation. These memories are past machine responses to similar input information, or past responses to multiple pieces of information that are partially similar to input information. The sender of these responses can be either the machine itself or other things. The machine takes the most frequently-occurring response between itself and the information source and related to the input information as the purpose of the information source. If there is no frequent interaction between the machine and the information source, then the machine considers the response most used by others as the purpose of the information source. This is reasonable, because the purpose of the information source is to get a response. The information source has preset possible responses based on its own experience. These pre-determined responses are established based on the interaction between the information source and the machine or the interactive experience of the information derived from others. These can all be achieved by using preset experience. When the machine understands the purpose of the information source, it also understands the input information.
  • the method for the machine to establish a response is: 4.1, the machine finds one or more segments that can be used as a reference to establish the memory of the response.
  • the specific method is: the machine converts the input linguistic information into non-verbal information streams through the association activation method, and uses these information streams and other input non-verbal information as the total input information.
  • the machine uses the converted information stream as the new virtual input information, and uses the association activation for the information again. After the Lenovo activation is completed, the information stream composed of information with high activation value.
  • the information before the input of similar information is the information about the cause, and the information after the input of the similar information is the information about the result.
  • the machine may find multiple memories about itself, or it may find multiple memories about others.
  • the real need of the machine is to find the change process of the demand state and the change process of the emotional state that are stored with these memories. If it is about the memory of others, the machine needs to replace the activities of others with its own activities, again as a new virtual input, and then activate it through association again, looking for the memory related to the cause of this virtual input and the memory related to the result, looking for these The changing process of the demand state and the changing process of the emotional state in memory.
  • the machine sets up one or more segments of memory organization with the highest total activation value in step 4.3 into one or more segments of response as the object of imitation. Because these activated messages have their own time and relationship (they are stored in memory simultaneously), they can be organized into one or more responses.
  • the machine analyzes whether it is positive or negative for the change process of one's own demand state and emotional state in one or more response-related memories to be imitated. Select these one or more responses in accordance with the principle of "seeking advantages and avoiding disadvantages". Since these memories contain changes in "demand value” and “emotional value”, only simple statistical algorithms are needed to realize the choice of "seeking advantages and avoiding disadvantages". These algorithms can be preset.
  • the machine combines one or more segments of responses that have passed the assessment of "seeking advantages and avoiding disadvantages" into a large process. There may be organizational information in time and space in these responses, so this organization is carried out in accordance with the information in time and space. If there is no clear time and space order information in these responses, then the machine needs to input these responses as a new virtual information again to find more memories through association activation to find their time and space order. This process is iterative until the order of these responses can be determined (it is also possible to find that the order of these responses can be arbitrary through memory).
  • the source of power that drives the machine is the motivation of the machine, and the motivation of the machine can be summarized as “seeking advantages and avoiding disadvantages.”
  • Profiles and “harms” are partly preset; partly they are established through acquired learning, because they are related to the needs of the machine itself. Analogous to human beings, for example, at the beginning, “water”, “milk” and “food” are pre-built “profits”. Later, through learning, we have obtained the connection between "exam scores", “banknotes” and our innate needs. Later, we discovered that the object of operation can also be non-substantial things such as “love” and "time”, and we even pursue domination in the group, which is a kind of "goal achievement” that exists in the bottom motivation of our genes. Stretch.
  • the profit value, safety value, risk value, goal achievement value, and dominance value of the machine are similar situations. They all continuously link behavior and behavior results through the machine's past experience. The way to connect them is to put them in the same memory frame. Even if the machine did not get timely feedback when the behavior occurred. The trainer may also point out the behavior itself and give feedback in the later stage, so that the behavior and the result are connected in a single memory frame. The trainer does not even need to specify which behavior is good or bad. The machine only needs to receive the correct feedback every time, and through memory and forgetting, it can gradually establish the connection between the correct behavior and the demand value. For example, those behaviors that will definitely receive rewards or punishments are memorized at the same time after each behavior and reward or punishment. Each time they repeat, their memory increases, and eventually the connection between the two will become closer and closer than the other connections.
  • the evaluation system of the machine is a preset program. This program determines whether a virtual output should be transformed into a real output based on the satisfaction state of the machine's demand for gains and losses, safety and risk values, goal achievement values, and dominance values. These types of needs are given by humans to machines. Of course, we can give machines more goals that humans expect them to have, such as "compliance with the robot convention”, “compliance with human laws”, “compassionate”, “ethical”, “behaving gracefully” and other goals. These goals can be achieved by setting demand symbols in the memory and adjusting the behavior of the machine through feedback from the trainer, so as to achieve human expectations. It needs to be pointed out that these goals can be increased or decreased in accordance with human expectations. The addition or reduction of these objectives does not affect the claims of the present application.
  • the application of the present invention proposes to use the actual satisfaction state of the machine's requirements as the input of the emotion system, and use a preset program to convert them into the emotion of the machine.
  • the purpose of this is to anthropomorphize, imitating the emotional response of human beings in different states of satisfying needs. Only in this way can machines better communicate with humans.
  • the machine can modify the parameters of the preset program by itself, and output emotions according to its own experience.
  • the machine can connect emotions and feedback.
  • Such emotions are not only a way of expression, but also a means that can be used. Because certain emotions are connected with certain external feedback. When the machine is looking for specific feedback, emotions may be incorporated into memory and become a kind of imitation object when the machine expects to reproduce specific results. It needs to be pointed out that the type and intensity of emotions can be increased or decreased according to human expectations. The addition or reduction of these objectives does not affect the claims of the present application.
  • the various evaluation values established by the machine also need to be combined with the internal state values of the machine itself (for example, is it lack of power, whether some of its own systems are broken, etc.) to make a judgment, and the result of the judgment is pass or fail.
  • the evaluation system of the machine is a preset program. It is a link that personalizes the machine, and different choices are equivalent to different personalities.
  • the machine can also retain some parameters that can be adjusted by itself, and try different options to bring different consequences, so as to gradually establish an evaluation system that best meets its needs. This step can be achieved by the existing publicly known technology, and will not be repeated here.
  • the machine If the machine establishes a response, it cannot pass the evaluation system. Then the machine needs to re-establish the response, and needs to remove the behaviors that brought heavy losses, dangers and other negative results in the last evaluation. These behaviors are the combined behaviors of the static feature maps and dynamic feature maps that bring losses. Getting rid of negative behaviors is also a more complicated machine thinking process. In this process, the machine needs to convert all the current goals into inheritance goals, leaving the computing power vacant for the calculation of a temporary goal such as removing negative behaviors. Then, the machine needs to look for all the memories of this negative behavior and find the experience of how to exclude it. After removing the behavior that brought the negative result, the machine re-established a new response.
  • the process of establishment is still to optimize dynamic feature maps, replace static feature maps with concepts, and then use similar memories to determine their combination. If the machine is repeated many times, it still cannot establish a response that can pass the evaluation. It is possible that there was an error in the previous steps, or the machine encountered an unsolvable problem. At this time, the machine enters the processing of the "unprocessable information" flow. In other words, "unable to process information” itself is a result of processing information. The machine builds a response to "unable to process information" based on its own experience. These responses may be ignored, may be reconfirming the information with the information source, or again using higher resolution to identify the information, etc. These are also reasonable responses similar to human behavior.
  • the machine needs to use the Lenovo activation process repeatedly. It needs to be pointed out here that due to the activation threshold, even if the activation value transfer coefficient between the feature maps is linear, the activation value accumulation function of the feature map is also linear, but due to the existence of the activation threshold, no matter in the single association activation process In the process of multiple association activations, the same feature map and the same initial activation value, but because the activation order is selected differently, the final activation value distribution is different. This is because of the non-linearity caused by the existence of the activation threshold. Different transmission paths bring different information losses. The preference of activation order selection is equivalent to the difference in machine personality. Therefore, under the same input information, different thinking results are produced. This phenomenon is consistent with human beings.
  • the strength of the relationship in the relationship network is related to the latest memory value (or connection value). Therefore, the machine will be preconceived. For example, if two machines with the same relationship network face the same feature map and the same initial activation value, one of the machines suddenly processed an input information about this feature map, then this machine is processing this additional piece of information Later, it will update the relevant part of the relationship network.
  • One of the relationship lines may increase according to the memory curve. This increased memory value will not fade in a short time. Therefore, when facing the same feature map and the same initial activation value, the machine that processes the additional information will spread more activation values along the newly enhanced relationship line, which will lead to a preconceived phenomenon.
  • the activation value in the association activation will decrease with time. . Because if the activation value in the relational network does not fade with time, the activation value changes brought about by the subsequent information will not be obvious enough, which will cause interference between information. If the activation value does not fade, after the subsequent information is entered, it will be strongly interfered by the previous information, resulting in the inability to find one's focus correctly. But if we completely clear the memory value of the previous information, then we will lose the possible connection relationship between the two pieces of information before and after.
  • the thinking time given to the machine is limited, or there is too much information, and the machine needs to complete the information response as soon as possible.
  • the machine can also adopt the method of output and then input. In this way, the machine emphasizes useful information and suppresses interference information.
  • These methods are commonly used by humans, and in the application of the present invention, we also introduce them into the thinking of machines.
  • the machine can determine whether the current thinking time exceeds the normal time based on the built-in program, or its own experience, or a mixture of the two, and it needs to refresh the attention information, or tell others that they are thinking, or emphasize the key points, and eliminate interference information.
  • the virtual output of the machine's self-information filtering or emphasizing method is usually speech, because this is the most common output method.
  • the machine outputs them the least energy.
  • this is closely related to a person's growth process. For example, people who learn about life from books may convert information into words and then re-enter it.
  • the search method that uses association activation uses the implicit connection relationship among the input information of language, text, image, environment, memory and other sensors to transfer activation values to each other, so that related feature maps, concepts and memories are mutually connected Support and highlight.
  • the difference between it and the traditional "context" to identify information is that the traditional recognition method requires manual establishment of the "context" relation database in advance.
  • we put forward the basic assumption of "similarity and implicit connection between information in the same environment”. Based on this basic assumption, all kinds of relationships are simplified, allowing the machine to build a network of relationships on its own. It contains not only semantics, but also common sense.
  • Lenovo activation is a search method, which itself is not a necessary step in the application of the present invention, and can be replaced by other search methods that can achieve similar purposes.
  • the machine can consider the feature map of each memory whose activation value exceeds the preset value as having been used once, and maintain their memory value according to the memory and forgetting mechanism in the memory bank to which the memory belongs.
  • Imitation is the ability of human beings to exist in genes. For example, for a babbling child, if every time he (she) returns home, we greet him (her) and say “you are back.” After several times, when he (she) goes home again, he (she) will take the initiative to say "you are back”. This shows that he (she) has begun to imitate others to learn without understanding the meaning of the information. In the same way, we let machine learning use the same method. The machine also imitates the experience of others or its own to understand and respond to the input information.
  • Performing the response step is a translation process.
  • the machine uses voice output, which is relatively simple. It only needs to convert the image feature map to be output into voice, and then use the relational network and memory to change the dynamic
  • the feature map (including the concept that represents the relationship) is combined with the static concept, organized into a language output sequence, and the pronunciation experience is used to implement it. It needs to be pointed out that the machine may choose some dynamic features that express the entire sentence based on experience (self or other people's experience) (such as using different movement patterns of tone, audio pitch, or stress changes to express doubts, mockery, distrust, emphasizing key points, etc.) Common way). Because machines learn these expressions from human life, in theory, machines can learn all the expressions that humans have.
  • the machine needs to target the image feature map sequence to be output (this is the intermediate target and the final target). According to these targets, different time and space are involved. The machine needs to divide them in time and space in order to coordinate their execution efficiency.
  • the method adopted is to select groups that are closely related in time and that are closely related in space. Because the dynamic feature map and the static feature map are combined to form an information combination, the environment space of the related memory contains time and space information, so this step can use the classification method. (This step is equivalent to rewriting from the overall script to the sub-script).
  • the machine needs to combine the intermediate targets in each link again with the real environment and adopt the method of segmented imitation to expand layer by layer.
  • the response plan proposed by the machine at the top level is usually only composed of highly generalized process features and highly generalized static concepts (because these highly generalized processes can find multiple similar memories, so learn from them to establish The response is also highly general). For example, under the total output response of "business trip", "going to the airport” is an intermediate link goal. But this goal is still very abstract, and machines cannot perform imitation.
  • the machine needs to be divided according to time and space, and the link that needs to be executed in the current time and space is the current goal. And take other goals in time and space as inheritance goals and put them aside for the time being. After the machine takes the intermediate link as the target, the machine still needs to further subdivide time and space (write down the score script again).
  • This is a process of increasing temporal and spatial resolution.
  • the process by which a machine converts a target into multiple intermediate links is still a process of creating various possible responses, using an evaluation system to evaluate them, and selecting their own responses according to the principle of "seeking advantages and avoiding disadvantages".
  • the above process is continuous iteration, and the process of dividing each goal into multiple intermediate goals is a completely similar processing flow.
  • the bottom experience is to mobilize muscles to make syllables.
  • it is broken down to issuing drive commands to related “muscles”.
  • This is a tower-shaped decomposition structure.
  • the machine starts from the top-level goal and decomposes a goal into multiple intermediate-link goals. This process is to create virtual intermediate process goals, if these intermediate process goals "meet the requirements", keep them. If "does not meet the requirements", re-create it. This process unfolds layer by layer, and finally establishes the colorful response of the machine.
  • the machine is to perform imitation tasks that can be performed while decomposing other goals into more detailed goals. So the machine is thinking while doing it. This is because the reality is very different, and it is impossible for the machine to know the external situation in advance and make a plan. So this is a process in which the environment and the machine interact to complete a goal.
  • step S1 the establishment of low-level features is mainly to use memory and forgetting mechanisms. Each time the machine finds a similar local feature through the local field of view, if there are already similar local features in the feature library, it will increase its memory value according to the memory curve. If there is no similar local feature in the feature library, store it in the feature map and give it an initial memory value. The memory values in all feature libraries gradually decrease according to the forgetting curve with time or training time (increasing with the number of training samples). In the end, the simple features that are widely present in various things will have high memory value and become the underlying feature map.
  • step S2 each time a bottom-level feature or feature map is found, if there are already similar bottom-level features or feature maps in the temporary memory library, feature library, or memory, its memory value increases according to the memory curve. They also follow the forgetting mechanism.
  • the machine first saves the environment space into the temporary memory bank. When the machine stores these environment spaces in the memory bank, it will also store the feature maps in the environment space and their memory values. The initial memory values of these feature maps are positively correlated with the activation values when their storage occurs.
  • steps S3, S4, S5 and S6 the memory value of the feature map in the memory bank complies with the memory and forgetting mechanism. Whenever a relationship in the memory is used once, the feature map involved in this relationship will increase the memory value according to the memory curve, and all the feature maps will forget the memory value according to the forgetting curve of the memory bank in which they are located.
  • 6.1 Directly adopt the time and space relationship of information input, store them in order, and establish three-dimensional coordinates to represent the distance between information.
  • the time axis of this coordinate can be driven by an event-driven mechanism: each time an event occurs, the memory is stored once, and the time axis increases by one unit.
  • each feature has its own three-dimensional coordinates in the memory space. In this way, the machine can quickly find all similar features, and according to the spatial coordinate information of these features, to achieve proximity activation and strong memory activation.
  • each feature receives the activated electrical signal and also imitates the brain nerves, and the feature with high memory value is used to receive strong, and the feature's receiving ability is positively correlated with the degree of matching between the activated electrical signal and itself.
  • FIG. 3 is a schematic diagram of a module for realizing general machine intelligence.
  • the core idea of the method shown in Figure 3 is to use a separate module to implement the association activation process. After this module gets the input information, it searches memory to find the proximity information, similarity information and strong memory value information to the input information. Then according to the corresponding algorithm, the activation value is directly assigned to the information in the memory.
  • the activation value is directly assigned to the information in the memory according to the corresponding algorithm again.
  • This process is iterative until the Lenovo activation process stops because each message has an activation preset.
  • This is a "God" perspective, using a preset algorithm, and directly using an external algorithm to complete the association activation according to the spatial distance, memory value and similarity of the memory space.
  • the algorithm that the memory value and the activation value in the memory fade with time can either be refreshed in the memory bank using software or hardware, or it can be implemented using the memory association activation module.
  • S600 is to establish a machine feature extraction module.
  • This module selects the static features and dynamic features of the data at different resolutions by comparing the local similarity, and establishes the contrast similarity or trains the neural network, or any other existing algorithms to extract the features of the data.
  • the S601 and S602 modules are modules that extract the features of multi-resolution information from external input information, and they involve different resolutions.
  • the machine may need to perform feature extraction on input data at multiple resolutions.
  • the same sensor data can be divided into multiple channels of data through preprocessing to extract different characteristics of the data.
  • different preprocessing algorithms can be used again at different resolutions to extract data features at different resolutions.
  • S601 and S602 can use separate hardware to complete the multi-resolution feature extraction function.
  • the machine can include two modules in S603.
  • One of them is a dedicated module for Lenovo activation, which can be a dedicated search hardware. The purpose of this is to solidify the search memory and assign activation value algorithms, and improve efficiency by using specialized hardware.
  • the other is a module that combines memory information and reality information, which is equivalent to software that realizes data reorganization. This step is mainly to find the dynamic process from the relevant memory, and then generalize the experience through the generalization ability of the action characteristics.
  • S604 is the entire memory bank (including the quick search library established to improve search efficiency, which contains commonly used memory information. It also includes temporary memory banks, long-term memory banks, and possibly other memory banks).
  • the memory bank is equivalent to storage space, but it carries the life cycle (memory value) of each information.
  • the memory bank can use a special memory value refresh module to maintain the memory value.
  • S605 is a demand assessment system, which uses the demand value obtained in the S603 process to make logical judgments. S605 can be implemented in software.
  • S606 is a segmented imitation process (a process of iterative concept development). This process requires constant calls to S603 and S604, which can be implemented by software.
  • S607 is a logical judgment, and it can be realized by software.
  • S608 is a newly memorized storage process, which can be implemented by software or dedicated hardware. The new memory contains the internal and external input information of the machine, the demand information of the machine and the emotional information of the machine.
  • S609 is the state of completing an information response cycle.
  • FIG. 3 it is characterized in that separate hardware can be used to implement multi-resolution feature extraction and separate hardware can be used to implement the association activation process.
  • FIG. 4 is a schematic diagram of another module for realizing general machine intelligence.
  • the core idea of the method shown in Figure 4 is to integrate the algorithm for realizing the association activation process in a memory module in a distributed manner.
  • S704 is a memory bank that can imitate the memory function of the brain and realize the functions of proximity activation, strong memory activation and similarity activation. It mimics the way of the brain to receive the electrical excitation signal transmitted by the characteristic, and realizes the propagation and attenuation of the electrical excitation signal in the memory according to the distance of the memory space, and also imitates the brain to achieve strong memory activation.
  • the memory module itself can also integrate search algorithms to achieve similarity activation, and there are many ways to achieve similarity activation. They need to be implemented according to different memory data organization methods.
  • the remaining parts in FIG. 4 are the same as those in FIG. 3.

Landscapes

  • Engineering & Computer Science (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • General Health & Medical Sciences (AREA)
  • Computing Systems (AREA)
  • Biomedical Technology (AREA)
  • Biophysics (AREA)
  • Computational Linguistics (AREA)
  • Data Mining & Analysis (AREA)
  • Evolutionary Computation (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Molecular Biology (AREA)
  • Artificial Intelligence (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Mathematical Physics (AREA)
  • Software Systems (AREA)
  • Health & Medical Sciences (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

A learning method for imitating an association activation process of human memory. Experience and input information are recombined into multiple response schemes by means of experience summarization and experience generalization, the response schemes are evaluated according to an approach of seeking advantages and avoiding disadvantages, and a response is implemented by dividing one response scheme into multiple intermediate links to find imitable experience and other means. By means of the learning method for imitating an association activation process of human memory, machines can gradually obtain simple to complex responses to the input information and have emotional expressions similar to those of humans. This method shows a huge difference from the existing machine learning methods in the industry, and no similar method exists in the industry at present.

Description

一种模仿人类记忆来实现通用机器智能的方法A way to imitate human memory to realize general machine intelligence 技术领域Technical field
本发明申请涉及人工智能领域，尤其涉及建立类似人类智能的通用人工智能领域。The application of the present invention relates to the field of artificial intelligence, in particular to the field of establishing general artificial intelligence similar to human intelligence.
背景技术Background technique
当前人工智能通常是为特定任务设计的，还没有能够完成多种不确定性任务的通用人工智能。当前人工智能通常是从大量的标记数据中寻找映射关系，它们无法从输入的信息推测原因、预测结果并做出选择和响应。所以，当前的机器智能和人类的智能差异很大。而本发明申请中，我们提出一种建立多分辨率对象，并在学习和生活中提取这些对象之间的多分辨率连接关系，建立多分辨率关系网络。在新的信息输入后，我们通过调用输入信息在多个分辨率下的组织关系，并寻找记忆中类似的组织关系，来推测信息的产生原因、预测可能的结果并做出合理的选择和响应。本发明申请在多分辨率关系网络的基础上，进一步提出了建立类似于人类的思维、情绪和个性的人工智能的方法和步骤。The current artificial intelligence is usually designed for specific tasks, and there is no general artificial intelligence that can complete a variety of uncertain tasks. Current artificial intelligence usually finds the mapping relationship from a large amount of labeled data. They cannot infer the cause, predict the result, and make choices and responses from the input information. Therefore, the current machine intelligence and human intelligence are very different. In the application of the present invention, we propose a method for establishing multi-resolution objects, and extracting the multi-resolution connection relationship between these objects in study and life to establish a multi-resolution relationship network. After the new information is input, we call the organizational relationship of the input information at multiple resolutions and look for similar organizational relationships in memory to speculate on the cause of the information, predict the possible results, and make reasonable choices and responses. . On the basis of the multi-resolution relational network, the application of the present invention further proposes methods and steps for establishing artificial intelligence similar to human thinking, emotion and personality.
发明内容Summary of the invention
在本发明申请中，我们首先提出几条关于人类大脑工作方式的基本假设，然后通过这些假设来分析大脑的工作流程。然后我们提出如何在机器上实现类似于大脑的基本假设，并提出了在机器上如何模仿大脑的工作流程，来实现通用的人工智能。In the present application, we first propose a few basic hypotheses about the working mode of the human brain, and then analyze the work process of the brain through these hypotheses. Then we proposed how to implement the basic hypothesis similar to the brain on the machine, and proposed how to imitate the work process of the brain on the machine to realize general artificial intelligence.
首先，我们假设大脑具有多分辨率下的对输入信息的特征提取能力，其目的是对输入信息在不同分辨率上建立联系。其次，我们认为大脑具有联想能力，它可以根据过去的经验来预测输入信息产生的原因和可能带来的结果。再其次，我们认为大脑具有泛化能力，这是为了把过去的经验应用于不同的对象上。再其次，我们认为大脑存在需求和情绪***，这是大脑在输入信息刺激下，创造出各种可能的响应，并从中选出符合大脑期望的响应。最后，我们认为大脑具有模仿的能力。大脑通过模仿过去多段经验，并通过泛化能力把这些经验和现实信息相结合来执行模仿，并在输出过程中根据实际情况不断调整。First, we assume that the brain has the ability to extract features of input information at multiple resolutions, and its purpose is to establish connections between input information at different resolutions. Secondly, we believe that the brain has the ability to associate. It can predict the causes and possible results of input information based on past experience. Secondly, we believe that the brain has the ability to generalize, which is to apply past experience to different objects. Secondly, we believe that there is a need and emotional system in the brain. This is when the brain creates a variety of possible responses under the stimulation of input information, and selects the response that meets the expectations of the brain. Finally, we believe that the brain has the ability to imitate. The brain imitates multiple past experiences and combines these experiences with real-life information through generalization capabilities to perform imitation, and constantly adjusts according to the actual situation during the output process.
图1是本发明申请中提出的实现通用人工智能的主要部分。S1是实现多分辨率数据特 征提取部分。通过S1部分，机器把输入信息变成多路输入，每一路是不同分辨率下的对应数据特征。S2是实现联想激活能力的部分。S2部分主要建立在3个方法的基础上：机器在记忆中寻找相关经验时，采用的方法是“临近激活”原则、“相似激活”原则和“强记忆激活”原则。其中“临近激活”是指记忆中特定的信息激活后，它只能激活它附近的信息。“相似激活”是指记忆中的特定特征，接收其他特征发出的激活信号时，接收能力和彼此之间相似度成正相关。“强记忆激活”是指记忆值越高的记忆，接收其他特征发出的激活信号的能力越强。在这3个原则的基础上，机器就能实现类似于人脑的联想能力。S3实现泛化能力的部分。在本发明申请中，我们提出的经验泛化方法是：在多个相似过程中，通过降低分辨率来寻找这些相似过程中的共有部分，把这些共有部分作为过程特征。在重建具体过程时，只要现实中的对象，在特定的分辨率下，符合过程特征中对象需要的所有特征，那么就是可以替换的。这是经验泛化的关键步骤，也是我们需要建立不同的事物、场景和过程的多个分辨率下的特征的原因。S4是需求和评估的部分。这个部分中最关键的基础是机器在建立关系网络时，把需求和情绪的改变与引起这些改变的信息存放在一起。通过一次次类似情况的重复，特定的信息就会和特定的需求和情绪建立更加紧密的联系。机器存储的信息包括外部输入信息、内部输入信息、机器自身的需求/情绪和这些信息之间的关系，并通过记忆和遗忘机制来量化这些连接关系的强度。有了这些关系后，机器通过记忆网络，使用联想能力寻找和输入信息相关的经验。机器使用泛化能力把这些经验泛化到输入信息上。机器把这些泛化后片段模仿过去经验中相似的组织方式把它们拼接起来，作为计划响应，然后在“趋利避害”的原则下评估计划响应带来的结果是否符合自己的预期。如果符合就输出计划响应。如果不符合就重新选择响应。在S5步骤中，机器是把计划响应里面的概念(中间目标环节)，通过分段模仿来层层展开(更加细致的中间目标环节)，展开到记忆可以具体执行的底层经验，并按照时间和空间划分来执行这些响应。Fig. 1 is the main part of the implementation of general artificial intelligence proposed in the application of the present invention. S1 is the part that realizes the feature extraction of multi-resolution data. Through the S1 part, the machine turns the input information into multiple inputs, each of which is a corresponding data feature at a different resolution. S2 is the part that realizes the ability of Lenovo to activate. The S2 part is mainly based on three methods: when the machine searches for relevant experiences in memory, the methods used are the "proximity activation" principle, the "similar activation" principle and the "strong memory activation" principle. Among them, "proximity activation" means that after the specific information in the memory is activated, it can only activate the information near it. "Similar activation" refers to specific features in memory. When receiving activation signals from other features, the receiving ability is positively correlated with the similarity between them. "Strong memory activation" refers to the higher the memory value, the stronger the ability to receive activation signals from other characteristics. On the basis of these three principles, the machine can realize the association ability similar to the human brain. S3 realizes the part of generalization capability. In the application of the present invention, the empirical generalization method we propose is to find the common parts of these similar processes by reducing the resolution in multiple similar processes, and use these common parts as process features. When reconstructing the specific process, as long as the object in reality meets all the features required by the object in the process feature at a specific resolution, then it can be replaced. This is the key step of experience generalization, and it is also the reason why we need to establish the characteristics of different things, scenes and processes at multiple resolutions. S4 is the needs and assessment part. The most critical foundation in this part is that when the machine builds a network of relationships, it stores the changes in needs and emotions with the information that caused these changes. Through repeated repetition of similar situations, specific information will be more closely related to specific needs and emotions. The information stored by the machine includes external input information, internal input information, the machine's own needs/emotions and the relationship between these information, and the strength of these connections is quantified through the memory and forgetting mechanism. With these relationships, the machine uses the memory network to find and input information-related experience using the ability of association. The machine uses generalization ability to generalize these experiences to the input information. The machine combines these generalized fragments to imitate similar organizational methods in the past experience and joins them together as a planned response, and then evaluates whether the results of the planned response meet their expectations under the principle of "seeking advantages and avoiding disadvantages". If it meets, output the planned response. If it does not match, select the response again. In step S5, the machine expands the concept (intermediate target link) in the plan response through segmented imitation (more detailed intermediate target link), and expands it to the underlying experience that can be specifically executed in memory, and according to time and Space division to perform these responses.
整个机器的思维过程是迭代的。机器每次处理新信息的方式都是：把目前的目标，转 为“继承目标”。机器对新信息做多分辨率特征提取。在关系网络中，通过联想能力，寻找和“继承目标”、“新信息”相关的经验。通过泛化能力，把这些经验泛化到输入信息上。通过借鉴这些经验过去的组织方式，把泛化后的经验片段按照时间和空间关系，组合成一个可能的响应计划。然后基于过去的经验，在“趋利避害”的原则下评估响应计划可能给自己带来的影响。如果无法通过评估，则重新建立响应计划；如果通过，则把响应计划作为输出计划，通过分段模仿的方式，把这个计划中的每个环节展开到更加具体中间环节。这个过程也是迭代进行，直到展开到机器可以立即执行的底层经验为止。这个过程是一个边想边做的过程，在这个过程中，一但有新信息输入，机器再次回到“把目前的目标，转为继承目标，并对新信息做多分辨率特征提取”的过程中去。所以，在本发明申请中，机器只需要使用很简单的步骤反复迭代，就可以实现类似于人类的思维过程和人类的需求以及情绪响应。The thinking process of the entire machine is iterative. The way the machine processes new information each time is to turn the current goal into an "inherited goal." The machine performs multi-resolution feature extraction on the new information. In the relationship network, through the ability of association, find the experience related to the "inheritance goal" and "new information". Through generalization ability, these experiences are generalized to the input information. By learning from the past organization of these experiences, the generalized experience fragments are combined into a possible response plan according to the time and space relationship. Then based on past experience, under the principle of "seeking advantages and avoiding disadvantages", evaluate the possible impact of the response plan on oneself. If the assessment fails, the response plan is re-established; if it is passed, the response plan is taken as the output plan, and each link in the plan is expanded to more specific intermediate links by means of segmented imitation. This process is also iterative until it reaches the bottom experience that the machine can execute immediately. This process is a process of thinking while doing. In this process, once new information is input, the machine returns to "turn the current target into the inherited target, and perform multi-resolution feature extraction of the new information". Go in the process. Therefore, in the application of the present invention, the machine only needs to use very simple steps to iterate repeatedly to achieve a human-like thinking process and human needs and emotional response.
附图说明Description of the drawings
图1是本发明申请的主要组成部分。Figure 1 is the main components of the present application.
图2是实现多分辨率特征提取的方法。Figure 2 is a method to achieve multi-resolution feature extraction.
图3是一种功能模块组织示意图。Figure 3 is a schematic diagram of a functional module organization.
图4是另外一种功能模块组织示意图。Figure 4 is a schematic diagram of another functional module organization.
具体实施方式Detailed ways
在本发明中，我们首先说明我们为什么采用上述方式来实现通用人工智能，然后具体来说明我们采用什么方式来实现这些目标。In the present invention, we first explain why we use the above methods to achieve general artificial intelligence, and then specifically explain what methods we use to achieve these goals.
我们首先分析一种大脑可能的工作流程。我们假设大脑已经具有使用多分辨率来提取输入信息特征的能力(假设这是基因里带来的能力)。当外界信息输入后，信息的预处理部分首先提取了1～K1(自然数)层不同分辨率下的信息特征，每一层分辨率可能有多个对应特征。然后大脑开始在记忆中搜索相似的特征。假设大脑的基本搜索方法就是激活需要搜素的特征，使其发出特定模式的激活电信号，这种电信号可以在记忆空间中传播，并且会随传播距离而衰减。假设记忆中，其他神经组织可以接收这种特定模式的电信号，但只有相似特征的记忆 (在之前通过类似信息记忆下来的神经组织)才能更好的接收激活电信号(因为模式匹配)。那么，显然这些电信号只能激活和它们相邻的神经组织(因为输入激励强)，也能激活和它们相似的更远的神经组织(因为对方接收能力好)，那些记忆深刻的特征(比如神经元或者突触多)因为接收单元多也可能被激活。假设被激活的神经组织，如果它们的激活值超过了预设阈值，它们也会发出自己的特定模式的电信号。那么，它们发出的电信号，同样也只能激活和它们相邻的神经组织和激活距离它们更远的、但结构相似的神经组织，还能激活那些和它们不相似，但记忆值高的记忆。We first analyze a possible workflow of the brain. We assume that the brain already has the ability to use multi-resolution to extract features of input information (assuming this is the ability brought by genes). After the external information is input, the preprocessing part of the information first extracts the information features at different resolutions from 1 to K1 (natural number) layers, and each layer of resolution may have multiple corresponding features. Then the brain starts searching for similar features in the memory. It is assumed that the basic search method of the brain is to activate the features that need to be searched, so that it emits a specific pattern of activation electrical signals. This electrical signal can propagate in the memory space and will attenuate with the propagation distance. Assuming that in memory, other nerve tissues can receive electrical signals of this specific pattern, but only memories with similar characteristics (neural tissues previously memorized through similar information) can better receive activation electrical signals (because of pattern matching). Then, obviously these electrical signals can only activate the nerve tissues adjacent to them (because the input excitation is strong), and can also activate the more distant nerve tissues similar to them (because the other party has good receiving ability), those memorable features (such as More neurons or synapses) may be activated because of more receiving units. Assuming that the activated nerve tissues, if their activation value exceeds the preset threshold, they will also send out their own specific pattern of electrical signals. Then, the electrical signals they send out can only activate the nerve tissues adjacent to them and the nerve tissues that are farther away from them, but with similar structures. They can also activate memories that are not similar to them but have high memory values. .
举例说明：当一组“餐桌”特征输入到我们的大脑后，大脑首先把它们转变成多分辨率下的信息特征(比如最粗糙分辨率下是一个整体的立体形象；然后是更细分辨率下的桌面的形状和桌腿的轮廓；然后是更细的分辨率下的桌面纹理、边缘轮廓的其他细节等)，然后每个分辨率下的特征依次发出这些特征对应的激活电信号(假设存在一种神经组织，可以在不同的输入信息特征激励下发出其对相应模式的激活电信号)。显然，这些激活电信号只能激活我们最近的记忆(因为记忆空间中距离近，衰减小，我们称之为临近激活)以及和输入信息相似的记忆(因为模式匹配，所以它们接收能力好，我们称之为相似性激活)，还有那些让我们印象深刻的记忆(因为这些记忆的神经元或者突触多，接收能力也更好)。所以通过“餐桌”，我们可能回忆起昨天我们放在餐桌上的一包零食，或者回忆起小时后和妈妈在餐桌边做手工的情景。要实现上述联想的能力，暗含了假设我们的记忆是按照时间顺序存储的，并暗含了假设每个时间段记忆中存储信息是按照和现实类似的方式存储的。所以，当“餐桌”信号激活了我们小时后“我和妈妈在餐桌边做手工”的情景后，还可能激活“当时一只皮球突然把窗户玻璃砸碎了”的记忆(尽管这个信息和餐桌距离较远，但它记忆值高。另外，信息的存储时间顺序是按照注意力关注的次序进行的，因为它们是信息输入的时间次序)，因为这个场景有更多的神经元或者突触来存储它们，也有更强的接收电信号的能力(我们称之为强记忆激活)。由于我们的大脑在存储记忆时，同时存储我们的情绪，那些强烈的情绪同样记忆深刻， 同样拥有更多神经元或者突触来存储它们，所以我们还能回忆起我们的情绪从“快乐”转为“受到惊吓”(强记忆激活)。我们把上述3种激活方式(临近激活、相似激活和强记忆激活)称之为联想激活。尽管我们提出的是一种假设，但我们认为这个假设和我们大脑的联想功能是非常类似的。而联想功能正是我们大脑的主要工作方式。For example: when a set of "dining table" features are input to our brain, the brain first transforms them into information features at multi-resolution (for example, at the coarsest resolution, it is an overall three-dimensional image; then it is at a finer resolution. The shape of the desktop and the outline of the table legs under the lower resolution; then the desktop texture at a finer resolution, other details of the edge contour, etc.), and then the features at each resolution emit the activation electrical signals corresponding to these features in turn (assuming There is a kind of nerve tissue that can send out its activation electrical signals to the corresponding mode under the excitation of different input information characteristics). Obviously, these activating electrical signals can only activate our nearest memory (because the distance in the memory space is short, the attenuation is small, we call it proximity activation) and memories similar to the input information (because the pattern matches, they have good receiving ability, we Call it similarity activation), and memories that impress us (because these memories have more neurons or synapses, and their ability to receive them is better). So through the "dining table", we may recall a pack of snacks that we put on the dining table yesterday, or recall the scene of making handicraft with our mother at the dining table after an hour. To realize the above-mentioned ability of association, it implies the assumption that our memory is stored in chronological order, and it implies the assumption that the information stored in the memory of each time period is stored in a manner similar to reality. Therefore, when the "dining table" signal activates the scene of "I and my mother are doing handwork at the table" after our hours, it may also activate the memory of "the window glass was suddenly smashed by a leather ball" (although this message is related to the dining table). The distance is far, but it has a high memory value. In addition, the time sequence of information storage is in the order of attention, because they are the time sequence of information input), because this scene has more neurons or synapses. Storing them also has a stronger ability to receive electrical signals (we call it strong memory activation). Since our brain stores our emotions while storing memories, those strong emotions are also deeply memorable, and they also have more neurons or synapses to store them, so we can still recall our emotions from "happy". It is "frightened" (strong memory activation). We call the above three activation methods (proximity activation, similar activation and strong memory activation) associative activation. Although we are proposing a hypothesis, we believe that this hypothesis is very similar to the associative function of our brain. The association function is the main working mode of our brain.
大脑的另外一种工作方式是“趋利避害”。在上面的例子中，当“我和妈妈在餐桌边做手工”之所以快乐，是因为这样的活动满足了我们的“需求”，比如一种安全需求或者亲情需求，这种满足给我们带来了快乐。而我们受到惊吓，也是因为这种突发情况破坏了我们的安全感，使得我们的危险符号被激活，让我们意识到发生了紧急情况，需要改变自己的状态来应对。当我们通过经验，对发生的事情类比分析后，明白发生了什么时，这时危险符号的激活值降低(危险符号的激活值增加和减小，也是通过不同输入特征和危险符号之间的连接强度决定的。而这种连接强度，一部分是通过先天预置的，还有一部分正是通过我们后天的记忆联想激活过程实现的)。但这时“窗玻璃破碎”破坏了我们财产安全的需求(这时一种安全需求的泛化)，所以这种财产安全被破坏，带给我们了“生气”的情绪。这种从“受到惊吓”到“生气”的情绪转换，也存在于我们的记忆中。从这个过程看，这些情绪的转变，是受到我们心理需求状态的转变而控制的，这种控制是一种“与生俱来”的能力(想想“婴儿”因为没有喝到奶而“生气”)。而我们在后天通过“趋利避害”的方法，慢慢学习到如何调整情绪，并把情绪作为我们表达信息的一种手段。我们把上述通过需求评估后的情绪反应作为“需求和情绪***”。尽管我们提出的是一种假设，但我们认为这个假设和我们大脑的情绪调节功能非常类似的。而这些功能正是我们大脑的主要工作方式。Another way of working of the brain is "seeking advantages and avoiding disadvantages." In the above example, when "Mum and I do handicrafts at the table" is happy because this kind of activity meets our "needs", such as a safety need or a family need, and this satisfaction brings us Happy. And we are frightened because this sudden situation undermines our sense of security, makes our danger symbol activated, and makes us realize that an emergency has occurred and we need to change our state to deal with it. When we understand what happened through experience and analogy analysis of what happened, then the activation value of the danger symbol decreases (the activation value of the danger symbol increases and decreases, also through the connection between different input features and the danger symbol Strength is determined. And this connection strength is partly preset through innateness, and partly through our acquired memory and association activation process). But at this time, the "broken window glass" destroys our property security needs (a generalization of security requirements at this time), so this property security is destroyed, which brings us "angry" emotions. This emotional transition from "frightened" to "angry" also exists in our memory. From the perspective of this process, these emotional changes are controlled by the changes in our psychological needs. This control is an "innate" ability (think about "babies" being "angry" because they did not drink milk. "). However, we gradually learn how to adjust emotions and use emotions as a means of expressing information through the method of "seeking advantages and avoiding disadvantages" the day after tomorrow. We take the above-mentioned emotional response after passing the needs assessment as the "needs and emotional system". Although we are proposing a hypothesis, we think that this hypothesis is very similar to the emotion regulation function of our brain. And these functions are the main way our brain works.
大脑在有了联想和“趋利避害”的基础上，就有了预测的能力。预测是大脑时时刻刻在运用的能力，它的基础就是大脑通过记忆和遗忘机制建立起来的事物之间、概念之间的关系网络。关系网络主要完成2件事：1，对需要增加记忆的信息增加记忆(用更多的神经元或者突触来记忆)。2，建立每个信息到需求和情绪的连接强度。一但关系网络建立了这两种关 系后，当有新信息输入时，通过联想激活***，我们就可以找到类似的记忆。并通过对类似记忆，在时间维度和相似性维度两个方向顺次激活，就能找到这些类似记忆的起因和结果。大脑就可以采用类比的方法，把过去的经验(起因或者结果)泛化到输入信息上，从而生成对输入信息的预测，这样就可以根据需求和情绪***来做出自己的选择。The brain has the ability to predict on the basis of association and "seeking advantages and avoiding disadvantages". Prediction is the ability that the brain uses all the time. Its foundation is the network of relationships between things and concepts established by the brain through the mechanism of memory and forgetting. The relationship network mainly accomplishes 2 things: 1. Increase memory for information that needs to be increased memory (use more neurons or synapses to remember). 2. Establish the strength of the connection of each information to needs and emotions. Once the relationship network has established these two relationships, when new information is input, we can find similar memories through the Lenovo activation system. And by sequentially activating similar memories in the two directions of the time dimension and the similarity dimension, the causes and results of these similar memories can be found. The brain can use an analogy method to generalize past experience (cause or result) to input information, thereby generating predictions for input information, so that it can make its own choices based on needs and emotional systems.
在预测的过程中，泛化是最关键的一步。在本发明申请中，我们提出的经验泛化方法是：在多个相似过程中，通过降低分辨率来寻找这些相似过程中的共有部分，把这些共有部分作为过程特征。在重建具体过程时，只要现实中的对象，在过程特征中指定的分辨率下，和经验中的对象是一样的，则就是可以替换的。这就是经验泛化的关键步骤。这也是我们需要建立不同的事物、场景和过程的多个分辨率下的特征的原因。In the process of forecasting, generalization is the most critical step. In the application of the present invention, the empirical generalization method we propose is to find the common parts of these similar processes by reducing the resolution in multiple similar processes, and use these common parts as process features. When reconstructing the specific process, as long as the object in reality, at the resolution specified in the process feature, is the same as the object in experience, it can be replaced. This is the key step of experience generalization. This is why we need to establish the characteristics of different things, scenes and processes at multiple resolutions.
在有了联想激活***、泛化能力、需求和评估***，大脑就有了预测和选择的能力。在这些能力的基础上，大脑通过分段模仿来执行自己的选择。分段模仿就是把大脑的响应计划去具体落实：把计划响应里面的主要概念(中间环节)，通过分段模仿来层层展开(更加细致的中间环节)，展开到记忆可以具体执行的底层经验，并按照时间和空间划分来执行这些响应。With the association activation system, generalization ability, demand and evaluation system, the brain has the ability to predict and choose. Based on these abilities, the brain executes its own choices through segmented imitation. Segmented imitation is the concrete implementation of the brain’s response plan: the main concepts (intermediate links) in the planned response are expanded layer by layer (more detailed intermediate links) through segmented imitation, to the underlying experience that can be specifically implemented in memory , And execute these responses according to time and space division.
下面，我们通过模仿我们提出的大脑工作工作方式，来实现通用人工智能。Next, we will realize general artificial intelligence by imitating the working method of the brain that we propose.
1，建立多分辨率特征提取的能力。1. Establish the capability of multi-resolution feature extraction.
我们认为，在我们的世界，不可能有两个完全一样的东西。当我们说两个物体是同类物体时，是指在我们使用的信息分辨率下，它们是相同的。所以，在本发明申请中，我们需要从细节到抽象，逐步使用不同的分辨率来识别信息。这就是建立不同分辨率下特征的过程。We believe that in our world, there cannot be two exactly the same things. When we say that two objects are of the same kind, we mean that they are the same at the information resolution we use. Therefore, in the application of the present invention, we need to gradually use different resolutions to identify information from details to abstraction. This is the process of establishing features at different resolutions.
在建立多分辨率特征之前，我们首先要解决的一个问题是哪些数据组合可以作为特征。事物纷繁复杂，我们要对每一类事物去建立其对应的特征，这是一个不可能完成的任务。所以在本发明中，我们提出了“采用局部相似性作为特征”的方法。采用这个方法的原因是：我们认为，在进化史上，生物在识别信息时，是沿最节省能量消耗的方向进化。因为对生物 而言，节省能量消耗意味着更高的生存机会。所以，我们把这个思路也引入到机器学习中。Before establishing multi-resolution features, one of the first issues we have to solve is which data combinations can be used as features. Things are complicated, and we have to establish corresponding characteristics for each type of thing. This is an impossible task. Therefore, in the present invention, we propose a method of "using local similarity as a feature". The reason for adopting this method is: We believe that in the history of evolution, when organisms recognize information, they evolve in the direction that saves most energy. Because for living things, saving energy consumption means a higher chance of survival. Therefore, we also introduce this idea into machine learning.
综合上述两个方面，我们提出信息特征的选取标准是：1，这些特征广泛存在于我们的世界中。只有这样我们才能在信息处理过程中复用这些特征，这样最节省能量。2，同一数据，在不同的分辨率下，有不同的数据特征。这样我们才能在不同的分辨率下来对比两者的相似性。Combining the above two aspects, we propose that the selection criteria for information features are: 1. These features are widely present in our world. Only in this way can we reuse these features in the information processing process, which saves energy the most. 2. The same data has different data characteristics under different resolutions. In this way, we can compare the similarities between the two at different resolutions.
1.1多分辨率特征的选择。1.1 Selection of multi-resolution features.
我们提出了如图2所示的多分辨率下信息特征的建立方法。S201是通过滤波器把输入数据分成多个通道。对于图像，这些通道包括针对图形的轮廓、纹理、色调、动态模式等方面做特定的滤波。对于语音，这些通道包括对音频组成、音调变化(一种动态模式)等语音识别方面做滤波。这些预处理方式可以和目前行业内已有的图像、语音预处理方法一样(比如采用小波变换就是一种较好的实现多分辨率特征提取的方法)，这里不再赘述。机器也可以对经过预处理后的数据，直接作为多分辨率数据输入，通过寻找其中的局部相似性来作为特征。也可以把这些数据，采用不同的窗口来细分数据区间，然后在所有数据的所有区间之间寻找其中的局部相似性来作为特征。这里需要指出，不同分辨率的窗口可以是时间窗口或者空间窗口。S202是对每个通道内数据，使用不同大小的窗口来寻找局部相似性。采用大小不同的窗口选取数据，这是模仿人类的注意力区间。通常大窗口的数据对应使用低分辨率，而小窗口内的数据对应使用高分辨率。在S202中，具体步骤可以如下：机器可以逐次使用局部窗口W1，W2，W3，...，Wn，其中W1＜...为自然数)，来对比所有输入下的所有窗口数据中，寻找能够重复出现局部相似性作为特征。We propose a method for establishing information features at multi-resolution as shown in Figure 2. In S201, the input data is divided into multiple channels through a filter. For images, these channels include specific filtering for the contour, texture, tone, and dynamic mode of the graphic. For speech, these channels include filtering for speech recognition such as audio composition and pitch change (a dynamic mode). These pre-processing methods can be the same as the existing image and voice pre-processing methods in the industry (for example, wavelet transform is a better method to achieve multi-resolution feature extraction), and will not be repeated here. The machine can also directly input the pre-processed data as multi-resolution data, and find the local similarity among them as features. You can also use different windows to subdivide the data interval for these data, and then look for the local similarity among all the intervals of all the data as features. It should be pointed out here that the windows of different resolutions can be time windows or spatial windows. In S202, for the data in each channel, windows of different sizes are used to find local similarity. The data is selected using windows of different sizes, which mimics the human attention interval. Generally, the data in the large window corresponds to the use of low resolution, and the data in the small window corresponds to the use of high resolution. In S202, the specific steps can be as follows: the machine can successively use the partial windows W1, W2, W3,..., Wn, where W1<... is a natural number) to compare all window data under all inputs and find Local similarities are repeated as features.
在对比窗口内的数据相似性时，可以使用相似性对比算法。因为相似性对比算法中是非常成熟的算法，本行业专业人员基于公知知识就可以实现，所以这里不再赘述。机器把找到的局部相似特征放入临时记忆库中。每新放入一个局部特征，就赋予其初始记忆值。每发现一个已有的局部特征，就对临时记忆库中的底层特征的记忆值按照记忆曲线增加。临时记 忆库中的信息都遵守临时记忆库的记忆和遗忘机制。那些在临时记忆库中存活下来的底层特征，达到进入长期记忆库阈值后，就可以放入特征图库中，作为长期记忆的特征。长期记忆库可以有多个，它们也遵从自己的记忆和遗忘机制。When comparing the similarity of the data in the window, the similarity comparison algorithm can be used. Because the similarity comparison algorithm is a very mature algorithm, professionals in the industry can implement it based on public knowledge, so I won't repeat it here. The machine puts the found local similar features into a temporary memory bank. Every time a new local feature is added, its initial memory value is assigned. Every time an existing local feature is found, the memory value of the underlying feature in the temporary memory bank is increased according to the memory curve. The information in the temporary memory bank complies with the memory and forgetting mechanism of the temporary memory bank. Those low-level features that survived in the temporary memory bank, after reaching the threshold of entering the long-term memory bank, can be put into the feature library as long-term memory features. There can be multiple long-term memory banks, and they also follow their own memory and forgetting mechanisms.
1.2多分辨率特征的提取的算法。1.2 The algorithm of multi-resolution feature extraction.
机器不仅仅需要建立底层特征图数据库，还需要建立能够提取这些底层特征的模型。在S203中，机器建立的底层特征提取算法。其中一种可能的算法就是寻找局部相似性中相似性对比算法A。当新的信息输入后，机器对信息采用预处理(比如各种坐标基变换后，去除或者压缩部分基底的系数)方法后，机器使用大窗口(低分辨率)和小窗口(高分辨率)来提取窗口内的数据特征。采用窗口相当于模仿人类的注意力，这样，我们才能得到多分辨率特征的同时提取到数据特征在输入中的位置。这些位置会用于我们使用特征来重建输入信息的镜像空间。而这种镜像空间的存储方式，就是我们实现临近激活原则的基础。The machine not only needs to build the underlying feature map database, but also needs to build a model that can extract these underlying features. In S203, the underlying feature extraction algorithm established by the machine. One of the possible algorithms is to find the similarity comparison algorithm A in the local similarity. When new information is input, the machine uses a method of preprocessing the information (for example, after various coordinate base transformations, removing or compressing some of the base coefficients), the machine uses large windows (low resolution) and small windows (high resolution) To extract the data features in the window. Using a window is equivalent to imitating human attention, so that we can obtain multi-resolution features while extracting the position of the data feature in the input. These positions will be used to reconstruct the mirror space of the input information using features. And this storage method of mirror space is the basis for us to realize the principle of proximity activation.
另外一种提取底层特征的算法是神经网络算法B。它是基于多层神经网络的算法模型。这种模型训练好后，比相似度算法的计算效率要高。机器采用选出的信息特征，作为可能的输出来训练多层神经网络。我们可以采用逐层训练方法。在S204中，机器逐次使用局部窗口W1，W2，...，Wn，其中W1＜...为自然数)，来训练算法模型。在优化时，一种是每次增加窗口大小后，就在对应的前一个网络模型上增加零到L(L为自然数)层神经网络层。在S205中，对这个增加了层的神经网络优化时，有两个选择：1，每次只优化增加的零到L(L为自然数)层神经网络层；这样，机器就可以把所有网络模型叠加起来，构成一个有中间输出的整体网络。这样计算效率最高。2，每次都把目前网络复制到新网络，然后优化增加了零到L层的新网络。这样机器最终得到n个神经网络。每个神经网络模型对应一个分辨率。在提取信息中的特征时，机器需要根据本次提取信息的目的，来选用一到多个神经网络。所以，机器可能得到两种提取信息特征的神经网络。一种是多输出层的单个算法网络，其优点是运算资源需求小，但对特征的抽取能力不如后者。另外一种是多个单输出神经网络。这种方式需 要的运算量大，但特征提取更优。需要指出，上述方法可以对图像、语音处理，也可以对任何其他传感器的信息采用类似的方法处理。还需要指出，选用不同的分辨率就是选用不同的窗口，选用不同的特征提取算法。所以提取的特征大小也是不一样的。有些底层特征可能和整个图像一样大。Another algorithm for extracting underlying features is neural network algorithm B. It is an algorithm model based on a multilayer neural network. After this model is trained, it is more efficient than the similarity algorithm. The machine uses the selected information features as possible outputs to train a multilayer neural network. We can use a layer-by-layer training method. In S204, the machine successively uses local windows W1, W2,..., Wn, where W1<... is a natural number) to train the algorithm model. In the optimization, one is to increase the neural network layer from zero to L (L is a natural number) layer on the corresponding previous network model every time the window size is increased. In S205, when optimizing the neural network with the added layer, there are two options: 1. Each time only the added zero to L (L is a natural number) layer of neural network layer is optimized; in this way, the machine can convert all network models Add them together to form an overall network with intermediate output. This is the most efficient calculation. 2. Copy the current network to the new network every time, and then optimize the new network that adds zero to L layers. In this way, the machine finally gets n neural networks. Each neural network model corresponds to a resolution. When extracting features in information, the machine needs to select one or more neural networks according to the purpose of extracting information this time. Therefore, the machine may obtain two kinds of neural networks for extracting information features. One is a single algorithm network with multiple output layers. Its advantage is that it requires less computing resources, but its ability to extract features is not as good as the latter. The other is multiple single-output neural networks. This method requires a large amount of calculation, but the feature extraction is better. It should be pointed out that the above method can process images and voices, and can also process information from any other sensors in a similar way. It should also be pointed out that choosing different resolutions means choosing different windows and different feature extraction algorithms. So the extracted feature size is also different. Some underlying features may be as large as the entire image.
这里，提出一种多层神经网络的训练方法：多分辨率训练方法。多分辨率训练方法是指把输入信息分解成不同的分辨率层。然后使用部分分辨率层来训练神经网络。比如优先使用分辨率低的信息数据来训练神经网络，然后在逐步增加分辨率来训练神经网络。当到达需要的精度后，就可以放弃其他分辨率层的信息。当然，也可以根据识别的目的，来调整使用分辨率层的次序。机器也可以针对不同分辨率下的输入信息特征，按照分辨率分组，单独训练多层神经网络，然后把多个神经网络的输出加权平均，作为总的输出。Here, a training method of a multilayer neural network is proposed: a multi-resolution training method. The multi-resolution training method refers to decomposing the input information into different resolution layers. Then use the partial resolution layer to train the neural network. For example, first use low-resolution information data to train the neural network, and then gradually increase the resolution to train the neural network. When the required accuracy is reached, the information of other resolution layers can be discarded. Of course, the order of using the resolution layers can also be adjusted according to the purpose of identification. The machine can also train multi-layer neural networks separately according to the input information characteristics at different resolutions, group them according to the resolution, and then weight the outputs of multiple neural networks as the total output.
多分辨率提取的计算机实现和相似度对比的计算机实现是目前图像处理非常成熟的算法，也不在本发明申请的权利要求中，所以这里不再赘述。The computer implementation of multi-resolution extraction and the computer implementation of similarity comparison are currently very mature algorithms for image processing, and they are not included in the claims of the present invention, so they will not be repeated here.
1.3静态特征图提取。1.3 Extraction of static feature maps.
需要指出，静态特征图是基于分辨率而建立的，它代表机器根据相似性而对事物的自建分类。比如两张桌子，在粗略的分辨率下它们可能属于同一个分类，而在细致的分辨率下，它们可能是不同的分类。机器只需要对输入信息按照不同的分辨率来提取它们的特征，并把这些特征作为一个整体来代表输入信息就可以了。在输入信息和记忆中信息做相似性对比时，机器采用在不同分辨率下分别做对比。比如同样两个事物，它们都是多个分辨率特征图组合构成的。在对比它们的相似度时，只需要在不同分辨率上对比，就能量化两者的相似度。It needs to be pointed out that the static feature map is established based on the resolution, which represents the machine's self-built classification of things based on similarity. For example, two tables may belong to the same category at a rough resolution, but they may belong to different categories at a fine resolution. The machine only needs to extract the features of the input information according to different resolutions, and take these features as a whole to represent the input information. When the input information is compared with the information in the memory, the machine uses different resolutions to make the comparison. For example, the same two things are composed of multiple resolution feature maps. When comparing their similarity, you only need to compare at different resolutions to quantify the similarity between the two.
1.4动态特征图提取。1.4 Extraction of dynamic feature maps.
在动态图像中，存在两种相似性。一种是其包含的图像和其他过程中的图像的相似性。机器只需要把过程中的特征图和其他过程中的特征图，按照静态特征图提取方法进行就可以了。它们本质上还是静态特征图。但在动态过程中，存在另外一类相似性，那就是运动模式 的相似性。运动模式是指机器忽略运动物体本身的构成细节，而重点对比它们的运动模式。同样，这也存在比较的分辨率问题，比如一个人向我们走过来，或者滑动着过来，或者跑过来，我们在粗略的层面上，甚至不会注意到这些运动模式的差异，所以这个时候，我们认为他们的运动模式是一样的。但当我们增加了分辨率，我们发现滑动过来的人是平稳的运动过来的，而走过来的人和跑过来的人，有各种的运动特征，这些特征包括人体的各个部分的相对运动和人体作为一个整体的整体运动，也包括变化的快慢，所以我们会发现他们的运动模式是不一样的。In dynamic images, there are two similarities. One is the similarity between the images it contains and the images in other processes. The machine only needs to process the feature maps in the process and the feature maps in other processes according to the static feature map extraction method. They are essentially static feature maps. But in the dynamic process, there is another kind of similarity, that is, the similarity of the movement pattern. The motion mode means that the machine ignores the details of the composition of the moving objects, and focuses on comparing their motion modes. Similarly, this also has a comparative resolution problem. For example, a person walks towards us, or slides over, or runs over. At a rough level, we will not even notice the difference in these motion modes, so at this time, We think their exercise patterns are the same. But when we increased the resolution, we found that the person who slid over moved smoothly, and that the person who walked over and the person who ran over had a variety of motion characteristics, including the relative motion and movement of various parts of the human body. The overall movement of the human body as a whole also includes the speed of change, so we will find that their movement patterns are different.
要解决这个问题，本发明申请提出了动态局部相似性对比方法。具体就是，采用不同大小的窗口跟踪事物的不同部分。比如一个人跑过来、走过来还是滑动过来，我们可以采用不同窗口代表不同的分辨率。比如，当我们采用一个大窗口，把整个人作为一个整体时，我们跟踪这个窗口的运动模式，我们就发现这三种情况下，运动模式是一样的。但当我们采用更小的窗口，把人的双手、双腿、头、腰、屁股等部分分别做运动模式提取时，我们就区别出了这三种运动模式的差异。进一步，如果我们对手部采用更多的窗口去关注手部的运动模式，我们就能得到更加精细分辨率的运动模式。除了空间的分辨率，机器还需要确立不同的时间分辨率。比如我们形容大街上人群川流不息，这是一种人群的运动模式。但从更加细微的时间分辨率，我们就能发现早晚上班。时间的人群流动高峰。我们对比不同时间分辨率下的运动轨迹变化，就能得到变化速率。而变化速率是运动在时间上的一个重要动态特征。所以，对运动模式的提取，就是建立在一定时间分辨率和一定空间分辨率基础上，机器通过对大量的动态数据做处理，来寻找常见的动态特征。To solve this problem, the present application proposes a dynamic local similarity comparison method. Specifically, windows of different sizes are used to track different parts of things. For example, if a person runs over, walks over or slides over, we can use different windows to represent different resolutions. For example, when we use a large window to treat the whole person as a whole, we track the movement pattern of this window, and we find that the movement patterns are the same in these three cases. But when we use a smaller window to extract the human hands, legs, head, waist, buttocks and other parts of the movement mode separately, we distinguish the difference of these three movement modes. Furthermore, if we use more windows to focus on the movement pattern of the hand, we can get a finer resolution movement pattern. In addition to the spatial resolution, the machine also needs to establish different temporal resolutions. For example, we describe the constant flow of people on the street, which is a mode of crowd movement. But from a more subtle time resolution, we can find morning and evening shifts. Crowd flow peaks in time. We compare the changes of the motion trajectory at different time resolutions to get the rate of change. The rate of change is an important dynamic feature of movement in time. Therefore, the extraction of motion patterns is based on a certain time resolution and a certain spatial resolution. The machine processes a large amount of dynamic data to find common dynamic features.
机器每发现一个相似的运动模式，机器就把表示这个运动模式的数据放入临时记忆库中，作为动态特征图的候选者，并给这个动态特征图候选者赋予一个记忆值。机器使用大小不同的窗口，对数据迭代使用上述过程，这样机器就能在临时记忆库中得到大量的动态特征图候选者。同静态特征图一样，机器也是使用记忆和遗忘机制来对提取到的动态特征图做优 胜劣汰。那些广泛存在于各种运动物体中的运动模式，会一次次被发现，从而一次次增加记忆值，最终进入长期记忆库中，成为我们长期记忆。Every time the machine finds a similar movement pattern, the machine puts the data representing this movement pattern into the temporary memory bank as a candidate for the dynamic feature map, and assigns a memory value to the candidate for the dynamic feature map. The machine uses windows of different sizes and iteratively uses the above process on the data, so that the machine can obtain a large number of dynamic feature map candidates in the temporary memory bank. Like the static feature map, the machine also uses the memory and forgetting mechanism to survive the fittest on the extracted dynamic feature map. Those movement patterns that are widely present in various moving objects will be discovered again and again, thereby increasing the memory value again and again, and finally entering the long-term memory bank and becoming our long-term memory.
同理，我们可以对图像之外的其他传感器信息做一样的处理。比如对于语音，我们可以采用大小不同的时间窗口作为分辨率，把某些特定的语言属性(某种特征)作为对象，然后对比这个观察对象的变化模式(运动模式)，从中寻找局部变化模式的相似性(比如升调、降调、颤音、***音等)。同理，对于触觉、感觉等数据，也可以采用类似的方法，我们只需要在这些数据的不同维度上，按照不同的分辨尺度，来把某种特征作为观察对象，来寻找观察对象的变化模式之间的相似性，就可以建立起这些对象的动态特征图。需要指出，动态特征图是基于空间和时间双重分辨率而建立的，它代表机器根据动态的相似性而对动态过程的自建分类。它们和被观察对象的静态特征没有关系。In the same way, we can do the same processing for sensor information other than the image. For example, for speech, we can use time windows of different sizes as the resolution, take some specific language attributes (a certain feature) as the object, and then compare the change pattern (motion pattern) of the observed object to find the local change pattern. Similarity (such as rising pitch, falling pitch, vibrato, popping, etc.). In the same way, a similar method can be used for data such as touch and sensation. We only need to use a certain feature as the observation object in different dimensions of the data according to different resolution scales to find the change pattern of the observation object. The similarity between these objects can establish the dynamic feature map of these objects. It should be pointed out that the dynamic feature map is established based on the dual resolution of space and time. It represents the machine's self-built classification of dynamic processes based on the similarity of dynamics. They have nothing to do with the static characteristics of the observed object.
在生活中，由于动态特征和实施这些动态特征的对象无关，所以动态特征在我们生活中使用的重复性非常高。比如不同的对象可能有相似的动态特征。比如小狗的跑步特征，通过一次次在记忆中重复后，那些粗略分辨率的静态和动态特征会一次次得到重复，从而逐步增加了记忆值。这些特征包括：小狗的粗略分辨率特征和小狗运动的粗略分辨率特征。而那些记忆中，具体到每个品种的小狗，每个具体的小狗的特征则记忆值很低，甚至可能被逐渐忘记。不同品种的小狗的运动姿态特征可能也有区别，但这些区别的记忆值比小狗的共有运动特征记忆值低。上述差异来自于：每一只小狗的运动都包含了粗略的小狗图像和粗略的小狗运动特征。这些粗略的小狗图像和粗略的小狗运动特征在所有过程中都是相似的，所以它们能一次次地重复，从而得到更高的记忆值。所以我们在识别小狗的运动时，最高的记忆值首先被激活，那就是：“物体”、“移动”两个特征，然后是“动物”和“跑”两个特征。假如我们从来没有见过猫，第一次看到猫跑步时，我们调用记忆识别时，会调用更加粗略的联系关系：“动物”和“跑”。这是因为我们并不认识猫，但我们从猫的其他特征可以判断这是一个“动物”。在预测猫跑步时，我们会借用关于各种“动物”和“跑”的记忆，通过寻找最相 似记忆，通过泛化来推测。所以，泛化的核心，就是在更加粗略的上次概念下，记忆中的事物和现实中的事物在我们借用的经验里是同类(属性一样)，所以可以采用概念内属性相同就可以替换这个原则来把经验泛化。泛化的基础是利用记忆和遗忘机制来提取事物之间在不同分辨率上的关系。那些看上去没有关系的事物，在其他分辨率上常常是同一类事物。在这种分辨率下，它们和动作特征的连接关系是一样的，所以它们可以用于泛化经验。而且，正是因为动态特征图连接的对象广泛，所以它们的对象通常是很概况的一大类事物，所以机器才能很方便地使用泛化机制(同概念内替换)，利用这些动态特征。所以动态特征和多分辨率关系是我们泛化经验的2个关键工具。In life, since dynamic features have nothing to do with the objects that implement these dynamic features, the repetitive use of dynamic features in our lives is very high. For example, different objects may have similar dynamic characteristics. For example, the running characteristics of a puppy, after repeated in the memory, those rough resolution static and dynamic characteristics will be repeated again and again, thereby gradually increasing the memory value. These features include: the rough resolution feature of the puppy and the rough resolution feature of the puppy movement. In those memories, specific to each breed of puppies, the characteristics of each specific puppies are very low, and may even be gradually forgotten. Different breeds of puppies may have different sports posture characteristics, but the memory value of these differences is lower than the common sports characteristics memory value of the puppies. The above difference comes from the fact that each puppy's movement contains rough puppy images and rough puppy movement characteristics. These rough puppy images and rough puppy motion characteristics are similar in all processes, so they can be repeated again and again to get a higher memory value. So when we recognize the movement of a puppy, the highest memory value is activated first, that is: the two features of "object" and "movement", and then the two features of "animal" and "running". If we have never seen a cat, the first time we see a cat running, when we call memory recognition, we will call a more crude relationship: "animal" and "running." This is because we don't know cats, but we can judge that it is an "animal" from other characteristics of cats. When predicting a cat's running, we will borrow the memories of various "animals" and "running", find the most similar memories, and speculate through generalization. Therefore, the core of generalization is that under a more rough last concept, things in memory and things in reality are the same in our borrowed experience (same attributes), so we can replace this with the same attributes in the concept. Principles to generalize experience. The basis of generalization is to use memory and forgetting mechanisms to extract the relationship between things at different resolutions. Things that seem to be irrelevant are often the same kind of things in other resolutions. At this resolution, the connection between them and action features is the same, so they can be used to generalize experience. Moreover, it is precisely because of the wide range of objects connected by dynamic feature graphs that their objects are usually a large class of things that are very general, so the machine can easily use the generalization mechanism (replacement within the same concept) to utilize these dynamic features. Therefore, dynamic features and multi-resolution relationships are two key tools for our generalization experience.
1.5特征图库的建立。1.5 The establishment of feature library.
在临时记忆库中，我们采用记忆和遗忘机制来维护这些特征图。具体就是：每发现一个相似的特征图候选者，那么这个特征图候选者的记忆值就按照记忆曲线增加其记忆值。同时，临时记忆库中的所有记忆值都按照遗忘曲线，随时间而逐渐递减。如果记忆值递减到零，那么这个特征图候选者就从临时记忆库中删除。如果某个特征图的记忆值增加到预设标准，那么这个特征图就被移入到长期记忆库，成为长期记忆。在这里，记忆值代表对应的特征图能在数据库中存在的时间。记忆值越大，存在的时间越长。记忆值为零时，对应的特征图就被从记忆库中删除。记忆值的增减按照记忆曲线和遗忘曲线来进行。而且不同的数据库可以有不同的记忆和遗忘曲线。机器在训练过程中，在日常生活中，不断使用上述过程，最终获得大量的特征图，这些特征图可以放入快速搜索记忆库中。同理，我们可以对图像之外的其他传感器信息做一样的处理。比如对于语音，我们可以分辨不同语音的频率组成、相对强度作为静态特征，从中寻找局部相似性。对于触觉、感觉等数据，也可以采用类似的方法，我们只需要在这些数据的不同维度上，按照不同的分辨尺度来寻找相似性就可以建立在不同分辨率下的相似性对比结果，从而建立其特征图。In the temporary memory bank, we use memory and forgetting mechanisms to maintain these feature maps. Specifically: every time a similar feature map candidate is found, the memory value of this feature map candidate increases its memory value according to the memory curve. At the same time, all memory values in the temporary memory bank follow the forgetting curve and gradually decrease over time. If the memory value decreases to zero, then the feature map candidate is deleted from the temporary memory bank. If the memory value of a feature map increases to the preset standard, then this feature map is moved to the long-term memory bank and becomes a long-term memory. Here, the memory value represents the time that the corresponding feature map can exist in the database. The larger the memory value, the longer the existence time. When the memory value is zero, the corresponding feature map is deleted from the memory bank. The increase or decrease of the memory value is carried out in accordance with the memory curve and the forgetting curve. And different databases can have different memory and forgetting curves. In the training process of the machine, in daily life, the above process is constantly used, and finally a large number of feature maps are obtained. These feature maps can be put into the fast search memory bank. In the same way, we can do the same processing for sensor information other than the image. For example, for speech, we can distinguish the frequency composition and relative intensity of different speech as static features, and find local similarities from them. Similar methods can be used for tactile and sensory data. We only need to find similarities in different dimensions of these data and at different resolution scales to establish similarity comparison results at different resolutions, thereby establishing Its characteristic map.
2，实现联想激活能力。2. Realize Lenovo activation ability.
实现联想的关键在于建立记忆网络和采用3个激活原则：临近激活原则”、“相似激活原则”和“强记忆激活原则”。The key to the realization of association lies in the establishment of a memory network and the use of three activation principles: the proximity activation principle, the similar activation principle, and the strong memory activation principle.
2.1记忆网络的建立。我们建立记忆时，需要保留事物之间的相似性、时间和空间关系，它们是“临近激活”、“相似性激活”和“强记忆激活”的基础，所以我们采用一种称之为镜像空间的方法来存储数据。当机器从输入中提取了多分辨率信息特征后，机器需要使用这些特征建立镜像空间。机器首先把提取的特征，通过缩放和旋转，按照和原始数据相似度最高的位置、角度和大小，来调整底层特征的位置、角度和大小，把它们和原始数据重叠放置，这样就能保留这些底层特征在时间和空间上的相对位置，并建立镜像空间。2.1 The establishment of memory network. When we build memory, we need to preserve the similarity, time and space relationship between things. They are the basis of "proximity activation", "similarity activation" and "strong memory activation", so we adopt a kind of mirror space Way to store data. After the machine extracts the multi-resolution information features from the input, the machine needs to use these features to build a mirror space. The machine first adjusts the position, angle and size of the underlying features according to the position, angle and size with the highest similarity to the original data by scaling and rotating the extracted features, and places them overlapping the original data so that these can be retained. The relative position of the underlying features in time and space, and the establishment of a mirror space.
机器对信息的记忆是按照输入的时间次序进行记忆的。所以如果机器关注区间在两个空间位置来回切换，那么在记忆中的临近空间并非是实际空间上的临近位置，而是记忆中，两个不断切换的空间位置，因为它们是按照相邻时间次序放置的。机器的“强记忆激活”是通过记忆和遗忘机制来完成的。每个信息输入机器后，都会转换成多分辨率信息特征图。这些特征图随时间关系存储在记忆中。同时，这些多分辨率特征图也会依次在记忆空间中发出自己的激活信号。它不仅仅激活临近记忆，也激活那些和它相似的记忆。需要强调，每个被提取出来的特征都会发出自己的激活信号。也就是说，同一个事物、场景和过程，在不同分辨率上，都可能发出多个对应分辨率特征的激活信号。那些收到激活信号，并且被激活的特征，由于被激活一次，它的记忆值就按照记忆曲线增加。同时，所有在记忆区间中的特征的记忆值，按照各自记忆库的遗忘曲线递减。这样，那些被反复激活的特征，记忆值会随被激活次数增加而增加，从而接收激活信号更强，从而形成一个正反馈，增加自己的记忆。所以那些记忆值高的记忆，要么是反复被激活而逐渐增加了记忆值，要么是一次性给予了足够的记忆值，使得这些信息被记忆下来。每一次新记忆信息产生时，机器在存储这些记忆时，需要确定给这些记忆信息的记忆值。这些记忆被赋予记忆值的原则是：它们的记忆值和存储发生时的激活值成正相关，但不一定是线性关系。它们的激活值，就是它们在存储发生时的激 活强度。The machine remembers the information according to the time sequence of input. So if the machine's attention interval switches back and forth between two spatial positions, then the adjacent space in the memory is not the actual adjacent position in space, but the two constantly switching spatial positions in the memory, because they are in the order of adjacent time. Placed. The "strong memory activation" of the machine is accomplished through the memory and forgetting mechanism. After each information is input to the machine, it will be converted into a multi-resolution information feature map. These feature maps are stored in memory over time. At the same time, these multi-resolution feature maps will in turn send out their own activation signals in the memory space. It activates not only nearby memories, but also memories similar to it. It needs to be emphasized that each extracted feature will emit its own activation signal. In other words, the same thing, scene, and process may emit multiple activation signals corresponding to the resolution characteristics at different resolutions. Those features that receive the activation signal and are activated, because they are activated once, their memory value increases according to the memory curve. At the same time, the memory value of all the features in the memory interval decreases according to the forgetting curve of the respective memory bank. In this way, the memory value of those features that are repeatedly activated will increase with the number of activations, so that the activation signal will be stronger, which will form a positive feedback and increase your own memory. Therefore, those memories with high memory value are either repeatedly activated and gradually increase the memory value, or they are given enough memory value at one time so that the information is memorized. Every time new memory information is generated, the machine needs to determine the memory value for these memory information when storing these memories. The principle that these memories are assigned memory values is that their memory values are positively correlated with the activation values when the storage occurs, but not necessarily linear. Their activation value is their activation intensity when storage occurs.
机器存储的记忆中，有三类数据，每一类都有自己的记忆值。第一类是外部输入的信息特征，包括所有外部传感器输入信息的特征，它们包括视觉、听觉、嗅觉、触觉、味觉、温度、湿度、气压等信息，这些信息和具体环境密切相关，它们按照原始数据的组织方法存储，可以重建立体镜像空间；它们按照记忆和遗忘机制来维护其记忆值。第二类是内部自身信息，包括电量、重力方向、肢体的姿态、各个功能模块运转情况等，这些信息和环境无关，它们的记忆值按照预设程序来设置。第三类是机器需求和需求所处状态的数据，包括安全值、危险值、收益值、损失值、目标达成值、支配值、自身身体状态评估值等数据；也包括由这些需求和需求的状态数据。同时，机器也根据自身需求被满足的情况来产生各种情绪。这些情绪和自身需求被满足的情况之间的关系是通过预置程序来设置的。同时机器也可以反向利用内部情况、外部情况和自身需求被满足的状态之间的关系，来调整情绪产生的预置程序参数，从而利用自己的情绪来影响外界。为了达到在这个目的，我们采用的方法就是：把机器的自身需求类型和情绪类型建立不同的符号代表。在机器的镜像空间发生一个事件时，机器需要把目前的镜像空间存入记忆库中。机器给所有特征图(包括特征图、需求符号和情绪符号)，以及其初始记忆值(和存储发生时的激活值成正相关，但不一定是线性关系)一起存入记忆中。我们把需求符号获得的记忆值，和需求符号一起称为需求状态。There are three types of data in the memory stored by the machine, and each type has its own memory value. The first category is the information characteristics of external input, including the characteristics of all external sensor input information. They include visual, auditory, smell, touch, taste, temperature, humidity, air pressure and other information. These information are closely related to the specific environment. They are based on the original The organization method of data storage can reconstruct the three-dimensional mirror space; they maintain their memory value according to the memory and forgetting mechanism. The second category is internal self-information, including power, gravity direction, body posture, operation of various functional modules, etc. These information have nothing to do with the environment, and their memory values are set according to a preset program. The third category is data on the state of machine needs and needs, including data such as safety value, dangerous value, profit value, loss value, goal achievement value, dominance value, and own body state evaluation value; it also includes data related to these needs and needs. Status data. At the same time, the machine also generates various emotions based on the satisfaction of its own needs. The relationship between these emotions and the situation where one's own needs are met is set through a preset program. At the same time, the machine can also reversely use the relationship between internal conditions, external conditions and the state in which its own needs are met to adjust the preset program parameters of emotion generation, thereby using its own emotions to influence the outside world. In order to achieve this goal, the method we adopted is to establish different symbolic representations of the machine's own demand type and emotional type. When an event occurs in the mirror space of the machine, the machine needs to store the current mirror space in the memory. The machine stores all feature maps (including feature maps, demand symbols, and emotional symbols) and their initial memory values (positively correlated with the activation value when the storage occurs, but not necessarily linear) in memory. We call the memory value obtained by the demand symbol and the demand symbol together as the demand state.
机器的需求可以多种多样，每类需求可以使用一个符号来代表。比如安全和危险、收益和损失、支配权和被支配、尊重和被忽视等。需求类型的差异和多少，不影响本发明申请的权利要求。因为在本发明申请中，所有的需求都是同样的处理方法。The requirements of the machine can be varied, and each type of requirement can be represented by a symbol. Such as safety and danger, gains and losses, dominance and dominance, respect and neglect, etc. The difference and amount of the demand types do not affect the claims of the present application. Because in the present application, all requirements are handled in the same way.
机器的情绪可以多种多样，每类情绪可以使用一个符号来代表。比如比如兴奋、生气、伤心、紧张、焦虑、尴尬、厌倦、冷静、困惑、厌恶、痛苦、嫉妒、恐惧、快乐、浪漫、悲伤、同情和满足等。情绪类型的差异和多少，不影响本发明申请的权利要求。因为在本发明申请中，所有的情绪都是同样的处理方法。The emotions of the machine can be varied, and each type of emotion can be represented by a symbol. For example, such as excitement, anger, sadness, tension, anxiety, embarrassment, boredom, calmness, confusion, disgust, pain, jealousy, fear, happiness, romance, sadness, sympathy and satisfaction. The difference and amount of emotion types do not affect the claims of the present application. Because in the present application, all emotions are handled in the same way.
机器对输入信息赋予的初始激活值，也会通过关系网络传播到机器的需求和情绪数据上，产生了机器对这些信息的本能反应。机器的需求和情绪数据，是一类非常重要的“拟人化”数据。它和外部输入信息和自己内部自身信息密切相关。它们的关系是：当外部数据或者内部数据输入时，机器会产生响应，这些响应又会得到外部反馈和改变内部状态(比如电量变少)。在本发明申请中，我们给机器赋予了类似于人类的需求类型和表示需求被满足的情况的需求获得值。同时，为了更好的和人类交流，我们通过预置程序，把机器需求的满足情况和机器的情绪连接起来。具体实现方法可以是：人类在训练机器的过程中，通过预置的符号(比如语言、动作或者眼神)，在训练中，告诉机器那些环境是安全的，那些环境是危险的，或者可以进一步告诉机器不同的等级。和训练一个孩子一样，告诉它“非常危险”、“比较危险”和“有一点危险”等就可以了。这样，机器就能通过训练，通过记忆和遗忘，逐渐把那些带来危险的环境或者过程中的共有特征，和危险这个内置需求符号的连接强度逐渐增加(因为出现的重复次数增多)。那么当下一次机器处理输入信息时，给予输入信息同样的初始激活值后，有些特征的激活值由于和危险这个符号连接关系紧密，它传递了一个大的激活值给危险这个符号。机器立即意识到危险，会立即根据自己的经验(可以是预置经验或者自己总结的经验)来处理这个危险信息。当然，由于人类已经有大量的经验可以传承，所以在训练中，我们也可以直接告诉机器那些具体的事物或者过程有多大的危险，这是一种给机器预置经验的方法。预置经验可以通过语言来让机器建立记忆帧把危险因素和危险连接起来，也可以通过直接修改机器已有的关系网络来实现(修改对应记忆帧中的危险符号的记忆值)。安全和危险两个值是告诉机器如何识别安全和危险因素，从而学习如果保护自己。收益值和损失值则是告诉机器哪些行为是我们鼓励的，而哪些行为会被惩罚的，这是一个奖励和惩罚***。和训练孩子一样，我们只需要在它做出特定行为后，给予奖励或者惩罚就可以了。或者奖励和惩罚发生时，告诉它原因就可以了。当然我们也可以预置经验(比如事先告诉它那些行为会有奖励，那些会有惩罚，或者直接修改它的大脑神经连接就可以达到目的。人类每达成一个 目标，带来快乐(受到奖励)，这是进化带给我们的礼物，这也是我们这个种族能够不断发展的动力。我们也可以给机器赋予类似的本能动机，让机器建立自我发展的动力。所以，当机器达成一个目标后，既可以通过人类给予的奖励，也可以通过预设程序给机器奖励值，从而激发机器愿意不断去尝试。支配与被支配，是通过收益和损失来告诉机器它可以支配的范围，这个范围随不同环境和不同过程变化而变化，它也是一个奖励和惩罚***。但它和利益损失***的差异在于，利益损失***着眼于行为的结果，而支配与被支配着眼于行为的范围。它和利益损失***采用一样的训练方法。我们也可以把机器自身身体状态评估值和需求与情绪、外部输入信息联系起来，目的是让机器理解自己身体状态评估值和它们之间的联系。比如在下雨天，机器如果发现自己的电量，或者其他性能在快速下降，它把这些记忆存储下来。如果多次重复一样的情况后，机器就会把性能下降和下雨之间建立更加紧密的联系。这些联系在后续机器选择自己的响应过程时，激活下雨这个特征，就会通过联想激活过程传递给损失这个符号较大的损失值。而损失值是机器用于评估选择什么样的响应的指标之一，所以机器就可能倾向于选择排除下雨带来损失值的方案。所以，在本发明中，我们只需要把奖励和惩罚与所有的外部和内部信息一起放入记忆中，机器就能把这些奖励和惩罚信息纳入自己的思维中，而不需要做很多“规则”来告诉机器该怎么识别环境、该做些什么和如何表达情绪(这实际上是不可能完成的任务)。The initial activation value assigned by the machine to the input information will also be propagated to the machine's needs and emotional data through the relationship network, resulting in the machine's instinctive response to this information. The demand and emotional data of machines are a very important type of "anthropomorphic" data. It is closely related to external input information and one's own internal information. Their relationship is: when external data or internal data is input, the machine will respond, and these responses will get external feedback and change the internal state (for example, the battery becomes less). In the application of the present invention, we give the machine a need type similar to that of a human and a demand gain value that represents the situation in which the demand is satisfied. At the same time, in order to better communicate with humans, we use preset programs to connect the satisfaction of the machine's needs with the emotions of the machine. The specific implementation method can be: in the process of training the machine, humans use preset symbols (such as language, action or eye contact) to tell the machine which environments are safe and those environments are dangerous, or can tell the machine further Different grades of machines. Just like training a child, just tell it "very dangerous", "more dangerous" and "a little dangerous". In this way, the machine can gradually increase the connection strength between the dangerous environment or the common features in the process and the built-in demand symbol of danger through training, memory and forgetting (because of the increased number of repetitions). Then when the machine processes the input information next time, after giving the input information the same initial activation value, the activation value of some features is closely connected with the danger symbol, and it transmits a large activation value to the danger symbol. The machine is immediately aware of the danger and will immediately process this dangerous information based on its own experience (which can be preset experience or self-summed experience). Of course, since humans already have a lot of experience to pass on, during training, we can also directly tell the machine how dangerous those specific things or processes are. This is a way to preset experience for the machine. The preset experience can use language to allow the machine to establish a memory frame to connect the dangerous factors with the danger, or it can be realized by directly modifying the existing relationship network of the machine (modifying the memory value of the danger symbol in the corresponding memory frame). The two values of safety and danger tell the machine how to identify safety and danger factors, so as to learn how to protect itself. The gain value and loss value tell the machine which behaviors we encourage and which behaviors will be punished. This is a reward and punishment system. Just like training children, we only need to reward or punish them after they perform certain behaviors. Or when rewards and punishments happen, just tell them why. Of course, we can also preset experience (such as telling it in advance that those behaviors will be rewarded and those will be punished, or directly modify its brain neural connections to achieve the goal. Every time a human reaches a goal, it brings happiness (rewarded), This is the gift that evolution brings to us, and it is also the motivation for our race to continue to develop. We can also give the machine similar instinctive motivation and let the machine build the motivation for self-development. Therefore, when the machine achieves a goal, it can either Through the rewards given by humans, the machine can also be rewarded through the preset program, which inspires the machine to be willing to continue to try. Domination and being dominated is to tell the machine the range it can control through gains and losses. This range varies with different environments and Different processes change and change, it is also a reward and punishment system. But the difference between it and the loss of profit system is that the loss of profit system focuses on the result of the behavior, and the domination and dominance focus on the scope of the behavior. It is used with the loss of profit system. The same training method. We can also associate the machine’s own body state evaluation and needs with emotions and external input information, the purpose is to let the machine understand its own body state evaluation value and the relationship between them. For example, on a rainy day, if the machine finds If its own power or other performance is declining rapidly, it stores these memories. If the same situation is repeated many times, the machine will establish a closer connection between the performance degradation and the rain. These connections are in the subsequent machine selection When the feature of rain is activated in its own response process, it will be transferred to the loss value with the larger symbol of loss through the association activation process. The loss value is one of the indicators used by the machine to evaluate which response to choose, so the machine will It may be inclined to choose a solution that excludes the loss value caused by rain. Therefore, in the present invention, we only need to put the rewards and punishments into memory along with all external and internal information, and the machine can put these rewards and punishments into memory. Incorporate into your own thinking, without having to do a lot of "rules" to tell the machine how to recognize the environment, what to do and how to express emotions (this is actually an impossible task).
机器的情绪是机器和人类交流的重要途径。所以在本发明申请中，我们把机器的情绪也纳入考虑。人类的情绪反应，是对自己需求是否被满足的一种与生俱来的反应，但通过后天的学习，我们逐步学会了调整这种反应，控制这种反应，甚至隐藏这种反应。同理，我们通过预置程序，把机器的情绪和机器的需求是否被满足联系起来。比如，识别到危险时，机器的情绪是“担心”、“畏惧”和“恐惧”，这要看危险程度有多大。比如机器的各个内部运转参数都在正确的区间，带给机器的是“舒适”、“放松”等情绪。如果有些参数脱离了正确的区间(相当于机器生病了)，机器的表情可能是“难受”和“担心”。所以，采用这样的方法， 我们可以把人类拥有的所有情绪，赋予给机器。而情绪本身，是通过机器的面部表情和肢体语言来表达的。同理，机器的这些本能情绪，会受到奖励和惩罚机制的调整。机器在生活中，在不同的环境或者过程中，训练者会不断告诉机器，它的情绪表现，哪些受到奖励，哪些受到惩罚。也可以直接告诉它，在特定或者过程中，合适的情绪是什么。当然也可以直接修改它的神经网络连接来调整它的情绪反应。所以，通过这样的方式，机器可以把情绪调整到和人类相似程度，而进一步，由于情绪和其他记忆是存放在一起的，在同一个记忆中。当机器需要某种结果时，它会模仿带来这个结果的记忆。比如某一类行为带来某种结果能够重复出现，那么机器就会模仿包含这类行为的记忆，当然也会模仿这些记忆中的情绪，所以它会为了某种目的而调整自己的情绪。这是一种情绪利用的方式。The emotion of the machine is an important way for the machine to communicate with human beings. Therefore, in the application of the present invention, we also take the emotion of the machine into consideration. Human emotional response is an innate response to whether one's own needs are met, but through acquired learning, we have gradually learned to adjust this response, control this response, and even hide this response. In the same way, we use preset programs to link the emotions of the machine with whether the needs of the machine are met. For example, when a danger is identified, the emotions of the machine are "worry", "fear" and "fear", depending on the degree of danger. For example, the various internal operating parameters of the machine are in the correct range, which gives the machine emotions such as "comfort" and "relaxation". If some parameters are out of the correct range (equivalent to the machine is sick), the machine's expression may be "uncomfortable" and "worry". Therefore, using this method, we can assign all the emotions that humans have to the machine. The emotion itself is expressed through the facial expressions and body language of the machine. In the same way, these instinctive emotions of the machine will be adjusted by the reward and punishment mechanism. In the machine's life, in different environments or processes, the trainer will continue to tell the machine its emotional performance, which ones are rewarded, and which ones are punished. You can also directly tell it what the appropriate emotion is in a particular or process. Of course, you can directly modify its neural network connection to adjust its emotional response. Therefore, in this way, the machine can adjust emotions to a degree similar to that of humans, and further, because emotions and other memories are stored together, in the same memory. When a machine needs a certain result, it will imitate the memory that brought that result. For example, a certain type of behavior brings a certain result that can be repeated, then the machine will imitate the memory that contains this type of behavior, and of course it will also imitate the emotions in these memories, so it will adjust its emotions for a certain purpose. This is a way of using emotions.
需要指出，通过本发明申请所提出的方法而建立的机器智能，其思维和情绪对人类而言，是可见的可控的，是完全可以理解的，它们是通过联想激活来连接起来的。所以这样的机器智能对人类是不会带来危险的，这也是本发明申请所提出的通用人工智能实现方法的一个特征。It should be pointed out that the thoughts and emotions of the machine intelligence established by the method proposed by the present application are visible and controllable to humans, and are completely understandable. They are connected through association activation. Therefore, such machine intelligence will not bring danger to humans, which is also a feature of the general artificial intelligence implementation method proposed in the present application.
在本发明中，机器对镜像空间的存储采用记忆筛选机制：事件驱动机制和临时记忆库机制。在镜像空间中，每发生一次事件，机器就把这个镜像空间做一个快照，保存下来。保存下来的内容包括镜像空间中的特征(包括信息、机器状态、需求和情绪)和它们的记忆值。它们的记忆值和存储发生时的激活值成正相关，但不一定是线性关系。一次镜像空间的快照存储数据，我们称之为一个记忆帧。它们像电影帧一样，通过多个帧连续回放，我们就能重现记忆发生时的动态场景。所不同的是，记忆帧中的信息可能会随时间而被遗忘。镜像空间中发生一次事件，是指镜像空间中特征组合和前一个镜像空间相比较，发生了超过预设值的相似度的改变，或者镜像空间中的记忆值发生了超过预设值的改变。记忆库就是指存放这些记忆帧的数据库。而临时记忆库是记忆库的一种，其目的是对记忆帧存储的信息做筛选。在临时记忆库中，如果某一个记忆帧里面包含有记忆值达到预设标准的特征，那么这个记忆帧 就可以被移到长期记忆库汇中保存。本发明申请中，我们采用有限容量的堆栈来限制临时记忆库容量的大小，并在临时记忆库中采用快速记忆和快速遗忘的方式，来对准备放入长期记忆库中的材料进行筛选。机器在面对大量的输入信息时，那些已经习以为常的事物、场景和过程，或者远离关注点的事物、场景和过程，机器对它们缺乏深入分析的动机，所以机器可能不去识别这些数据，或者赋予给它们的激活值很低。机器在按照事件驱动的方式把信息存入临时记忆库时，机器对每个信息特征赋予的记忆值和其存储发生时的激活值正相关。那些记忆值低的记忆有可能很快就从临时记忆库中被忘记，而不会进入长期记忆库。这样我们只需要把那些我们关注的信息放入长期记忆库，而不用把每天琐碎的、不需要再提取连接关系的事物都记忆下来。另外，因为临时记忆库容量有限制，所以临时记忆库也会因为堆栈容量接近饱和而被动加快遗忘速度。In the present invention, the machine uses a memory screening mechanism for the storage of the mirror space: an event-driven mechanism and a temporary memory library mechanism. In the mirror space, every time an event occurs, the machine takes a snapshot of the mirror space and saves it. The saved content includes the features in the mirror space (including information, machine states, needs, and emotions) and their memory values. Their memory value is positively related to the activation value when the storage occurs, but not necessarily linear. A snapshot of the mirror space stores data, which we call a memory frame. They are like movie frames. Through continuous playback of multiple frames, we can reproduce the dynamic scene when the memory occurs. The difference is that the information in the memory frame may be forgotten over time. An event in the mirror space means that the feature combination in the mirror space is compared with the previous mirror space, and the similarity changes beyond the preset value, or the memory value in the mirror space changes beyond the preset value. Memory bank refers to the database that stores these memory frames. The temporary memory bank is a kind of memory bank, and its purpose is to filter the information stored in the memory frame. In the temporary memory bank, if a memory frame contains features whose memory value reaches the preset standard, then this memory frame can be moved to the long-term memory bank for storage. In the application of the present invention, we use a limited-capacity stack to limit the size of the temporary memory bank, and use the fast memory and fast forgetting methods in the temporary memory bank to screen the materials to be put into the long-term memory bank. When the machine is faced with a large amount of input information, those things, scenes and processes that are already accustomed to, or things, scenes and processes far away from the focus of attention, the machine lacks the motivation for in-depth analysis of them, so the machine may not recognize these data, or The activation value assigned to them is very low. When the machine stores information in the temporary memory bank in an event-driven manner, the memory value assigned by the machine to each information feature is positively correlated with the activation value when the storage occurs. Those memories with low memory value may soon be forgotten from the temporary memory bank and will not enter the long-term memory bank. In this way, we only need to put the information that we care about into the long-term memory, instead of memorizing the trivial things that do not need to extract the connection relationship every day. In addition, because the capacity of the temporary memory bank is limited, the temporary memory bank will passively accelerate the forgetting speed because the stack capacity is close to saturation.
如果我们把记忆看作是一个包含了无数信息特征图的立体空间，那么关系网络，就是这个空间中的脉络。这些脉络的出现，是因为记忆和遗忘机制，那些不能被反复激活的关系都被遗忘了，而那些能得到反复激活的关系得到了加强。那些通过粗大的关系脉络连接起来的特征图就组成了概念。它连接同类信息的图像、语音、文字或者其他任何表达形式。由于这些表达形式频繁出现在一起，并频繁相互转换，所以它们之间的连接更加紧密。最紧密的局域的连接关系就构成了基础概念(包括静态特征图及其语言，动态特征图及其语言)；比基础概念松散一点的是静态扩展概念和动态概念扩展概念(包括代表关系的概念和过程特征图)，比概念松散就是记忆。在关系网络中，那些静态特征图(或者概念)通常就是小零件，而那些动态特征图(包括表示关系的概念)就是连接件，而那些过程特征就是大框架，它是多个小零件(静态对象)和、连接件(动态特征)，按照一定的时间和空间次序组织起来的。过程特征是我们可以借鉴的大框架。而动态特征图(包括表示关系的概念)就是可以具体实施经验泛化的工具，而静态特征图(或者概念)就是在泛化中被替代的对象。泛化过程是通过对激活值高的信息流，把现实对象和记忆中的对象替换后，重新组织起来实施模仿的过程。If we regard memory as a three-dimensional space containing countless information feature maps, then the network of relationships is the context in this space. The emergence of these contexts is due to the memory and forgetting mechanism. Those relationships that cannot be repeatedly activated are forgotten, and those relationships that can be repeatedly activated are strengthened. Those feature maps that are connected through the coarse relationship context constitute the concept. It connects images, voice, text or any other form of expression of similar information. Because these forms of expression frequently appear together and frequently switch to each other, the connection between them is closer. The tightest local connection relationship constitutes the basic concept (including static feature map and its language, dynamic feature map and its language); a bit looser than the basic concept is the static expansion concept and the dynamic concept expansion concept (including the representative relationship Concept and process characteristic diagram), looser than concept is memory. In the relational network, those static feature maps (or concepts) are usually small parts, and those dynamic feature maps (including concepts that represent relationships) are connectors, and those process features are large frames, which are multiple small parts (static Objects) and connectors (dynamic features) are organized according to a certain time and space sequence. Process characteristics are a large framework that we can learn from. The dynamic feature map (including the concept that represents the relationship) is a tool that can implement empirical generalization, and the static feature map (or concept) is the object to be replaced in the generalization. The generalization process is a process of reorganizing and implementing imitation after replacing the real objects and the objects in memory through the information flow with high activation value.
为了提高搜索效率，我们可以把关系网络从记忆中分离出来，建立一个单独的网络。一种可能的方法就是：把每个记忆帧中的特征图先建立连接线，它们的连接值是每个连接线两端的特征图的记忆值的函数。然后对每个特征图发出的连接值归一化。这样就会导致两个特征图彼此之间的连接值不是对称的。然后把记忆帧之间的相似特征图按照相似度的程度连接起来，连接值就是相似度。通过上述步骤后，获得的网络就是从记忆库中提取出来的认知网络。我们可以把认知网络单独放到一个快速搜索库(记忆库的一种)，用于一些需要快速的本能反应中，比如自动驾驶应用中，或者一些只需要简单的智能应用中(比如生产线)。这种关系网络中的记忆和遗忘采用对连接值作记忆和遗忘机制：关系每使用一次，连接值就按照记忆曲线增加。而所有连接值都按照遗忘曲线随时间而递减。还有一种方法就是把包含经常调用的概念和过程特征的记忆，保持记忆库的组织形式，放到一个单独快速搜索库中。在这些记忆中，具体的细节可能已经被忘记，保留下来的都是强记忆值记忆。这些强记忆值记忆通过联想激活，就能快速调用相关的概念和过程特征。这样能加快机器的记忆搜索效率。这种方法也可以用于那些需要快速反应的应用中，比如自动驾驶应用中，或者一些只需要简单的智能应用中(比如生产线)。In order to improve search efficiency, we can separate the relationship network from memory and build a separate network. One possible method is to first establish a connection line for the feature maps in each memory frame, and their connection value is a function of the memory value of the feature maps at both ends of each connection line. Then normalize the connection value sent by each feature map. This will cause the connection values between the two feature maps to be non-symmetrical. Then the similar feature maps between the memory frames are connected according to the degree of similarity, and the connection value is the similarity. After passing the above steps, the obtained network is the cognitive network extracted from the memory bank. We can put the cognitive network alone in a quick search library (a kind of memory library) for some instinctive responses that require fast, such as in autonomous driving applications, or in some simple smart applications (such as production lines) . The memory and forgetting in this kind of relationship network adopts the mechanism of remembering and forgetting the connection value: each time the relationship is used, the connection value increases according to the memory curve. And all the connected values decrease with time according to the forgetting curve. Another method is to put the memory containing frequently called concepts and process features, maintaining the organizational form of the memory bank, into a separate quick search library. In these memories, specific details may have been forgotten, and what remains are memories with strong memory values. These strong memory value memories are activated by association, and related concepts and process features can be quickly recalled. This can speed up the memory search efficiency of the machine. This method can also be used in applications that require rapid response, such as autonomous driving applications, or in some simple smart applications (such as production lines).
需要指出，建立单独的关系网络可以采用很多形式，但只要这种关系网络是基于本发明申请提出的基础假设之上的，它们都是本发明申请中关系网络的一种变形方式，和本发明申请中所提出的关系网络并没有本质区别，所以它们依然处于本发明申请的权利要求中。It should be pointed out that the establishment of a separate relationship network can take many forms, but as long as this relationship network is based on the basic assumptions proposed in the present application, they are all a variant of the relationship network in the present application, and the present invention There is no essential difference between the relationship networks proposed in the application, so they are still in the claims of the present application.
2.2联想能力的实现。2.2 The realization of Lenovo's ability.
在有了记忆空间后，通过“临近激活”、“相似性激活”和“强记忆激活”，机器就可以实现联想能力了。任何能够实现“临近激活”、“相似性激活”和“强记忆激活”的算法，都可以应用于本发明申请中。这里，我们提出几种实现上述激活原则的方法(但不限于这些方法)：After having memory space, through "proximity activation", "similarity activation" and "strong memory activation", the machine can realize the ability of association. Any algorithm that can realize "proximity activation", "similarity activation" and "strong memory activation" can be applied to the application of the present invention. Here, we propose several methods to realize the above activation principle (but not limited to these methods):
方法1：采用记忆值(实数)来代表神经元或者突触的数量；使用激活值来代表特征发出的 激活电信号强度；使用特定的编码来代表不同的特征发出的不同模式激活信号；使用总线来代替整个记忆空间来传播激活值；使用三维立体坐标点位置来代表不同特征信息在记忆空间中的位置，并使用空间距离(激活源和接收特征之间的空间距离)来计算衰减量。当输入特征通过通用激励模块把自己对应编码的激活电信号发布到总线上，而且使用编码中的数字来代表自己被赋予的初始强度，记忆中的特征可以通过周期性的对总线信息读取，来接收总线上的信息，并计算应该的衰减量。如果存在和自己相似的激活信息，比如可能属于一个大类，或者属于一个子类等，那么就有不同的接收能力。如果收到的激活信号通过自己的接收通道后，得到的激活值超过自己预设的激活阈值，那么这个特征就把收到的激活值作为初始值，并激活自己。通常可能存在多个输入特征同时激活一个小记忆区间的情况，比如一张“餐桌”有多个不同分辨率的特征，它们依次通过记忆区间的总线，可能激活多个小的区间。每个区间都可能有多个关于“餐桌”的特征被激活。这些小区间中集中的特征图，彼此再次激活时，通过临近激活又给彼此赋予激活值。所以它们的激活值就可能值记忆空间中“凸显”出来。而在它们共同的临近激活作用下，某一个小区间可能激活当时餐桌上一个“美味”的蛋糕的记忆。这是因为蛋糕通过味觉传感器相关的预置程序，给食物相关的“正面需求”符号赋予了很高的激活值。当记忆存储发生时，食物相关的“正面需求”符号的激活值按照正相关转化为记忆值(不一定是线性关系)。所以，在这里，食物相关的“正面需求”符号(比如对美味的需求)是一个强记忆。它存在“餐桌”记忆附近，由于它的记忆值高，所以按照“强记忆激活”原则，它也获得了很高的激活值。当它被激活后，和它很临近的记忆“蛋糕”(因为两者可能是同时存储到记忆中的)也可能被激活。另外，并且“蛋糕”和“对美味的需求被满足”常常一起被激活，在记忆中，它们的记忆越来越强，所以任何时候，但一个被激活后，另外一个也常常被激活，我们就在“蛋糕”和“对美味的需求被满足”之间建立了正确的连接。另外，在本发明申请中，我们通过预置一套机器的需求被满足的情况和机器的情绪之间的预置程序，来实现机器情绪。这套预置程序在和“对美味的需求被满足”输入激励下，会 向“愉悦”、“满足”等情绪符号发出较高的定向激活值。于是机器的“愉悦”和“满足”等情绪符号获得了较高的激活值。当存储发生时，这些激活值也是按照正相关方式转化为记忆值(不一定是线性的)，所以在这些记忆中，情绪也被记忆下来。当机器激活了“蛋糕”和“对美味的需求被满足”的记忆后，这些情绪符号也一并可能被激活，从而使得机器体会到了“愉悦”、“满足”等情绪。Method 1: Use the memory value (real number) to represent the number of neurons or synapses; use the activation value to represent the strength of the activation electrical signal emitted by the feature; use a specific code to represent the different mode activation signals emitted by different features; use the bus Instead of the entire memory space to propagate the activation value; use the three-dimensional coordinate point position to represent the position of different feature information in the memory space, and use the spatial distance (the spatial distance between the activation source and the receiving feature) to calculate the attenuation. When the input feature releases its own corresponding coded activation electrical signal to the bus through the universal excitation module, and uses the number in the code to represent its initial strength, the feature in memory can be read by periodically reading the bus information. To receive the information on the bus, and calculate the amount of attenuation that should be. If there is activation information similar to yourself, for example, it may belong to a large category or a subcategory, etc., then there are different receiving capabilities. If the activation value obtained after the received activation signal passes through its own receiving channel exceeds its preset activation threshold, then this feature uses the received activation value as the initial value and activates itself. Usually, there may be multiple input features simultaneously activating a small memory interval. For example, a "dining table" has multiple features with different resolutions, and they pass through the bus of the memory interval in turn, possibly activating multiple small intervals. Each section may have multiple features related to the "dining table" activated. When the feature maps concentrated in these cells are activated again, they are given activation values to each other through adjacent activations. Therefore, their activation values may be "highlighted" in the memory space. And under their common proximity activation, a certain neighborhood may activate the memory of a "delicious" cake on the table at that time. This is because the cake gives a high activation value to the food-related "positive demand" symbol through the preset program related to the taste sensor. When memory storage occurs, the activation value of the food-related "positive demand" symbol is converted into a memory value according to a positive correlation (not necessarily a linear relationship). So, here, the symbol of "positive demand" related to food (such as the demand for delicious food) is a strong memory. It exists near the memory of the "dining table". Because of its high memory value, it also obtains a high activation value according to the principle of "strong memory activation". When it is activated, the memory "cake" that is very close to it (because the two may be stored in the memory at the same time) may also be activated. In addition, "cake" and "the need for deliciousness are satisfied" are often activated together. In memory, their memory is getting stronger and stronger. So anytime, after one is activated, the other is often activated. We The right connection is established between "cake" and "the demand for deliciousness is satisfied". In addition, in the application of the present invention, we realize the machine emotions by presetting a set of preset procedures between the situation where the requirements of the machine are satisfied and the emotions of the machine. This set of preset programs will send out higher directional activation values to emotional symbols such as "pleasure" and "satisfaction" under the stimulation of the input of "demand for deliciousness is satisfied". As a result, emotional symbols such as "pleasure" and "satisfaction" of the machine have obtained higher activation values. When storage occurs, these activation values are also converted into memory values (not necessarily linear) in a positive correlation manner, so in these memories, emotions are also memorized. When the machine activates the memory of "cake" and "the demand for deliciousness is satisfied", these emotional symbols may also be activated, so that the machine can experience emotions such as "pleasure" and "satisfaction".
当机器需要寻求“愉悦”、“满足”等情绪时(比如给机器赋予这样的本能需求)，机器就会寻找关于“愉悦”、“满足”相关的记忆，它可能就会激活“蛋糕”、“餐桌”等记忆。这些记忆就可能成为一个响应目标，机器就有可能通过这些目标联想得到“蛋糕”和“餐桌”的经验，进而通过泛化能力泛化这些经验，在现有条件下通过模仿过去的经验，把泛化之后的各种过程特征组织起来，通过分段模仿，把这个组织起来的过程逐层细分成大量的中间环节目标，再一步一步去实现这些中间环节目标。比如去完成订购“蛋糕”、寻找“餐桌”并满足自己的需求的过程。When the machine needs to seek emotions such as "pleasure" and "satisfaction" (for example, to give the machine such an instinctive need), the machine will look for memories related to "pleasure" and "satisfaction", and it may activate the "cake", "Dining table" and other memories. These memories may become a response target, and the machine may obtain the experience of "cake" and "dining table" through these target associations, and then generalize these experiences through generalization ability, and under existing conditions, by imitating past experience, After the generalization, various process characteristics are organized, and through segmented imitation, the organized process is subdivided into a large number of intermediate link goals, and then these intermediate link goals are achieved step by step. For example, to complete the process of ordering "cakes", finding "tables" and satisfying their own needs.
以上过程是一种分布式计算的过程。这个方法还可以变成2层结构。比如每一小段记忆放置一个和总线连接的计算模块作为和总线信息交换的门户，这个计算模块承担把辖区外的激活信号识别后，决定是否传入辖区内。也负责把辖区内的激活，再次传到总线上去。这样做的目的是减少计算模块的数量。当然，这个结构还可以迭代自己，采用类似的多层结构来进一步减小计算模块。The above process is a process of distributed computing. This method can also be turned into a two-layer structure. For example, in each small segment of memory, a computing module connected to the bus is placed as a gateway for information exchange with the bus. This computing module is responsible for identifying the activation signal outside the jurisdiction and deciding whether to transfer it to the jurisdiction. It is also responsible for transferring the activation in the jurisdiction to the bus again. The purpose of this is to reduce the number of computing modules. Of course, this structure can also iterate itself, using a similar multi-layer structure to further reduce the calculation module.
方法2：方法2是一种集中计算方法。就是采用专门的计算模块来搜索记忆(记忆搜索模块)。每当发现一个多分辨率下的输入信息特征后，机器直接激活目前时间上最近的记忆，并按照它们的记忆值赋予其相应的激活值。这就完成了临近激活和强记忆激活。也在记忆中直接去寻找相关相似特征，找到后，按照相似度直接给这些特征赋予激活值。相似度既可以采用现场对比的方法，也可以采用预编码逐层分类的方法。Method 2: Method 2 is a centralized calculation method. It uses a special calculation module to search memory (memory search module). Whenever a feature of input information under multi-resolution is found, the machine directly activates the most recent memory in the current time and assigns its corresponding activation value according to their memory value. This completes the proximity activation and strong memory activation. It also directly searches for related similar features in the memory, and after finding, directly assigns activation values to these features according to the similarity. The degree of similarity can either use the method of on-site comparison or the method of precoding layer by layer classification.
那些被激活特征图再次发出激活电信号时，记忆搜索模块可以采用一样的方法。通过 对发起激活的特征图，搜索附近的记忆发起临近激活，搜索更远的那些拥有高记忆值的记忆发起强记忆激活，通过搜索其他记忆中相似的特征发起相似性激活。而且每个被激活的模块，发出的激活电信号有自己的编码和强度信息。这个过程可以反复迭代下去。The memory search module can use the same method when the activated feature map sends out the activation signal again. Through the feature map that initiates activation, searching for nearby memories initiates proximity activation, searching for those further away with high memory value initiates strong memory activation, and searches for similar features in other memories to initiate similarity activation. And each activated module, the activation signal sent out has its own code and intensity information. This process can be iterated over and over again.
方法3：方法3是一种混合模式。机器通过记忆搜索模块完成相似性激活搜索后，进一步的激活可以通过在每段记忆的局部网络中进行。通过记忆中特征之间建立的连接网络来实现临近激活和强记忆激活。这种局部网络的一种实现方法是：记忆空间中每个特征都和临近特征之间建立连接神经，当自己被激活后，通过这些连接线可以把激活值传递出去，这就是临近激活。而两个特征之间的传递系数和两个特征的记忆值正相关，这就是强记忆激活。Method 3: Method 3 is a hybrid mode. After the machine completes the similarity activation search through the memory search module, further activation can be carried out in the local network of each memory. Proximity activation and strong memory activation are realized through the connection network established between the features in the memory. One way to realize this kind of local network is: each feature in the memory space establishes a connection nerve with the neighboring feature, and when it is activated, the activation value can be transmitted through these connection lines, which is adjacent activation. The transfer coefficient between the two features is positively correlated with the memory value of the two features, which is the strong memory activation.
以上3种方法都可以实现在记忆网络中的联想能力。而能够实现“临近激活”、“相似性激活”和“强记忆激活”的方法很多，各种具体方式都可以建立在本行内的公知知识上。所以，本发明申请所列举的3种实现方式不是限制范围，而是演示其中的基本原理。任何其他方式，只要是建立在“临近激活”、“相似性激活”和“强记忆激活”3个原则基础上的联想激活实现算法，都涉及到本发明申请的权利要求。由于在机器中，我们可以采用数值来代表一个信息在记忆中的强度，采用编码来代表被激活电信号的类别，采用总线来代表激活电信号传播空间，采用立体坐标距离来模拟传播损耗，所以机器的联想查找速度可以远高于大脑的神经激活工作方式。The above three methods can all realize the ability of association in the memory network. There are many ways to achieve "proximity activation", "similarity activation" and "strong memory activation", and various specific methods can be built on the public knowledge in the industry. Therefore, the three implementations listed in the application of the present invention are not to limit the scope, but to demonstrate the basic principles. Any other method, as long as it is an association activation realization algorithm based on the three principles of "proximity activation", "similarity activation" and "strong memory activation", is related to the claims of the present application. Because in the machine, we can use numerical values to represent the strength of a piece of information in memory, use codes to represent the types of activated electrical signals, use buses to represent the propagation space of activated electrical signals, and use three-dimensional coordinate distances to simulate propagation loss, so The machine's association search speed can be much higher than the brain's neural activation work.
在比较输入特征图和关系网络中的特征图的相似性过程中，机器可能需要处理大小缩放和角度匹配的问题。一种处理方法包括：(1)机器把各种角度的特征图都记忆下来。记忆中的特征图，是通过对每一次输入信息提取底层特征后建立的简图。它们是在关系提取机制下保留下来相似事物的共有特征。虽然它们彼此相似，但它们可能存在不同的观察角度。机器把生活中同一个事物，但不同角度的特征图都记忆下来，构成不同的特征图，但它们可以通过学习来归属于同一个概念。(2)机器用所有角度的视图，重叠这些特征图的共有部分，模仿它们的原始数据，把它们组合起来，构成一个立体特征图。(3)在机器内部嵌入对立体 图像做大小缩放和空间旋转后的视图变化程序。这一步是业内已经非常成熟的技术，这里不再赘述。(4)机器在记忆中寻找相似的底层特征时，包括了在记忆中寻找经过空间旋转后能匹配的特征图。同时机器把目前角度的特征图存入记忆，保留原始视角。后续再次有类似视角的底层特征输入时，就能快速的搜索到。所以这种方法下，机器是采用了不同视角记忆和进行空间角度旋转相结合的方法来寻找相似特征图，这会带来我们对熟悉视角识别更快的现象。当然，机器也可以只使用空间角度旋转后进行相似度对比的方法。In the process of comparing the similarity between the input feature map and the feature map in the relational network, the machine may need to deal with the problems of size scaling and angle matching. One processing method includes: (1) The machine memorizes feature maps of various angles. The feature map in memory is a simplified map created by extracting the underlying features of each input information. They are the common features of similar things retained under the relationship extraction mechanism. Although they are similar to each other, they may have different viewing angles. The machine memorizes the feature maps of the same thing in life but from different angles to form different feature maps, but they can belong to the same concept through learning. (2) The machine uses views from all angles, overlaps the common parts of these feature maps, imitates their original data, and combines them to form a three-dimensional feature map. (3) Embedded in the machine the view change program after the size scaling and spatial rotation of the stereo image. This step is a very mature technology in the industry, so I won't repeat it here. (4) When the machine searches for similar underlying features in the memory, it includes searching for a feature map that can be matched after spatial rotation in the memory. At the same time, the machine saves the feature map of the current angle in memory, keeping the original angle of view. When the underlying features with similar perspectives are input again later, they can be quickly searched. Therefore, in this method, the machine uses a combination of different perspective memory and spatial angle rotation to find similar feature maps, which will bring us to the phenomenon of faster recognition of familiar perspectives. Of course, the machine can also only use the method of comparing the similarity after rotating the space angle.
3，实现泛化能力。3. Realize generalization ability.
泛化能力是建立在多分辨率概念的基础上的，所以机器需要首先需求建立不同分辨率下的概念。The generalization ability is based on the concept of multi-resolution, so the machine needs to establish the concept of different resolutions first.
3.1基础概念的建立。我们的祖先，发明了语言，并使用这些语言来代表那些通过对比相似性而建立的分类，比如石头、树、无花果、兔子和狮子等和生活密切相关的事物。也使用语言来代表那些通过对比相似性而建立的动态分类，比如跑、跳、敲、磨、刨、投掷和流动等和生活密切相关的动态模式。在有了这些语言后，就能通过一定的组织方式来组织这些语言，表达思想，这是一个约定成俗的过程。这些语言符号和它们所代表的事物和动作一起，构成了基础概念。3.1 The establishment of basic concepts. Our ancestors invented languages and used these languages to represent those categories established by comparing similarities, such as stones, trees, figs, rabbits, and lions that are closely related to life. Language is also used to represent those dynamic classifications established by comparing similarities, such as running, jumping, knocking, grinding, planing, throwing, and flowing dynamic patterns closely related to life. After having these languages, we can organize these languages and express our thoughts through certain organizational methods. This is a process of convention. These language symbols, together with the things and actions they represent, constitute the basic concepts.
机器建立概念的具体方法，采用和人类一样的方式。比如，当某一个图像特征图输入到机器中时，我们同步赋予其代表这个图像特征图的语言。这两个被关注的信息是作为相邻信息被记忆的，它们存在临近关系。那么机器在通过多次重复后，就能在关系网络中，把这个图像特征图和对应的语言特征图建立起非常紧密的联系，它们能彼此激活，产生联想，并且能通过相似性激活存在于不同记忆区间的语言或者其他形式的信息，这就是概念的联想。The concrete method of the machine to establish the concept, adopts the same way as the human. For example, when a certain image feature map is input into the machine, we give it a language that represents the image feature map simultaneously. The two concerned information is memorized as adjacent information, and they have a close relationship. Then the machine can establish a very close relationship between this image feature map and the corresponding language feature map in the relationship network after repeated repetitions. They can activate each other, generate associations, and can exist in the relationship through similarity activation. Language or other forms of information in different memory intervals are conceptual associations.
机器建立概念的方法还可以是人为的植入记忆。把概念的语言符号和概念包含的其他形式内容(这些是人类总结的经验)，一起放在记忆中相邻的位置，并赋予其较高的记忆值，那么语言符号和其他形式信息之间、其他形式信息之间能够彼此激活，产生联想，它们就构 成了概念。The way machines build concepts can also be artificially implanted memories. Put the linguistic signs of the concept and the other forms of content contained in the concept (these are the experiences summarized by humans) together in the adjacent position in the memory, and give them a higher memory value, then the linguistic signs and other forms of information, Other forms of information can activate each other and produce associations, and they constitute concepts.
由于那些存在于不同记忆中的相似图像特征图，它们在不同记忆中的相似度，可能没有语言在不同记忆中的相似度高。当我们通过图像和语言把不同的记忆串接起来后，那些语言符号(比如语音或者文字)由于使用频繁(导致记忆值高)，彼此间相似度高(导致记忆间激活值传递系数大)，那么同一个概念包含的信息中(比如各种苹果图像、各种苹果语音和各种苹果文字)，语言符号很可能拥有最高的记忆值(因为使用频繁和相似度高)。在记忆中搜索概念时，机器常常会首先找到语言符号，并使用语言符号来代表概念。当我们使用语言来表达思想时，其实是使用一种方法(语音或者文字符号)来依次激活和这些语言连接紧密的其他形式信息(比如图像、味道、感觉、声音等)，正是这些其他形式的信息流让我们理解了语言所代表的信息。需要特别指出，通过语言信息流建立的图像和其他感知形式信息流，可以成为我们记忆中的一部分，就好像它们真的发生过一样。这是因为语言信息流激活的图像和其他感知形式信息流，和外界输入信息激活的图像和其他感知形式信息流一样。所以两者都会带来新记忆和记忆值，也会影响原有的记忆中的记忆值。Because of the similar image feature maps that exist in different memories, their similarity in different memories may not be as high as the similarity of language in different memories. When we connect different memories through images and languages, those language symbols (such as voice or text) are frequently used (resulting in high memory value) and have high similarity to each other (resulting in a large activation value transfer coefficient between memories). Then, in the information contained in the same concept (such as various apple images, various apple voices and various apple texts), language symbols are likely to have the highest memory value (because of frequent use and high similarity). When searching for concepts in memory, machines often find language symbols first and use language symbols to represent concepts. When we use language to express our thoughts, we actually use a method (voice or text symbols) to sequentially activate other forms of information (such as images, tastes, feelings, sounds, etc.) that are closely connected to these languages. It is these other forms. The flow of information allows us to understand the information represented by language. It needs to be pointed out in particular that images and other perceptual forms of information flow established through language information flow can become part of our memory, as if they actually happened. This is because the images activated by the language information flow and the information flow of other forms of perception are the same as the images activated by the external input information and the information flow of other forms of perception. So both will bring new memory and memory value, and also affect the memory value in the original memory.
扩展概念的建立。The establishment of the expansion concept.
在语言的运用中，我们必须把那些经常使用的信息组合，使用一个符号来代表，并在群体中形成共识。这样，我们在信息交流时，就可以使用这个符号来简洁地代表这一串信息组合。这就是在基础概念的基础上创造新的概念。In the use of language, we must combine the frequently used information, use a symbol to represent it, and form a consensus among the group. In this way, when we exchange information, we can use this symbol to concisely represent this string of information combinations. This is to create new concepts on the basis of basic concepts.
我们能够把不同的基础概念归属于一个概念下，是因为这些不同的基础概念中包含有某种共有属性。而这种共有属性就是在生活中，逐步归纳而建立的。比如在我们的祖先记忆中，“打猎”可能激活“长矛”，也可能激活“石斧”。那么“长矛”和“石斧”就通过“打猎”在记忆中建立了更加紧密的联系。我们的祖先可能为了更加方便的表达这种联系，就创造了“武器”这个语言符号来指所有和“打猎”活动相关的工具。如果有一次隔壁部落给了我们祖先一把“***”，并告诉它这是一件“武器”。虽然他没有使用这个事物的经验，但通过“武 器”这个属性，我们的祖先会借用武器相关的经验，在打猎时，把这把“***”作为一块石头扔向动物。因为降低分辨率后，***和石头在武器里是同一类事物，所以使用它们的经验就可以泛化。这就是我们提出的“同概念下属性相同就可以替换”的泛化原则。We can attribute different basic concepts to one concept because these different basic concepts contain certain common attributes. And this common attribute is gradually established in life. For example, in the memory of our ancestors, "hunting" may activate "spear" or "stone axe". Then "Lance" and "Stone Axe" established a closer connection in memory through "hunting." In order to express this connection more conveniently, our ancestors created the language symbol "weapon" to refer to all tools related to "hunting" activities. If once the tribe next door gave our ancestor a "pistol" and told it it was a "weapon". Although he has no experience in using this thing, through the attribute of "weapon", our ancestors will borrow experience related to weapons and throw this "pistol" as a stone at animals when hunting. Because after reducing the resolution, the pistol and the stone are the same thing in the weapon, so the experience of using them can be generalized. This is the generalization principle we put forward that "the same attribute can be replaced under the same concept".
比如我们把来就餐的人统称为顾客，把顾客额外给我们的各种数目不等的钱统称为小费，这是降低事物的分辨率，只保留它们的共有属性，所以它们彼此相似，被归纳为一个概念。同理，我们也把苹果分为红富士苹果、美国蛇果和烟台苹果。这是增加事物的分辨率，来区分差异。而之所以能够建立这样的新概念，是因为这些事物存在共同的特征。而这些特征是某些动作、场景或者过程中共有的特征。For example, we collectively refer to the people who come to eat as customers, and collectively refer to the various amounts of money that customers give us as tips. This is to reduce the resolution of things and only retain their common attributes, so they are similar to each other and are summarized. As a concept. In the same way, we also divide apples into Red Fuji apples, American snake fruit and Yantai apples. This is to increase the resolution of things to distinguish differences. The reason why such a new concept can be established is because these things have common characteristics. These features are common features in certain actions, scenes, or processes.
动态概念的扩展，也是通过增加或者降低分辨率来建立新的动态特征分类。比如把“跑步”和“跳舞”统称为“运动”，把“跑步”分为“快跑”、“慢跑”和“长跑”等。这些也是通过动态特征联系的不同属性来建立的新分类，并创建新语言符号来代表这些分类。The expansion of the dynamic concept also establishes a new dynamic feature classification by increasing or reducing the resolution. For example, "running" and "dancing" are collectively referred to as "sports", and "running" is divided into "fast running", "jogging" and "long-distance running". These are also new categories established through different attributes linked by dynamic features, and new language symbols are created to represent these categories.
所以，扩展的概念，就是在我们的记忆中的一类语言符号，它们是通过和其他语言符号建立了更加紧密的连接而存在，它们的内容是通过其他语言符号代表的内容中的共有特征而体现出来的。对于人类而言，在人类历史发展中，已经总结了大量的扩展概念。我们今天在生活中，绝大部分概念都是通过学习获得(直接获得前人的总结结果)，少量概念是通过同类事物在记忆中都和某一个语言符号建立密切联系而建立的。对于机器，也可以采用同样的方法来学习。比如，人类很多概念就是通过解释的方法来学习的。比如，出现“幸福”的场景或者感觉时，我们被告知这就是“幸福”。再比如，我们可能通过词典学习到“太阳系”这个概念的解释。同理，机器学习也可以这样。我们可以把扩展概念和它们的解释，直接赋予机器：Therefore, the concept of expansion is a type of language symbols in our memory. They exist by establishing a closer connection with other language symbols, and their content is based on the common characteristics of the content represented by other language symbols. Reflected. For mankind, a large number of extended concepts have been summarized in the development of human history. In our lives today, most of the concepts are obtained through learning (directly obtained from the summary results of the predecessors), and a small number of concepts are established through the close connection of similar things in memory with a certain language symbol. For machines, the same method can also be used to learn. For example, many human concepts are learned through interpretation. For example, when a scene or feeling of "happiness" appears, we are told that this is "happiness". For another example, we may learn the explanation of the concept of "solar system" through a dictionary. The same can be said for machine learning. We can directly assign extended concepts and their explanations to the machine:
方法1：直接让机器学习。比如让机器自己通过文字、语音学习一个概念所包含的含义。把这些信息放入到记忆中，机器通过重复学习来建立这些概念和它们的解释之间的相似激活、临近激活和强记忆激活的连接就可以了。Method 1: Let the machine learn directly. For example, let the machine learn the meaning of a concept through text and voice. Put this information into memory, and the machine can establish similar activation, proximity activation, and strong memory activation connections between these concepts and their interpretation through repeated learning.
方法2：直接建立一段“伪造的”机器记忆，在这些记忆中人为赋予机器相似激活、临近激活和强记忆激活的连接关系(比如赋予相关信息高记忆值，记忆位置放到一起，按照信息的编码来帮助机器能更快地找到相似的特征等)。Method 2: Directly establish a "fake" machine memory, in which artificial connections between similar activation, proximity activation, and strong memory activation are assigned to the machine (such as assigning high memory value to related information, and memory locations are put together, according to the information Encoding to help the machine find similar features faster, etc.).
通过上述方式，在输入这些扩展概念的语言符号时，机器就能联想到其包含的基础概念上，进而联想到其他形式的信息上(比如图像、声音、气味、触觉、情绪、感觉等)，并使用这些形式的信息来加入到语言符号形成的信息流中，机器就能通过联想来寻找类似的信息流，从而借鉴过去类似的信息流来推测信息流出现的原因和可能的结果。并把这些因果关系放入自己的需求和情绪评估***中，来决定自己的响应，这就是智能。Through the above method, when inputting the language symbols of these extended concepts, the machine can associate the basic concepts it contains, and then associate it with other forms of information (such as images, sounds, smells, touch, emotions, feelings, etc.). And using these forms of information to join the information flow formed by the language symbols, the machine can find similar information flows through association, and use similar information flows in the past to speculate on the causes and possible results of the information flow. And put these causal relationships into your own needs and emotion evaluation system to determine your own response. This is intelligence.
3.2，泛化的实现。3.2, the realization of generalization.
如果找不到和现实输入的信息流一样的记忆中信息流时，这时就需要泛化。泛化的意思是通过类似的记忆中信息流来推测现实输入的信息流的起因和结果。在本发明申请中，泛化的工具主要是表示动作特征的概念。因为动态特征是一种动态运动方式，其主体是一种泛化的主体。机器可以采用质点或者立体图形来代表抽象的运动主体。正是因为运动主体是泛化的主体，所以机器才可以把同类概念作带入运动特征中，从而实现经验的泛化能力。而这种同类概念可以是直观看并不相似，但它们通过降低分辨率，可以归纳到同一个类事物。If there is no information flow in memory that is the same as the actual input information flow, then generalization is needed. Generalization means to infer the cause and result of the actual input information flow through the similar information flow in the memory. In the application of the present invention, the tools of generalization are mainly the concept of expressing action characteristics. Because the dynamic feature is a dynamic way of movement, its subject is a generalized subject. The machine can use mass points or three-dimensional graphics to represent abstract moving subjects. It is precisely because the subject of motion is the subject of generalization, so that the machine can bring similar concepts into the characteristics of motion, thereby realizing the generalization ability of experience. And this kind of similar concepts can be viewed directly and not similar, but they can be summarized into the same kind of things by reducing the resolution.
表示事物之间关系的概念也是一种动态特征。它把关系两端的对象作为一个虚拟整体来考虑。所以，在本发明申请中，通过给表示关系的概念赋予一个动态特征，机器就能通过这个动态特征正确的运用这个表示关系的概念。比如语言“虽然...但是...”、“可是...”、“尽管...”、“然而...”等代表的关系，可以使用一个转折的动态特征来表示。“一边...一边...”、“既...又...”这样的并行概念，可以使用并行进行的动态特征来表示。“包含于...之中”这样的关系概念，可以使用包含的动态特征来表示。The concept of expressing the relationship between things is also a dynamic feature. It considers the objects at both ends of the relationship as a virtual whole. Therefore, in the application of the present invention, by assigning a dynamic feature to the concept representing the relationship, the machine can correctly use the concept representing the relationship through this dynamic feature. For example, the relationships represented by languages such as "although...but...", "but...", "though...", "but..." can be represented by a dynamic feature of transition. Parallel concepts such as "on one side... on the other side..." and "both... and..." can be represented by dynamic characteristics of parallel operations. The relational concept of "contained in" can be expressed by the dynamic feature of inclusion.
这种关系动态特征的具体建立方法是：1，机器通过对大量的语言采用记忆和遗忘机制，寻找它们的共同点，这些共同点通常就是表示动态模式或者关系的概念，因为它们和具 体对象的无关性，导致它们可以被广泛使用。这些词语的组织方式，逐渐变成常用语、常用句型和语法等形式。这种方法类似于目前人工智能中语言的组织方法，是一种机械模仿的方法。2，在本发明申请中，机器需要进一步去理解这些概念的含义。机器理解的方法就是把每次使用这些概念所联系的具体静态特征图和动态特征图记忆下来，然后通过记忆和遗忘机制来保存这些概念。因为在描述关系的过程中，具体对象总是变化的(它们只存在于粗略分辨率上的相似性，因为通过记忆和遗忘机制，只有那些粗略分辨率上的特征是它们的公共特征。所以它们是很广泛的对象，是可以把大量不同对象带入到同一个动作或者关系中)，而不变的是代表关系的动态特征。比如“一边...一边...”这样的关系应用时，常常使用在两个对象并列活动的动态特征中。所以经过积累，机器就能把“一边...一边...”这样的表述关系的词语，表示成两个泛化对象和一个代表“两个对象并列活动”的动态特征。The specific methods for establishing the dynamic characteristics of this relationship are: 1. The machine uses memory and forgetting mechanisms for a large number of languages to find their common points. These common points are usually the concept of dynamic patterns or relationships, because they are related to specific objects. Irrelevance, leading to them can be widely used. The organization of these words has gradually become common words, common sentence patterns, and grammar. This method is similar to the current method of language organization in artificial intelligence, and is a method of mechanical imitation. 2. In the application of the present invention, the machine needs to further understand the meaning of these concepts. The method of machine understanding is to memorize the specific static feature maps and dynamic feature maps associated with each use of these concepts, and then save these concepts through the memory and forgetting mechanism. Because in the process of describing the relationship, the specific objects always change (they only exist in the similarity at the rough resolution, because through the memory and forgetting mechanism, only those features at the rough resolution are their common features. So they It is a very wide range of objects, which can bring a large number of different objects into the same action or relationship), and what does not change is the dynamic characteristics of the relationship. For example, in relational applications such as "one side... one side...", it is often used in the dynamic characteristics of the parallel activities of two objects. Therefore, after accumulation, the machine can express the expressions of relations such as "on one side... on the other side..." as two generalized objects and a dynamic feature representing "two objects side by side activity".
动态扩展的另外一个方面是：我们的生活中，有很多过程是由多个实体概念或者扩展后的抽象概念，构成的一种广义的运动模式，我们称之为过程特征。过程特征是一种扩展了的动态特征，它的特征是：1，多个观察对象，它们不一定是一个整体。2，整个运动方式没有明确重复的轨迹线。比如回家、出差、洗手、做饭等过程，它们是多个实体概念或者扩展后的抽象概念，构成的一种广义的运动模式。之所以称之为模式，是因为这些概念在我们生活中是能够不断重复的。既然能重复，就说明这些概念代表的过程中，存在共有特征，否者，我们就不可能用一个概念来代表它们。Another aspect of dynamic expansion is: in our lives, many processes are composed of multiple entity concepts or expanded abstract concepts, which constitute a generalized movement mode, which we call process characteristics. Process feature is an extended dynamic feature. Its features are: 1. Multiple observation objects, they are not necessarily a whole. 2. There is no clear repeating trajectory in the whole movement mode. For example, the processes of going home, going on business, washing hands, cooking, etc., are multiple physical concepts or expanded abstract concepts that constitute a generalized movement mode. It is called a pattern because these concepts can be repeated in our lives. Since it can be repeated, it means that there are common features in the process of representation of these concepts. Otherwise, it is impossible for us to use a concept to represent them.
过程特征通常是涉及空间大、时间长的动态过程。实现它的具体细节和环境密切相关，所以很难从中找到相似性。但这些环节，通常都有语言符号来代表。所以，我们寻找一个过程特征时，可以先寻找每个环节的语言符号的重复性。比如机器通过记忆每次去机场，每个环节所对应的语言符号，构成一个逐步展开的塔形概念关系。举例说明：这个概念的顶层是“去机场”，下一层是“准备去”、“途中”、“到达”，再下一层是“准备行李”、“找车”、“告别朋友”、“坐车”、“途中”、“到达机场车库”、“出车库”、“到达机场入口”。再下一层是“准 备衣物”、“准备洗漱用品”、“准备钱”、“准备工作相关材料”....。这个过程可以不断细分下去。一开始，每个环节的区分可以是带有随意性的。但在每一次去机场后，我们都得到一个塔形的概念组织。这个塔形的概念组织经过记忆和遗忘机制，最终在每个分辨率层次上，只有少量的，必不可少的，频繁出现的概念才能在记忆中保留下来。它们就是在对应分辨率上的过程特征。这些过程特征是一连串概念，带有时间和空间次序组织起来的。尤其是在底层，通常只能留下每一次去机场都可能有的静态特征图和动态特征图。这些特征图数量很少，但它们缺一不可。这些就是代表关键环节的静态特征图或者动态特征图，比如“安检”或者“登机”。和关键环节相连的上层概念，也是缺一不可的(它们可能数量上更少)。依次向上推，最后就只有“去机场”这样一个顶层概念。所以，建立过程特征是从正向选择(自己借鉴他人经验而刻意记住的环节)和逆向选择(每次都有的事情对应的上层环节)，通过记忆和遗忘机制来实现的。Process characteristics are usually dynamic processes involving large space and long time. The specific details of its implementation are closely related to the environment, so it is difficult to find similarities. But these links are usually represented by language symbols. Therefore, when we look for a process feature, we can first look for the repetitiveness of the language symbols in each link. For example, each time the machine goes to the airport through memory, the language symbols corresponding to each link form a gradually unfolding tower-shaped conceptual relationship. For example: the top level of this concept is "going to the airport", the next level is "ready to go", "on the way", "arrival", and the next level is "preparing luggage", "finding a car", "farewell to friends", "By car", "On the way", "Arriving at the airport garage", "Out of the garage", "Arriving at the airport entrance". The next level is "Prepare clothes", "Prepare toiletries", "Prepare money", "Prepare related materials".... This process can be subdivided continuously. At the beginning, the distinction of each link can be arbitrary. But every time we go to the airport, we get a tower-shaped conceptual organization. This tower-shaped conceptual organization goes through a memory and forgetting mechanism, and finally at each resolution level, only a small amount, indispensable, and frequently appearing concepts can be retained in memory. They are process characteristics at the corresponding resolution. These process characteristics are a series of concepts, organized in a temporal and spatial order. Especially on the ground floor, usually only static feature maps and dynamic feature maps that may be available every time you go to the airport can be left. These feature maps are few in number, but they are indispensable. These are static feature maps or dynamic feature maps that represent key links, such as "security check" or "boarding". The upper-level concepts connected to the key links are also indispensable (they may be fewer in number). Push upwards one by one, and in the end there is only a top-level concept of "going to the airport". Therefore, the establishment of process characteristics is realized through the mechanism of memory and forgetting from positive selection (the link deliberately memorized by learning from other people's experience) and adverse selection (the upper link corresponding to something every time).
这些保留下来的塔形概念和底层特征图，就是我们每次去机场的模仿对象。我们只需要把现实环境中的具体事物，按照类比的方法放入这个过程特征中，我们就能建立起从任何地方去机场的各个阶段目标规划能力。而在具体实施时，需要使用分段模仿来把这些抽象的概念逐层展开，加入符合现实情况的更多环节。这样我们就建立了机器在各种不同环境下去机场的能力。而这种能力就是一种经验上的泛化。整个过程就是模仿和泛化的反复迭代进行的过程。These preserved tower-shaped concepts and underlying feature maps are the objects of imitation every time we go to the airport. We only need to put specific things in the real environment into the characteristics of this process according to the analogy method, and we can build up the ability to plan goals at all stages of going to the airport from anywhere. In the specific implementation, it is necessary to use segmented imitation to unfold these abstract concepts layer by layer, adding more links in line with the reality. In this way, we have established the ability of the machine to go to the airport in a variety of different environments. And this ability is an empirical generalization. The whole process is an iterative process of imitation and generalization.
在建立了以上基础后，机器就可以采用计算机能理解的方式来实现泛化了。下面以机器收到“去机场”这个指令为例来说明泛化过程。机器收到“去机场”这个指令后，通过联想，激活了“去”这个语言符号的相关信息(可能是动作图像和感觉)，也激活了“机场”这个语言符号的相关信息(可能是一些机场图像)。如果机器曾经去过机场，那么这两个激活点会通过“临近激活”、“相似激活”和“强记忆激活”直接激活一大串其他记忆，这些记忆可能包含静态图像、动态图像、语音和文字还有感觉和情绪。它们在记忆中都是有时间和空间 关系的。机器只需要把这一串信息按照时间和空间关系作为中间环节目标，逐步模仿并实现，就能达到“去机场”的目的。如果机器没有去机场的经验，那么可能激活的是“去火车站”、“去商店”、“去旅游”的相关图像，也有“机场”的相关图像。机器通过对比，发现这些图像中，除了“机场”外，“火车站”的激活值是最高的。这是因为存在多个激活通道给“火车站”传递激活值。比如“去...”的经验激活了去“火车站”，比如“机场”和“火车站”在粗略的分辨率上存在相似性。比如通过学习获得他人的经验“机场和火车站都是乘坐交通工具的地方”，这个途径也会从“机场”给“火车站”传递激活值。所以，在建立了扩展的静态和动态概念的基础上，机器对输入的信息，都可以通过概念之间的连接关系联系到一系列的底层静态或者动态非语言信息(包括声音)，这些信息就可以采用不同的分辨率来比较相似度，来实现相似度激活、临近激活和强记忆激活。然后这被激活的非语言信息，会再次通过语言概念作为桥梁，激活其他并不相似的，或者并临近的，或者记忆值不是很高的记忆。当所有联想激活完成后，机器选取其中的1～N(自然数)个激活值最高的信息，把它们按照它们自带的时间次序或者空间次序组织起来，它们就是机器需要模仿的泛化经验。所以泛化能力并不需要机器刻意去建立。机器只需要建立正确的概念，和概念之间的关系，通过联想激活，可以利用的泛化经验就能自动涌现。而本发明申请中，能够做到这一点的两个关键就是：1，动作特征和关系概念需要单独提取，把它们和具体对象脱钩。2，建立合理的关系网络，实现正确的联想能力，而且这个联想能力必须是能够量化的。After establishing the above foundation, the machine can realize generalization in a way that the computer can understand. The following is an example of the machine receiving the instruction "go to the airport" to illustrate the generalization process. After the machine receives the instruction of "go to the airport", through association, it activates the relevant information of the language symbol of "go" (may be action images and feelings), and also activates the relevant information of the language symbol of "airport" (may be some Airport image). If the machine has ever been to an airport, then these two activation points will directly activate a large number of other memories through "proximity activation", "similar activation" and "strong memory activation". These memories may include static images, moving images, speech and text There are also feelings and emotions. They are all related to time and space in memory. The machine only needs to take this string of information as the intermediate link goal in accordance with the time and space relationship, and gradually imitate and realize it, to achieve the goal of "going to the airport". If the machine has no experience in going to the airport, it may activate related images of "going to the train station", "going to the store", "going to travel", and also related images of the "airport". Through the comparison, the machine found that in these images, except for the “airport”, the activation value of the “train station” is the highest. This is because there are multiple activation channels to transfer activation values to the "train station". For example, the experience of "going to..." activates going to the "train station". For example, "airport" and "train station" are similar in rough resolution. For example, through learning to gain the experience of others "airports and railway stations are places to take transportation", this approach will also transfer activation values from "airports" to "train stations". Therefore, based on the establishment of extended static and dynamic concepts, the input information of the machine can be connected to a series of underlying static or dynamic non-verbal information (including sound) through the connection relationship between the concepts. Different resolutions can be used to compare similarity to achieve similarity activation, proximity activation and strong memory activation. Then the activated non-verbal information will once again use the language concept as a bridge to activate other memories that are not similar, or that are nearby, or whose memory value is not very high. When all the associations are activated, the machine selects the information with the highest activation value from 1 to N (natural numbers) and organizes them according to their own time sequence or spatial sequence. They are the generalized experience that the machine needs to imitate. Therefore, the generalization ability does not need to be established deliberately by the machine. The machine only needs to establish the correct concept, and the relationship between the concept, through the association activation, the generalization experience that can be used will automatically emerge. In the application of the present invention, the two keys to achieve this are: 1. Action features and relationship concepts need to be extracted separately to decouple them from specific objects. 2. Establish a reasonable relationship network to realize the correct association ability, and this association ability must be quantifiable.
4，实现需求和评估能力。4. Realize requirements and assessment capabilities.
在信息输入时，机器首先在记忆中找到一段或者多段最相关记忆，它们就是一连串和输入信息相关的信息流，这是通过联想激活实现的。这些记忆是过去机器对类似输入信息的响应，或者是过去对局部类似于输入信息的多个信息的响应。这些响应的发出者既可以是机器本身，也可以是其他事物。机器把自己和信息源之间发生次数最多的、和输入信息相关的响应作为信息源的目的。如果机器和信息源之间没有频繁的互动，那么机器就把他人使用最 多的响应，认为是信息源发出信息的目的。这是合理的，因为信息源发出信息的目的是为了得到响应。信息源根据自己的经验，已经预设了可能的响应。而这些预设响应正是基于信息源与机器的互动或者信息源于他人的互动经验来建立的。这些都可以采用预置经验来实现。当机器理解了信息源的目的，也就理解了输入信息。When inputting information, the machine first finds one or more segments of the most relevant memory in the memory. They are a series of information streams related to the input information, which is achieved through association activation. These memories are past machine responses to similar input information, or past responses to multiple pieces of information that are partially similar to input information. The sender of these responses can be either the machine itself or other things. The machine takes the most frequently-occurring response between itself and the information source and related to the input information as the purpose of the information source. If there is no frequent interaction between the machine and the information source, then the machine considers the response most used by others as the purpose of the information source. This is reasonable, because the purpose of the information source is to get a response. The information source has preset possible responses based on its own experience. These pre-determined responses are established based on the interaction between the information source and the machine or the interactive experience of the information derived from others. These can all be achieved by using preset experience. When the machine understands the purpose of the information source, it also understands the input information.
机器在理解了信息源的目的后，机器需要建立对应的响应。机器建立响应的方法是：4.1，机器找到一段或者多段可以作为参考来建立响应的记忆。具体方法是：机器把输入的语言信息通过联想激活方法，把它们转换为非语言信息流，并把这些信息流和其他输入的非语言信息作为总的输入信息。After the machine understands the purpose of the information source, the machine needs to establish a corresponding response. The method for the machine to establish a response is: 4.1, the machine finds one or more segments that can be used as a reference to establish the memory of the response. The specific method is: the machine converts the input linguistic information into non-verbal information streams through the association activation method, and uses these information streams and other input non-verbal information as the total input information.
4.2，机器把转换后的信息流，作为新的虚拟输入信息，再次对这些信息使用联想激活。联想激活完成后，那些获得激活值高的信息组成的信息流，时间在类似信息输入之前的信息就是关于起因的信息，在类似信息输入之后的信息就是关于结果的信息。4.2. The machine uses the converted information stream as the new virtual input information, and uses the association activation for the information again. After the Lenovo activation is completed, the information stream composed of information with high activation value. The information before the input of similar information is the information about the cause, and the information after the input of the similar information is the information about the result.
4.3，机器可能找到多段关于自己的记忆，也可能找的是多段关于他人的记忆。机器真实的需求是找到和这些记忆存储在一起的需求状态变化过程和情绪状态变化过程。如果关于他人的记忆，机器还需要把他人的活动替换成自己的活动，再次作为一个新的虚拟输入，然后再次通过联想激活，寻找和这个虚拟输入起因相关的记忆和结果相关的记忆，寻找这些记忆中的需求状态变化过程和情绪状态变化过程。4.3. The machine may find multiple memories about itself, or it may find multiple memories about others. The real need of the machine is to find the change process of the demand state and the change process of the emotional state that are stored with these memories. If it is about the memory of others, the machine needs to replace the activities of others with its own activities, again as a new virtual input, and then activate it through association again, looking for the memory related to the cause of this virtual input and the memory related to the result, looking for these The changing process of the demand state and the changing process of the emotional state in memory.
4.4，机器把步骤4.3中总激活值最高的一段或者多段记忆组织起立成一段或者多段响应，作为模仿的对象。因为这些被激活的信息中都带有自己的时间和关系关系(它们都同步存储在记忆中)，所以它们可以组织成一段或者多段响应。4.4. The machine sets up one or more segments of memory organization with the highest total activation value in step 4.3 into one or more segments of response as the object of imitation. Because these activated messages have their own time and relationship (they are stored in memory simultaneously), they can be organized into one or more responses.
4.5，机器分析准备模仿的一段或者多段响应相关记忆中，自己的需求状态变化过程和情绪状态变化过程，是正面的还是负面的。按照“趋利避害”的原则评估来选取这些一段或者多段响应。由于这些记忆中包含有“需求值”的变化和“情绪值”的变化，所以只需要简单的统计算法就可以实现“趋利避害”的选择。这些算法可以是预置的。4.5. The machine analyzes whether it is positive or negative for the change process of one's own demand state and emotional state in one or more response-related memories to be imitated. Select these one or more responses in accordance with the principle of "seeking advantages and avoiding disadvantages". Since these memories contain changes in "demand value" and "emotional value", only simple statistical algorithms are needed to realize the choice of "seeking advantages and avoiding disadvantages". These algorithms can be preset.
4.6，机器把通过了“趋利避害”评估的一段或者多段响应组合成一个大的过程。这些响应中可能存在时间和空间上的组织信息，那么这种组织就按照时间和空间上信息进行。如果这些响应中没有明确的时间和空间次序信息，那么机器需要把这些响应再次作为一个新的虚拟信息输入，来通过联想激活，找到更多的记忆，来寻找它们的时间和空间次序。这个过程迭代进行，直到能确定这些响应的次序为止(也有可能通过记忆发现这些响应的次序可以是任意的)。4.6. The machine combines one or more segments of responses that have passed the assessment of "seeking advantages and avoiding disadvantages" into a large process. There may be organizational information in time and space in these responses, so this organization is carried out in accordance with the information in time and space. If there is no clear time and space order information in these responses, then the machine needs to input these responses as a new virtual information again to find more memories through association activation to find their time and space order. This process is iterative until the order of these responses can be determined (it is also possible to find that the order of these responses can be arbitrary through memory).
4.7，机器在动机驱动下选择响应。4.7. The machine chooses to respond under motivation.
驱动机器的动力来源是机器的动机，而机器的动机可以概括为“趋利避害”。“利”和“害”一部分是预置的；一部分是后天学习中建立的，因为它们和机器自身的需求相关。类比于人类，比如一开始是“水”，“奶”、“食物”是先天预置的“利”，后来通过学习获得了“考试分数”、“钞票”和我们先天需求之间的联系，再后来我们还发现操作对象还可以是“爱情”和“时间”这样的没有实体的东西，甚至我们还追求群体中的支配权，这是一种存在于我们基因中底层动机“目标达成”的延展。采用类似的方法，我们也可以给机器赋予我们希望它们拥有的动机。因为在我们的关系网络中，所有的记忆帧存储时，同时存储了当时机器的需求符号和对应的记忆值。这些记忆值是和当时需求符号的状态值正相关的。举例说明，如果机器在某种行为后，收到了责备。由于责备是一种损失(这个经验既可以预置，也可以通过训练者语言表达，还可以直接修改关系网络来实现)，而且责备的程度(比如语言里面表示程度的词)给机器带来不同的损失值。责备越强烈，机器给这个记忆中的损失符号赋予的记忆值也相应比较高。那么在这个记忆中，由于损失符号记忆值比较高，所以这个记忆帧中所有其他记忆值比较高的特征图都和损失符号之间的连接比较强。如果在类似环境，类似动作发出对象或者接受对象，再次发生了类似受到责备的行为，那么这个记忆帧中的带来损失的特征图和损失符号本身由于被重复了，它们的记忆值在这个记忆帧中都按照记忆曲线增加了，从而增加了带来损失的特征图和损失符号之间的关系。通过一次次重复，那些真正带来损失 的特征图和损失符号之前的关系就按照记忆和遗忘机制挑选出来了。机器从一开始不清楚为什么被责骂，到后面就能清楚是什么东西给自己带来了被责骂的后果。这个过程和人类孩子的学习过程是类似的。The source of power that drives the machine is the motivation of the machine, and the motivation of the machine can be summarized as "seeking advantages and avoiding disadvantages." "Profits" and "harms" are partly preset; partly they are established through acquired learning, because they are related to the needs of the machine itself. Analogous to human beings, for example, at the beginning, "water", "milk" and "food" are pre-built "profits". Later, through learning, we have obtained the connection between "exam scores", "banknotes" and our innate needs. Later, we discovered that the object of operation can also be non-substantial things such as "love" and "time", and we even pursue domination in the group, which is a kind of "goal achievement" that exists in the bottom motivation of our genes. Stretch. In a similar way, we can also give machines the motivation we want them to have. Because in our relational network, when all memory frames are stored, the demand symbols of the machine at the time and the corresponding memory values are stored at the same time. These memory values are positively related to the state value of the demand symbol at the time. For example, if the machine receives blame after a certain behavior. Because blame is a loss (this experience can be preset, expressed through the language of the trainer, or directly modified by the relationship network), and the degree of blame (such as the words in the language that express the degree) brings different effects to the machine The loss value. The stronger the blame, the higher the memory value assigned by the machine to the loss symbol in memory. Then in this memory, since the memory value of the loss symbol is relatively high, all other feature maps with higher memory value in this memory frame have a stronger connection with the loss symbol. If in a similar environment, a similar action sends out an object or accepts an object, and a behavior similar to being blamed occurs again, then the loss-causing feature map and loss symbol themselves in this memory frame have been repeated, and their memory value is in this memory. The frames are all increased according to the memory curve, thereby increasing the relationship between the loss-causing feature map and the loss symbol. Through repeated repetitions, the relationship between the feature map and the loss symbol that actually caused the loss was selected according to the memory and forgetting mechanism. From the beginning, the machine didn't know why it was scolded, but later it would be clear what caused the scolding consequences. This process is similar to the learning process of human children.
同理，机器的收益值、安全值、危险值、目标达成值、支配值等就是类似的情况。它们都是通过机器在过去的经验中，不断的把行为和行为结果联系在一起。联系在一起的方法就是把它们放入同一个记忆帧中。即使机器在行为发生时没有得到及时反馈。训练者在后期也可能通过指出行为本身并发出反馈，这样就是在一个单独的记忆帧中把行为和结果连接起来了。训练者甚至无需去指明具体哪个行为好和不好，机器只需要每次收到正确的反馈，通过记忆和遗忘，就能逐步建立正确的行为和需求值之间的连接关系。比如那些一定会收到奖励或者惩罚的行为，每次行为和奖励或者惩罚发生后，它们被同时记忆下来。每重复一次，它们的记忆就增加，最终两者之间的连接，比其他连接会越来越紧密。In the same way, the profit value, safety value, risk value, goal achievement value, and dominance value of the machine are similar situations. They all continuously link behavior and behavior results through the machine's past experience. The way to connect them is to put them in the same memory frame. Even if the machine did not get timely feedback when the behavior occurred. The trainer may also point out the behavior itself and give feedback in the later stage, so that the behavior and the result are connected in a single memory frame. The trainer does not even need to specify which behavior is good or bad. The machine only needs to receive the correct feedback every time, and through memory and forgetting, it can gradually establish the connection between the correct behavior and the demand value. For example, those behaviors that will definitely receive rewards or punishments are memorized at the same time after each behavior and reward or punishment. Each time they repeat, their memory increases, and eventually the connection between the two will become closer and closer than the other connections.
机器的评估***，是一个预置的程序。这个程序是基于机器需求的收益和损失值、安全和危险值、目标达成值、支配值等满足状态来决定一个虚拟输出是否要转变成一个真正的输出。这些需求类型，是人类赋予给机器的。当然，我们可以赋予机器更多人类期望他们拥有的目标，比如“遵守机器人公约”、“遵守人类法律”、“富有同情心”、“讲道德”、“行为优雅”等目标。这些目标都可以通过在记忆中设定需求符号，并通过训练者反馈来调整机器的行为，从而实现人类的期望。需要指出，这些目标都可以按照人类的期望来增减。而对这些目标的增减不影响本发明申请的权利要求。The evaluation system of the machine is a preset program. This program determines whether a virtual output should be transformed into a real output based on the satisfaction state of the machine's demand for gains and losses, safety and risk values, goal achievement values, and dominance values. These types of needs are given by humans to machines. Of course, we can give machines more goals that humans expect them to have, such as "compliance with the robot convention", "compliance with human laws", "compassionate", "ethical", "behaving gracefully" and other goals. These goals can be achieved by setting demand symbols in the memory and adjusting the behavior of the machine through feedback from the trainer, so as to achieve human expectations. It needs to be pointed out that these goals can be increased or decreased in accordance with human expectations. The addition or reduction of these objectives does not affect the claims of the present application.
为了更好的和人类交流。本发明申请提出把机器需求的实际满足状态作为情绪***的输入，采用预置程序，把它们转换成机器的情绪。这样做的目的是拟人化，模仿人类自身在不同的需求满足状态下的情绪反应。只有这样，机器才能更好的和人类交流。同时，我们采用如下方法来实现机器自身可以利用自己的情绪来达到自己的目的：1，机器每次存储记忆时，同步存储自己的情绪。2，训练者需要对机器的情绪做出反馈。通过训练者的反馈，机器来确 定情绪应该怎么调整。3，机器可以自己修改预置程序的参数，根据自己的经验来输出情绪。有了以上3点，机器就能把情绪和反馈联系起来。这样情绪既是一种表达方式，又是一种可以利用的手段。因为特定的情绪和特定的外界反馈是相联系的。机器在寻找特定的反馈过程中，情绪就可能被纳入记忆，成为机器期望再现特定结果时的一种模仿对象。需要指出，情绪的种类和强度，都可以按照人类的期望来增减。而对这些目标的增减不影响本发明申请的权利要求。In order to better communicate with humans. The application of the present invention proposes to use the actual satisfaction state of the machine's requirements as the input of the emotion system, and use a preset program to convert them into the emotion of the machine. The purpose of this is to anthropomorphize, imitating the emotional response of human beings in different states of satisfying needs. Only in this way can machines better communicate with humans. At the same time, we use the following methods to realize that the machine itself can use its own emotions to achieve its own goals: 1. Each time the machine stores a memory, it stores its own emotions synchronously. 2. The trainer needs to give feedback on the emotions of the machine. Through the trainer's feedback, the machine determines how emotions should be adjusted. 3. The machine can modify the parameters of the preset program by itself, and output emotions according to its own experience. With the above three points, the machine can connect emotions and feedback. Such emotions are not only a way of expression, but also a means that can be used. Because certain emotions are connected with certain external feedback. When the machine is looking for specific feedback, emotions may be incorporated into memory and become a kind of imitation object when the machine expects to reproduce specific results. It needs to be pointed out that the type and intensity of emotions can be increased or decreased according to human expectations. The addition or reduction of these objectives does not affect the claims of the present application.
机器建立的各种评估值，还需要结合机器自身内部状态值(比如是不是缺电了，是不是自己有些***坏了等)来做出判断，判断的结果就是通过或者不通过。机器的评估***是一个预置程序。它是一个对机器赋予个性化的环节，不同的选择就是相当于不同的性格。机器也可以保留一些可以让自己来调整的参数，通过尝试不同的选择，来带不同的后果，从而逐步建立最符合自己需求的评估***。这一步通过现有的公知技术就可以实现，这里不再赘述。The various evaluation values established by the machine also need to be combined with the internal state values of the machine itself (for example, is it lack of power, whether some of its own systems are broken, etc.) to make a judgment, and the result of the judgment is pass or fail. The evaluation system of the machine is a preset program. It is a link that personalizes the machine, and different choices are equivalent to different personalities. The machine can also retain some parameters that can be adjusted by itself, and try different options to bring different consequences, so as to gradually establish an evaluation system that best meets its needs. This step can be achieved by the existing publicly known technology, and will not be repeated here.
如果机器建立的响应，无法通过评估***。那么机器需要重新建立响应，需要去掉上次评估中，带来重大损失、危险等各种负面结果的行为。这些行为就是由那些带来损失的静态特征图和动态特征图结合后的行为。去掉负面行为也是一个比较复杂的机器思维过程。在这个过程中，机器需要把目前的目标全部转为继承目标，把计算能力空置出来，用于去掉负面行为这样一个临时目标的计算。然后，机器需要寻找关于这个负面行为的所有记忆，从中找到如何排除它的经验。在去掉带来负面结果的行为后，机器重新建立新的响应。而建立的过程，依然是优选动态特征图，概念替换静态特征图，然后借助相似记忆来确定它们的结合方式。如果机器反复多次还是无法建立能通过评估的响应。则有可能是在前面的步骤中有差错，或者机器碰到了无法解决的难题。这时机器进入对“无法处理的信息”流程的处理。也就是说，“无法处理信息”本身就是一种对信息的处理结果。机器根据自己的经验，建立对“无法处理信息”的响应。这些响应可能是置之不理，可能是再次和信息源确认信息，或者再次 采用更高分辨率来识别信息等。这些也都是类似于人类行为的合理响应。If the machine establishes a response, it cannot pass the evaluation system. Then the machine needs to re-establish the response, and needs to remove the behaviors that brought heavy losses, dangers and other negative results in the last evaluation. These behaviors are the combined behaviors of the static feature maps and dynamic feature maps that bring losses. Getting rid of negative behaviors is also a more complicated machine thinking process. In this process, the machine needs to convert all the current goals into inheritance goals, leaving the computing power vacant for the calculation of a temporary goal such as removing negative behaviors. Then, the machine needs to look for all the memories of this negative behavior and find the experience of how to exclude it. After removing the behavior that brought the negative result, the machine re-established a new response. The process of establishment is still to optimize dynamic feature maps, replace static feature maps with concepts, and then use similar memories to determine their combination. If the machine is repeated many times, it still cannot establish a response that can pass the evaluation. It is possible that there was an error in the previous steps, or the machine encountered an unsolvable problem. At this time, the machine enters the processing of the "unprocessable information" flow. In other words, "unable to process information" itself is a result of processing information. The machine builds a response to "unable to process information" based on its own experience. These responses may be ignored, may be reconfirming the information with the information source, or again using higher resolution to identify the information, etc. These are also reasonable responses similar to human behavior.
在上述过程中，机器需要反复使用联想激活过程。这里需要特别指出，由于存在激活阈值，所以即使特征图之间激活值传递系数是线性的，特征图的激活值累计函数也是线性的，但由于激活阈值的存在，无论是在单次联想激活过程中，还是在多次联想激活过程中，相同特征图和相同初始激活值，但因为激活次序选择不一样，最终的激活值分布是不一样的。这是因为激活阈值的存在带来的非线性。不同的传递路径，带来的信息损失是不一样的。激活次序选择的偏好，这相当于机器个性的差异，所以在相同输入信息下，产生不同的思考结果，这个现象和人类是一致的。In the above process, the machine needs to use the Lenovo activation process repeatedly. It needs to be pointed out here that due to the activation threshold, even if the activation value transfer coefficient between the feature maps is linear, the activation value accumulation function of the feature map is also linear, but due to the existence of the activation threshold, no matter in the single association activation process In the process of multiple association activations, the same feature map and the same initial activation value, but because the activation order is selected differently, the final activation value distribution is different. This is because of the non-linearity caused by the existence of the activation threshold. Different transmission paths bring different information losses. The preference of activation order selection is equivalent to the difference in machine personality. Therefore, under the same input information, different thinking results are produced. This phenomenon is consistent with human beings.
另外，关系网络中的关系强度和最新的记忆值(或者连接值)是相关的。所以机器会有先入为主的现象。比如拥有同样的关系网络的两个机器，面对同样一个特征图和同样的初始激活值，其中一个机器突然处理了一条关于这个特征图的输入信息，那么这个机器在处理了额外的这条信息后，它会更新关系网络中的相关部分。其中某一个关系线可能会按照记忆曲线增加。这个增加的记忆值在短时间内不会消退。所以在面临同样的特征图和同样的初始激活值时，处理了额外信息的机器，将会把更多的激活值沿刚刚增强了的关系线传播，从而出现先入为主的现象。In addition, the strength of the relationship in the relationship network is related to the latest memory value (or connection value). Therefore, the machine will be preconceived. For example, if two machines with the same relationship network face the same feature map and the same initial activation value, one of the machines suddenly processed an input information about this feature map, then this machine is processing this additional piece of information Later, it will update the relevant part of the relationship network. One of the relationship lines may increase according to the memory curve. This increased memory value will not fade in a short time. Therefore, when facing the same feature map and the same initial activation value, the machine that processes the additional information will spread more activation values along the newly enhanced relationship line, which will lead to a preconceived phenomenon.
另外，为了合理地处理信息输入的先后次序，确保后面输入的信息带来的激活值，不会被前面的信息所屏蔽，在本发明申请中，联想激活中的激活值，会随时间而递减。因为如果关系网络中的激活值不随时间消退，后面信息带来的激活值变化就不够明显，这会带来信息间干扰。如果激活值不消退，后面的信息输入后，会受到前面信息的强烈干扰，导致无法正确的寻找自己的关注点。但如果我们完全清空前面信息的记忆值，那么我们又丢失了前后两段信息可能存在的连接关系。所以，在本发明中，我们提出采用渐进消退的方法来实现前后段信息的隔离和连接之间的平衡。这个消退参数需要在实践中优选。但这带来了维护一个信息的激活状态的问题。如果机器在思考过程中，迟迟无法找出满足机器评估***的响应方 案，随时间流逝，这些激活值就会消退，导致机器遗忘了这些相关信息，忘了自己要干什么。这时机器需要把记忆中的激活值再次刷新。一种刷新方法是：把那些激活值最高的信息转变成虚拟输出，再把这个虚拟输出作为信息输入，走一遍流程，来强调这些关注点，这就是我们在思考时，为什么有时候，不理解时或者找不到思路时，喜欢喃喃自语，或者自己在心中默念。这种虚拟的输入，和真实的输入流程一样，同样可以搜寻记忆和更新记忆值。所以，这种方法可以用于机器有意去增加某些信息的记忆。这就是使用朗读或者默念的方法来增加记忆。另外，在这种情况下，如果出现新的输入信息，机器不得不打断思考过程，去处理新的信息。所以，从节省能量的角度看，机器是倾向于完成思维，避免浪费的。这时机器可能会主动发出“嗯…啊…”等缓冲辅助词来发出输出信息，表示自己正在思维，请勿打扰。还有一种可能是给予机器的思考时间有限，或者信息过多，机器需要尽快完成信息响应，这时机器也可以采用输出再转输入的方式。通过一次这样的方式，机器就强调了有用信息，抑制干扰信息。这些方式在人类普遍使用，在本发明申请中，我们也把它也引入机器的思维。机器可以根据内置的程序，或者自己的经验，或者两者混合，来确定是不是目前的思考时间超过了正常时间，需要刷新关注信息，或者告诉别人自己正在思考，或者强调重点，排除干扰信息。In addition, in order to reasonably process the sequence of information input, to ensure that the activation value brought by the information input later will not be shielded by the previous information. In the application of the present invention, the activation value in the association activation will decrease with time. . Because if the activation value in the relational network does not fade with time, the activation value changes brought about by the subsequent information will not be obvious enough, which will cause interference between information. If the activation value does not fade, after the subsequent information is entered, it will be strongly interfered by the previous information, resulting in the inability to find one's focus correctly. But if we completely clear the memory value of the previous information, then we will lose the possible connection relationship between the two pieces of information before and after. Therefore, in the present invention, we propose to adopt a method of gradual fading to achieve a balance between the isolation and connection of the preceding and subsequent segments of information. This regression parameter needs to be optimized in practice. But this brings about the problem of maintaining the active state of a message. If the machine is unable to find a response plan that satisfies the machine evaluation system during its thinking process, these activation values will fade over time, causing the machine to forget the relevant information and forget what it wants to do. At this time, the machine needs to refresh the activation value in the memory again. One way to refresh is to convert the information with the highest activation value into virtual output, and then use this virtual output as information input, and go through the process to emphasize these concerns. This is why we sometimes don’t understand when we think. Sometimes or when you can't find ideas, I like to mutter to myself, or mutter in my heart. This kind of virtual input, like the real input process, can also search for memories and update memory values. Therefore, this method can be used for machines to deliberately increase the memory of certain information. This is the method of using reading aloud or silently to increase memory. In addition, in this case, if new input information appears, the machine has to interrupt the thinking process to process the new information. Therefore, from the perspective of energy saving, machines tend to complete thinking and avoid waste. At this time, the machine may actively send out buffer auxiliary words such as "Um...ah..." to send out output information, indicating that you are thinking, please do not disturb. Another possibility is that the thinking time given to the machine is limited, or there is too much information, and the machine needs to complete the information response as soon as possible. At this time, the machine can also adopt the method of output and then input. In this way, the machine emphasizes useful information and suppresses interference information. These methods are commonly used by humans, and in the application of the present invention, we also introduce them into the thinking of machines. The machine can determine whether the current thinking time exceeds the normal time based on the built-in program, or its own experience, or a mixture of the two, and it needs to refresh the attention information, or tell others that they are thinking, or emphasize the key points, and eliminate interference information.
由于人类交流最频繁的是语音和文字，所以一个概念的局部网络中，当其他特征图从关系网络的各个支路获得激活值，并都向语音或者文字传送，所以通常的关注点就是概念的语音和文字。所以，机器的自我信息过滤或者强调的方法，虚拟输出通常是语音，因为这是最常见的输出方式。机器输出它们耗能最少。当然，这和一个人的成长过程密切相关。比如，从书本中学习生活的人，有可能是把信息转变成文字，再重新输入。Since the most frequent human communication is voice and text, in a local network of a concept, when other feature maps obtain activation values from each branch of the relationship network and transmit them to voice or text, the usual focus is on the concept Voice and text. Therefore, the virtual output of the machine's self-information filtering or emphasizing method is usually speech, because this is the most common output method. The machine outputs them the least energy. Of course, this is closely related to a person's growth process. For example, people who learn about life from books may convert information into words and then re-enter it.
使用联想激活的搜索方法，利用了语言、文字、图像、环境、记忆和其他传感器的输入信息之中的隐含的连接关系，来相互传递激活值，从而让相关的特征图、概念和记忆彼此支持而凸显出来。它和传统的“上下文”来识别信息的差异在于，传统的识别方法需要预先 人工去建立“上下文”关系库。而本发明申请中，我们提出了“相似性、同环境中信息彼此存在隐含的连接”这个基础假设。在这个基础假设上，简化了形形色色的关系，从而让机器自己去建立关系网络。它不仅仅包含语义，更包含常识。这里需要指出，联想激活是一种搜索方法，它本身不是本发明申请中的必要步骤，可以被其他能达到类似目的的搜索方法所代替。在使用联想激活时，机器可以把每个记忆中，激活值超过预设值的特征图，认为是使用了一次，按照记忆所属记忆库中的记忆和遗忘机制来维护它们的记忆值。The search method that uses association activation uses the implicit connection relationship among the input information of language, text, image, environment, memory and other sensors to transfer activation values to each other, so that related feature maps, concepts and memories are mutually connected Support and highlight. The difference between it and the traditional "context" to identify information is that the traditional recognition method requires manual establishment of the "context" relation database in advance. In the application of the present invention, we put forward the basic assumption of "similarity and implicit connection between information in the same environment". Based on this basic assumption, all kinds of relationships are simplified, allowing the machine to build a network of relationships on its own. It contains not only semantics, but also common sense. It needs to be pointed out here that Lenovo activation is a search method, which itself is not a necessary step in the application of the present invention, and can be replaced by other search methods that can achieve similar purposes. When using associative activation, the machine can consider the feature map of each memory whose activation value exceeds the preset value as having been used once, and maintain their memory value according to the memory and forgetting mechanism in the memory bank to which the memory belongs.
5，建立响应执行能力。5. Establish response and execution capabilities.
模仿是人类存在于基因里的能力。比如对一个呀呀学语的孩子，如果每次他(她)回家后，我们和他(她)打招呼，说“你回来了”。经过几次后，当他(她)再次回家时，他(她)会主动说“你回来了”。这表明他(她)在并不理解信息含义的情况下，就已经开始模仿他人进行学习。同理，我们让机器学习也采用同样的方法。机器也是模仿他人或者自己的经验来对输入信息理解并做出响应的。Imitation is the ability of human beings to exist in genes. For example, for a babbling child, if every time he (she) returns home, we greet him (her) and say "you are back." After several times, when he (she) goes home again, he (she) will take the initiative to say "you are back". This shows that he (she) has begun to imitate others to learn without understanding the meaning of the information. In the same way, we let machine learning use the same method. The machine also imitates the experience of others or its own to understand and respond to the input information.
执行响应步骤是一个翻译过程。如果在选择各种可能的响应步骤中，机器选用的是语音输出，这就比较简单，只需要把准备输出的图像特征图转变为语音，然后利用关系网络和记忆，采用概念替换的方法把动态特征图(包括表示关系的概念)和静态概念结合起来，组织成语言输出序列，并调用发音经验来实施就可以了。需要指出，机器可能根据经验(自己或者他人经验)，选用一些表达整个句子的动态特征(比如使用语气、音频音调或者重音变化的不同运动模式，来表达疑问、嘲弄、不信任、强调重点等人类常用方式)。因为机器是从人类生活中学习到这些表达方式的，所以人类有的表达方式，理论上机器都可以学习到。Performing the response step is a translation process. If in selecting various possible response steps, the machine uses voice output, which is relatively simple. It only needs to convert the image feature map to be output into voice, and then use the relational network and memory to change the dynamic The feature map (including the concept that represents the relationship) is combined with the static concept, organized into a language output sequence, and the pronunciation experience is used to implement it. It needs to be pointed out that the machine may choose some dynamic features that express the entire sentence based on experience (self or other people's experience) (such as using different movement patterns of tone, audio pitch, or stress changes to express doubts, mockery, distrust, emphasizing key points, etc.) Common way). Because machines learn these expressions from human life, in theory, machines can learn all the expressions that humans have.
如果机器选用的是动作输出，或者是语音和动作混合输出，那么问题就会变得复杂很多。这相当于组织起一场活动。机器的响应计划中，可能只有主要步骤和最终目标，其余都需要在实践中随机应变。If the machine uses motion output, or a mixed output of voice and motion, the problem will become much more complicated. This is equivalent to organizing an event. In the machine's response plan, there may only be the main steps and the final goal, and the rest need to be changed in practice.
5.1，机器需要把准备输出的图像特征图序列作为目标(这是中间目标和最终目标)，按 照这些目标涉及到不同的时间和空间。机器需要对它们在时间和空间上做划分，便于协调自己的执行效率。采用的方法是通过选择时间上紧密联系的目标和空间上紧密联系的目标作为分组。因为动态特征图和静态特征图结合后构成的信息组合，其相关记忆的环境空间是带有时间和空间信息的，所以这一步可以采用归类方法。(这一步相当于从总剧本改写到分剧本)。5.1. The machine needs to target the image feature map sequence to be output (this is the intermediate target and the final target). According to these targets, different time and space are involved. The machine needs to divide them in time and space in order to coordinate their execution efficiency. The method adopted is to select groups that are closely related in time and that are closely related in space. Because the dynamic feature map and the static feature map are combined to form an information combination, the environment space of the related memory contains time and space information, so this step can use the classification method. (This step is equivalent to rewriting from the overall script to the sub-script).
5.2，机器需要把每个环节中的中间目标，再次结合现实环境，采用分段模仿的方法，来逐层展开。机器在顶层提出的响应计划，通常只是使用概括性很高的过程特征，和概括性很高的静态概念组成的(因为这些概括性很高的过程才能找到多个相似的记忆，所以借鉴它们建立的响应也是高度概括的)。比如“出差”这个总输出响应下面，“去机场”是一个中间环节目标。但这个目标依然很抽象，机器是无法执行模仿的。5.2. The machine needs to combine the intermediate targets in each link again with the real environment and adopt the method of segmented imitation to expand layer by layer. The response plan proposed by the machine at the top level is usually only composed of highly generalized process features and highly generalized static concepts (because these highly generalized processes can find multiple similar memories, so learn from them to establish The response is also highly general). For example, under the total output response of "business trip", "going to the airport" is an intermediate link goal. But this goal is still very abstract, and machines cannot perform imitation.
所以机器需要按照时间和空间划分，把在目前时间和空间中，需要执行的环节作为目前的目标。而把其他时间和空间的目标作为继承目标，暂时放到一边。机器把中间环节作为目标后，机器还是需要进一步细分时间和空间(再次写下级分剧本)。这是一个时间和空间分辨率不断增加的过程。机器把一个目标转换成多个中间环节目标的过程，依然是一个创建各种可能响应，并使用评估***来评估，按照“趋利避害”的原则来选择自己的响应的过程。上述过程是不断迭代，每一个目标划分成多个中间目的的过程是完全相似的处理流程。一直要分解到机器的底层经验为止。底层经验对语言来说就是调动肌肉发出音节。对动作而言，就是分解到对相关“肌肉”发出驱动命令。这是一个塔形分解结构。机器从顶层目标开始，把一个目标分解成多个中间环节目标。这个过程就是创建虚拟的中间过程目标，如果这些中间过程目标“符合要求”就保留。如果“不符合要求”就重新创建。这个过程逐层展开，最终建立机器丰富多彩的响应。Therefore, the machine needs to be divided according to time and space, and the link that needs to be executed in the current time and space is the current goal. And take other goals in time and space as inheritance goals and put them aside for the time being. After the machine takes the intermediate link as the target, the machine still needs to further subdivide time and space (write down the score script again). This is a process of increasing temporal and spatial resolution. The process by which a machine converts a target into multiple intermediate links is still a process of creating various possible responses, using an evaluation system to evaluate them, and selecting their own responses according to the principle of "seeking advantages and avoiding disadvantages". The above process is continuous iteration, and the process of dividing each goal into multiple intermediate goals is a completely similar processing flow. It has to be broken down to the bottom experience of the machine. For language, the bottom experience is to mobilize muscles to make syllables. In terms of action, it is broken down to issuing drive commands to related “muscles”. This is a tower-shaped decomposition structure. The machine starts from the top-level goal and decomposes a goal into multiple intermediate-link goals. This process is to create virtual intermediate process goals, if these intermediate process goals "meet the requirements", keep them. If "does not meet the requirements", re-create it. This process unfolds layer by layer, and finally establishes the colorful response of the machine.
5.3，在这个过程中，机器随时可能碰到新信息，导致机器需要处理各种信息，而这些原来的目标就变成继承动机。这就相当于组织活动的过程中，不断碰到新情况，需要立即解决，否者活动就无法组织下去了。于是导演叫停其他活动，先来解决眼前碰到的问题。解决 后，活动继续进行。另外一种情况就是在这个过程中，导演突然接到一个新任务，于是导演权衡利弊后，决定活动先暂停，优先处理新任务。5.3. In this process, the machine may encounter new information at any time, causing the machine to process all kinds of information, and these original goals become inheritance motivation. This is equivalent to the process of organizing activities, constantly encountering new situations that need to be resolved immediately, otherwise the activities will not be able to be organized. So the director called to stop other activities, first to solve the immediate problems. After resolution, the activity continues. Another situation is that during this process, the director suddenly received a new task, so after weighing the pros and cons, the director decided to suspend the activity first and deal with the new task first.
5.4，机器是一边执行可以进行的模仿任务，一边分解其他目标到更细致目标的。所以机器是边做边想的。这是因为现实情况千差万别，机器不可能事先都知道外界情况而做出计划。所以这是一个环境和机器互动来完成的一个目标的过程。5.4, the machine is to perform imitation tasks that can be performed while decomposing other goals into more detailed goals. So the machine is thinking while doing it. This is because the reality is very different, and it is impossible for the machine to know the external situation in advance and make a plan. So this is a process in which the environment and the machine interact to complete a goal.
至此，机器就完成了对一次信息输入的理解和响应。这个过程作为机器和外界互动的一个最小周期，会不断被重复使用来完成更大的目标。At this point, the machine has completed the understanding and response to an information input. This process is a minimal cycle of interaction between the machine and the outside world, and it will be repeatedly used to accomplish greater goals.
6，更新记忆库。6. Update the memory bank.
更新记忆库是贯穿于所有步骤中的，它不是一个单独的步骤，是关系提取机制的实现。在S1步骤中，建立底层特征主要是使用记忆和遗忘机制。机器通过局部视野每发现一个相似的局部特征，如果特征图库中已经有相似的局部特征，就按照记忆曲线增加它的记忆值。如果特征图库中没有相似的局部特征，就把它存入特征图，并赋予它初始记忆值。所有特征图库中的记忆值随时间或者训练时间(随训练样本数量增长)而按照遗忘曲线逐渐递减。最终那些广泛存在于各种事物中的，共有的简单特征会拥有高记忆值，成为底层特征图。Updating the memory bank runs through all the steps. It is not a separate step, but the realization of the relationship extraction mechanism. In step S1, the establishment of low-level features is mainly to use memory and forgetting mechanisms. Each time the machine finds a similar local feature through the local field of view, if there are already similar local features in the feature library, it will increase its memory value according to the memory curve. If there is no similar local feature in the feature library, store it in the feature map and give it an initial memory value. The memory values in all feature libraries gradually decrease according to the forgetting curve with time or training time (increasing with the number of training samples). In the end, the simple features that are widely present in various things will have high memory value and become the underlying feature map.
在S2步骤中，每发现一个底层特征或者特征图，如果临时记忆库、特征图库或者记忆中已经有相似的底层特征或者特征图，它的记忆值就按照记忆曲线增加。它们也遵从遗忘机制。在S2步骤中，机器首先把环境空间存入到临时记忆库。机器在记忆库中存储这些环境空间时，会同时存储环境空间中的特征图和它们的记忆值，这些特征图的初始记忆值和其存储发生时的激活值正相关。在S3、S4、S5和S6步骤中，记忆库中特征图记忆值遵从记忆和遗忘机制。记忆中某一个关系每当被使用一次，就对这个关系涉及到的特征图按照记忆曲线增加记忆值，同时所有特征图按照自己所在的记忆库的遗忘曲线对记忆值进行遗忘。In step S2, each time a bottom-level feature or feature map is found, if there are already similar bottom-level features or feature maps in the temporary memory library, feature library, or memory, its memory value increases according to the memory curve. They also follow the forgetting mechanism. In step S2, the machine first saves the environment space into the temporary memory bank. When the machine stores these environment spaces in the memory bank, it will also store the feature maps in the environment space and their memory values. The initial memory values of these feature maps are positively correlated with the activation values when their storage occurs. In steps S3, S4, S5 and S6, the memory value of the feature map in the memory bank complies with the memory and forgetting mechanism. Whenever a relationship in the memory is used once, the feature map involved in this relationship will increase the memory value according to the memory curve, and all the feature maps will forget the memory value according to the forgetting curve of the memory bank in which they are located.
本发明中，我们可以采用多种记忆组织形式，比如：In the present invention, we can use a variety of memory organization forms, such as:
6.1，直接采用信息输入的时间和空间关系，按照顺序存储，并建立立体坐标来表示信息之间 的距离。这个坐标的时间轴可以按照事件驱动机制：每发生一个事件驱动，存储一次记忆，时间轴就增加一个单位。6.1. Directly adopt the time and space relationship of information input, store them in order, and establish three-dimensional coordinates to represent the distance between information. The time axis of this coordinate can be driven by an event-driven mechanism: each time an event occurs, the memory is stored once, and the time axis increases by one unit.
6.2，把特征建立编号，每个编号和特征自身采用表格的形式对应起来。在记忆空间中，使用编码来代替特征(或者使用特征本身，但附带上编码)。这些编码可以按照相似性来逐层分类，机器只需要根据编码的分类信息就可以快速找到相似的特征。6.2. Create a number for the feature, and each number corresponds to the feature itself in the form of a table. In the memory space, use codes instead of features (or use the feature itself, but with the code attached). These codes can be classified layer by layer according to similarity, and the machine can quickly find similar features only according to the classification information of the code.
6.3，把相似的特征放在一起，但每个特征都带有自己的记忆空间中的立体坐标。这样，机器就能迅速找到所有的相似特征，并根据这些特征的空间坐标信息，去实现临近激活和强记忆激活。6.3. Put similar features together, but each feature has its own three-dimensional coordinates in the memory space. In this way, the machine can quickly find all similar features, and according to the spatial coordinate information of these features, to achieve proximity activation and strong memory activation.
6.4，模仿大脑神经组织，在相邻的记忆之间建立连接关系。通过这种连接关系模仿激活电信号的传播和衰减。同时，每个特征接收激活电信号也模仿大脑神经，采用记忆值高的特征接收能力强，并且特征的接收能力与激活电信号和自己的匹配程度正相关。6.4, imitating the brain's nerve tissue and establishing connections between adjacent memories. Through this connection relationship, the propagation and attenuation of the active electrical signal are simulated. At the same time, each feature receives the activated electrical signal and also imitates the brain nerves, and the feature with high memory value is used to receive strong, and the feature's receiving ability is positively correlated with the degree of matching between the activated electrical signal and itself.
6.5，还可以是上述形式的组合形式。6.5, it can also be a combination of the above forms.
无论采用哪种形式的信息存储组织方式，只要组织的目的是为了实现联想激活过程，那么它就是本发明申请中所提出方法的一种具体实施方式。No matter which form of information storage organization is adopted, as long as the purpose of the organization is to realize the association activation process, it is a specific implementation of the method proposed in the application of the present invention.
对记忆库中的记忆值的刷新过程，也可以采用不同的方式，比如在每个记忆库中采用单独的记忆值刷新模块，或者整机采用一个记忆值刷星模块，或者通过实现联想激活过程的程序或者硬件来实现记忆刷新，只要它们的目的是针对类似于本发明申请中的记忆库实现记忆和遗忘机制，它们就是本发明申请中所提出方法的一种具体实施方式。For the refresh process of the memory value in the memory bank, different methods can also be adopted, for example, a separate memory value refresh module is used in each memory bank, or the whole machine uses a memory value refresh module, or through the realization of the association activation process As long as their purpose is to achieve a memory and forgetting mechanism similar to the memory bank in the application of the present invention, they are a specific implementation of the method proposed in the application of the present invention.
7，实施示意图。7. Schematic diagram of implementation.
[根据细则91更正 13.08.2020]　
在本发明申请中，运算量最大的是多分辨率特征提取和联想激活两个过程。所以在本发明申请中提出可以采用单独的硬件算法来提高机器运算效率。图3是一种实现通用机器智能的模块组成示意图。图3所示方法的核心思路是采用一个单独的模块来实现联想激活过程。这个模块得到输入信息后，就搜索记忆找到和输入信息临近信息、相似性信息和强记忆值信 息。然后按照相应的算法直接给记忆中这些信息赋予激活值。然后，在通过寻找这些被激活的信息相关的临近信息、相似性信息和强记忆值信息，再次按照相应的算法直接给记忆中这些信息赋予激活值。这个过程迭代进行，直到联想激活过程因为每个信息都存在激活预置而停止。这是一个从“上帝”视角，采用预置算法，根据记忆空间的空间距离、记忆值和相似性直接采用外部算法来完成联想激活。记忆中的记忆值和激活值随时间而消退的算法，既可以在记忆库中使用软件或者硬件来刷新，也可以使用记忆联想激活模块来实现。其中S600是建立机器特征提取模块。这个模块是通过对比局部相似性来选择数据在不同分辨率下的静态特征和动态特征，并建立对比相似性或者训练神经网络，或者其他任何已有算法来提取数据的特征。其中S601和S602模块是从外部输入信息中提取多分辨率信息特征的模块，它们涉及到不同的分辨率。机器可能需要在多种分辨率下对输入数据进行特征提取。在S601中，可以通过预处理把同一传感器数据分成多路数据来提取数据的不同特征。S602中可以通过不同分辨率下，再次使用不同的预处理算法，来提取不同分辨率下的数据特征。为了提高效率，S601和S602可以采用单独的硬件来完成多分辨率特征提取功能。完成输入信息提取后，机器在S603中，可以包含两个模块。其中一个是用于联想激活专用模块，它可以是一块专用的搜索硬件。这样做的目的是为了把搜索记忆和赋予激活值算法固化，通过采用专门的硬件来提高效率。另外一个是组合记忆信息和现实信息的模块，它相当于实现数据重组的软件。这一步，主要是通过从相关记忆中寻找动态过程，然后通过动作特征的泛化能力，把经验泛化。S604是整个记忆库(包括为了提高搜索效率而建立的快速搜索库，它包含常用记忆信息。也包含临时记忆库、长期记忆库和可能有的其他记忆库)。记忆库相当于存储空间，但它带有每个信息的生命周期(记忆值)。记忆库可以采用专门的记忆值刷新模块来维护记忆值。S605是需求评估***，它利用S603过程中获得的需求值来做逻辑判断。S605可以是软件实现。S606是分段模仿过程(迭代进行概念展开的过程)，这个过程需要不断调用S603和S604，它可以是软件实现。S607是一个逻辑判断，它可以是软件实现。S608是新记忆的存储过程， 它可以是软件实现，也可以使用专门的硬件来实现。新记忆包含机器的内外输入信息，机器的需求信息和机器的情绪信息。它们是首先存储到临时记忆库中。S609是完成一个信息响应周期的状态。在图3的实施方案中，其特征在于可以使用单独的硬件来实施多分辨率特征提取和采用单独的硬件来实施联想激活过程。[Corrected according to Rule 91 13.08.2020]
In the application of the present invention, the two processes with the largest amount of calculation are multi-resolution feature extraction and association activation. Therefore, it is proposed in the application of the present invention that a separate hardware algorithm can be used to improve the computing efficiency of the machine. Figure 3 is a schematic diagram of a module for realizing general machine intelligence. The core idea of the method shown in Figure 3 is to use a separate module to implement the association activation process. After this module gets the input information, it searches memory to find the proximity information, similarity information and strong memory value information to the input information. Then according to the corresponding algorithm, the activation value is directly assigned to the information in the memory. Then, by looking for the proximity information, similarity information and strong memory value information related to the activated information, the activation value is directly assigned to the information in the memory according to the corresponding algorithm again. This process is iterative until the Lenovo activation process stops because each message has an activation preset. This is a "God" perspective, using a preset algorithm, and directly using an external algorithm to complete the association activation according to the spatial distance, memory value and similarity of the memory space. The algorithm that the memory value and the activation value in the memory fade with time can either be refreshed in the memory bank using software or hardware, or it can be implemented using the memory association activation module. Among them, S600 is to establish a machine feature extraction module. This module selects the static features and dynamic features of the data at different resolutions by comparing the local similarity, and establishes the contrast similarity or trains the neural network, or any other existing algorithms to extract the features of the data. Among them, the S601 and S602 modules are modules that extract the features of multi-resolution information from external input information, and they involve different resolutions. The machine may need to perform feature extraction on input data at multiple resolutions. In S601, the same sensor data can be divided into multiple channels of data through preprocessing to extract different characteristics of the data. In S602, different preprocessing algorithms can be used again at different resolutions to extract data features at different resolutions. In order to improve efficiency, S601 and S602 can use separate hardware to complete the multi-resolution feature extraction function. After the input information is extracted, the machine can include two modules in S603. One of them is a dedicated module for Lenovo activation, which can be a dedicated search hardware. The purpose of this is to solidify the search memory and assign activation value algorithms, and improve efficiency by using specialized hardware. The other is a module that combines memory information and reality information, which is equivalent to software that realizes data reorganization. This step is mainly to find the dynamic process from the relevant memory, and then generalize the experience through the generalization ability of the action characteristics. S604 is the entire memory bank (including the quick search library established to improve search efficiency, which contains commonly used memory information. It also includes temporary memory banks, long-term memory banks, and possibly other memory banks). The memory bank is equivalent to storage space, but it carries the life cycle (memory value) of each information. The memory bank can use a special memory value refresh module to maintain the memory value. S605 is a demand assessment system, which uses the demand value obtained in the S603 process to make logical judgments. S605 can be implemented in software. S606 is a segmented imitation process (a process of iterative concept development). This process requires constant calls to S603 and S604, which can be implemented by software. S607 is a logical judgment, and it can be realized by software. S608 is a newly memorized storage process, which can be implemented by software or dedicated hardware. The new memory contains the internal and external input information of the machine, the demand information of the machine and the emotional information of the machine. They are first stored in the temporary memory bank. S609 is the state of completing an information response cycle. In the embodiment of FIG. 3, it is characterized in that separate hardware can be used to implement multi-resolution feature extraction and separate hardware can be used to implement the association activation process.
图4是另外一种实现通用机器智能的模块组成示意图。图4所示方法的核心思路是把实现联想激活过程的算法分布式地集成在记忆库模块中。在图4中，S704是一个能模仿大脑记忆功能，实现临近激活、强记忆激活和相似性激活功能的记忆库。它模仿大脑的方式来接收特征传递过来的激励电信号，并按照记忆空间的距离在记忆中实现激励电信号的传播和衰减，同时也模仿大脑来实现强记忆激活。记忆模块本身也可以集成搜索算法来实现相似性激活，而实现相似性激活也可以有很多方法。它们需要根据不同的记忆库数据组织方法来具体实施。图4中的其余部分和图3中相同。Figure 4 is a schematic diagram of another module for realizing general machine intelligence. The core idea of the method shown in Figure 4 is to integrate the algorithm for realizing the association activation process in a memory module in a distributed manner. In Figure 4, S704 is a memory bank that can imitate the memory function of the brain and realize the functions of proximity activation, strong memory activation and similarity activation. It mimics the way of the brain to receive the electrical excitation signal transmitted by the characteristic, and realizes the propagation and attenuation of the electrical excitation signal in the memory according to the distance of the memory space, and also imitates the brain to achieve strong memory activation. The memory module itself can also integrate search algorithms to achieve similarity activation, and there are many ways to achieve similarity activation. They need to be implemented according to different memory data organization methods. The remaining parts in FIG. 4 are the same as those in FIG. 3.
当然，还可以采用类似于图3中的集中式和图4中的分布式建立混合实现联想激活功能的方式。Of course, it is also possible to adopt a method similar to the centralized establishment in FIG. 3 and the distributed establishment in FIG. 4 to realize the association activation function.

Claims (14)

  1. 一种在记忆中寻找和输入信息相关记忆的方法，其特征包括：A method for finding and inputting information-related memories in memory, its characteristics include:
    当信息输入时，采用“临近激活”、“强记忆激活”和“相似性激活”的方法，来寻找和输入信息相关的记忆。When information is input, the methods of "proximity activation", "strong memory activation" and "similarity activation" are used to find memories related to the input information.
  2. 一种把信息存储到记忆中的方法，其特征包括：A method of storing information in memory, its characteristics include:
    机器对输入信息做多分辨率特征提取并存储这些多分辨率特征时，按照保留信息之间原来的相似性关系、时间和空间关系来存储信息特征到记忆库中；机器使用数值或者符号来表示这些信息能在记忆库中存在的时间，它们称为记忆值；记忆值可以和其对应的特征存储在一起，也可以分开存储。When the machine extracts multi-resolution features from the input information and stores these multi-resolution features, it stores the information features in the memory according to the original similarity relationship, time and space relationship between the retained information; the machine uses values or symbols to represent The time this information can exist in the memory bank is called the memory value; the memory value can be stored with its corresponding characteristics or stored separately.
  3. 根据权利要求2所述的方法中，一种记忆信息的组织方式，其特征包括：The method according to claim 2, a way of organizing memory information, characterized by:
    直接采用信息输入的时间和空间关系，按照顺序存储，并建立立体坐标来表示信息之间的距离；这个坐标的时间轴可以按照事件驱动机制：每发生一个事件驱动，存储一次记忆，时间轴就增加一个单位。Directly use the time and space relationship of information input, store them in order, and establish a three-dimensional coordinate to represent the distance between information; the time axis of this coordinate can be driven by an event-driven mechanism: every time an event occurs, the memory is stored once, and the time axis is Add a unit.
  4. 根据权利要求2所述的方法中，另外一种记忆信息的组织方式，其特征包括：The method according to claim 2, wherein another way of organizing memory information is characterized by:
    把输入特征建立编码，每个编码和特征自身采用表格的形式对应起来；在记忆空间中，使用编码来代替特征(或者使用特征本身，但附带上编码)；这些编码可以按照相似性来逐层分类，机器只需要根据编码的分类信息就可以快速找到相似的特征。Create codes for the input features, and each code corresponds to the feature itself in the form of a table; in the memory space, use the code to replace the feature (or use the feature itself, but with the code attached); these codes can be layered according to the similarity Classification, the machine can quickly find similar features only according to the encoded classification information.
  5. 根据权利要求2所述的方法中，另外一种记忆信息的组织方式，其特征包括：The method according to claim 2, wherein another way of organizing memory information is characterized by:
    把相似的特征放在一起，但每个特征都带有自己的记忆空间中的立体坐标。Put similar features together, but each feature has its own three-dimensional coordinates in the memory space.
  6. 根据权利要求2所述的方法中，另外一种记忆信息的组织方式，其特征包括：The method according to claim 2, wherein another way of organizing memory information is characterized by:
    在记忆中相邻的信息之间建立连接关系，通过这种连接关系模仿激活电信号的传播和衰减；同时，每个特征接收激活电信号的能力也和自身的记忆值成正相关，也和激活源和自己的相似程度正相关。A connection relationship is established between adjacent information in memory, and the propagation and attenuation of the activation electrical signal is simulated through this connection relationship; at the same time, the ability of each feature to receive the activation electrical signal is also positively correlated with its own memory value, and also with activation The similarity between the source and itself is positively correlated.
  7. 一种通用机器智能实现方法，其特征包括：A general machine intelligence realization method, its characteristics include:
    机器包含对输入信息做多分辨率特征提取模块和联想激活算法模块，这些模块可以采用硬件实现。The machine includes a multi-resolution feature extraction module and an association activation algorithm module for the input information. These modules can be implemented by hardware.
  8. 根据权利要求7所述的方法中，一种联想激活方式，其特征包括：The method according to claim 7, an association activation method, characterized by:
    机器的联想激活功能由联想激活算法模块通过修改记忆中特征的激活值来实现。The association activation function of the machine is realized by the association activation algorithm module by modifying the activation value of the feature in the memory.
  9. 根据权利要求2所述的方法中，一种提高在记忆中搜索数据的方法，其特征包括：The method according to claim 2, a method for improving the search for data in memory, characterized by:
    机器把存在于记忆中的常用关系网络，提取出来，构成一个可以提高搜索效率的单独常用记忆库。The machine extracts the common relational network existing in the memory to form a separate common memory bank that can improve the search efficiency.
  10. 一种记忆信息的组织方式，其特征包括：An organization method for remembering information, its characteristics include:
    机器存储的记忆中，有三类数据，第一类是外部输入的信息特征；第二类是内部自身信息；第三类是机器需求和需求所处状态的数据、情绪和情绪所处状态的数据。In the memory stored by the machine, there are three types of data. The first type is the information characteristics of external input; the second type is internal information; the third type is data about the state of machine needs and needs, and data about the state of emotions and emotions. .
  11. 根据权利要求10所述的方法中，其特征包括：The method according to claim 10, characterized by comprising:
    机器按照时间顺序存储这三类数据，并且存储时，数据被赋予的初始记忆值和存储发生时数据的激活值成正相关。The machine stores these three types of data in chronological order, and when storing, the initial memory value assigned to the data is positively correlated with the activation value of the data when the storage occurs.
  12. 一种训练多层神经网络的方法，其特征包括：A method for training a multilayer neural network, its characteristics include:
    机器首先提取输入信息的多分辨率下的信息特征，然后采用部分分辨率下的特征来训练神经网络。The machine first extracts the information features of the input information at multiple resolutions, and then uses the features at partial resolutions to train the neural network.
  13. 根据权利要求12所述的方法中，其特征包括：The method according to claim 12, wherein the features include:
    机器可以针对不同分辨率下的输入信息特征，按照分辨率分组，单独训练多层神经网络，然后把多个神经网络的输出加权平均，作为总的输出。The machine can train multi-layer neural networks separately according to the input information characteristics at different resolutions, group them according to the resolution, and then weight the outputs of multiple neural networks as the total output.
  14. 一种记忆调用和存储方法，其特征包括：A memory recall and storage method, its characteristics include:
    机器把语言输入的信息流，转变成非语言信息流，并把这些非语言信息流作为输入信息流，按照存储输入信息流的方式存储这些信息流。The machine converts the information flow of language input into non-verbal information flow, uses these non-verbal information flows as input information flows, and stores these information flows in the same way as the input information flows.
PCT/CN2020/000109 2020-05-11 2020-05-15 Method for imitating human memory to realize universal machine intelligence WO2021226731A1 (en)

Applications Claiming Priority (2)

Application Number Priority Date Filing Date Title
CN202010400739.7 2020-05-11
CN202010400739.7A CN111582457A (en) 2020-05-11 2020-05-11 Method for realizing general machine intelligence by simulating human memory

Publications (1)

Publication Number Publication Date
WO2021226731A1 true WO2021226731A1 (en) 2021-11-18

Family

ID=72117273

Family Applications (1)

Application Number Title Priority Date Filing Date
PCT/CN2020/000109 WO2021226731A1 (en) 2020-05-11 2020-05-15 Method for imitating human memory to realize universal machine intelligence

Country Status (2)

Country Link
CN (1) CN111582457A (en)
WO (1) WO2021226731A1 (en)

Families Citing this family (4)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN112016664A (en) * 2020-09-14 2020-12-01 陈永聪 Method for realizing humanoid universal artificial intelligence machine
CN112215346B (en) * 2020-10-20 2021-11-02 陈永聪 Method for realizing humanoid universal artificial intelligence machine
WO2021218614A1 (en) * 2020-04-30 2021-11-04 陈永聪 Establishment of general artificial intelligence system
WO2022109759A1 (en) * 2020-11-25 2022-06-02 陈永聪 Method for implementing humanlike artificial general intelligence

Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6052679A (en) * 1997-09-11 2000-04-18 International Business Machines Corporation Artificial neural networks including Boolean-complete compartments
CN101576445A (en) * 2009-06-03 2009-11-11 重庆大学 Data reappearing method for structure health monitoring failure sensor simulating memory of people
US20100306227A1 (en) * 2009-06-02 2010-12-02 Fleming James S Methods, systems and computer program products for providing a distributed associative memory base
CN105981053A (en) * 2013-10-28 2016-09-28 英特尔公司 Methods, systems and computer program products for using a distributed associative memory base to determine data correlations and convergence therein
CN109202921A (en) * 2017-07-03 2019-01-15 北京光年无限科技有限公司 The man-machine interaction method and device based on Forgetting Mechanism for robot
CN109993707A (en) * 2019-03-01 2019-07-09 华为技术有限公司 Image de-noising method and device
CN111047482A (en) * 2019-11-14 2020-04-21 华中师范大学 Knowledge tracking system and method based on hierarchical memory network

Patent Citations (7)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US6052679A (en) * 1997-09-11 2000-04-18 International Business Machines Corporation Artificial neural networks including Boolean-complete compartments
US20100306227A1 (en) * 2009-06-02 2010-12-02 Fleming James S Methods, systems and computer program products for providing a distributed associative memory base
CN101576445A (en) * 2009-06-03 2009-11-11 重庆大学 Data reappearing method for structure health monitoring failure sensor simulating memory of people
CN105981053A (en) * 2013-10-28 2016-09-28 英特尔公司 Methods, systems and computer program products for using a distributed associative memory base to determine data correlations and convergence therein
CN109202921A (en) * 2017-07-03 2019-01-15 北京光年无限科技有限公司 The man-machine interaction method and device based on Forgetting Mechanism for robot
CN109993707A (en) * 2019-03-01 2019-07-09 华为技术有限公司 Image de-noising method and device
CN111047482A (en) * 2019-11-14 2020-04-21 华中师范大学 Knowledge tracking system and method based on hierarchical memory network

Also Published As

Publication number Publication date
CN111582457A (en) 2020-08-25

Similar Documents

Publication Publication Date Title
WO2021226731A1 (en) Method for imitating human memory to realize universal machine intelligence
Kreutzer et al. Understanding artificial intelligence
Bhattacharya et al. Text2gestures: A transformer-based network for generating emotive body gestures for virtual agents
Rutkowski Computational
WO2021217282A1 (en) Method for implementing universal artificial intelligence
WO2021223042A1 (en) Method for implementing machine intelligence similar to human intelligence
Mehrotra Basics of artificial intelligence & machine learning
US11715291B2 (en) Establishment of general-purpose artificial intelligence system
Custers et al. Law and artificial intelligence: regulating AI and applying AI in legal practice
CN112215346B (en) Method for realizing humanoid universal artificial intelligence machine
Suglia et al. Imagining grounded conceptual representations from perceptual information in situated guessing games
Ayesh et al. Models for computational emotions from psychological theories using type I fuzzy logic
Hill et al. Understanding early word learning in situated artificial agents
CN114492465B (en) Dialogue generation model training method and device, dialogue generation method and electronic equipment
CN113962353A (en) Method for establishing strong artificial intelligence
CN112016664A (en) Method for realizing humanoid universal artificial intelligence machine
Луценко РЕВОЛЮЦИЯ В СИСТЕМАХ ИСКУССТВЕННОГО ИНТЕЛЛЕКТА 20-Х ГОДОВ XXI ВЕКА И СИСТЕМЫ С ИНТЕРФЕЙСОМ «ДУША-КОМПЬЮТЕР» КАК БЛИЖАЙШИЙ ОЧЕРЕДНОЙ ЭТАП РАЗВИТИЯ ИНТЕЛЛЕКТУАЛЬНЫХ ТЕХНОЛОГИЙ
WO2022016299A1 (en) Method for establishing strong artificial intelligence
Viliunas et al. Shape-finding in biophilic architecture: application of AI-based tool
CN114127748A (en) Memory in personal intelligence
WO2022109759A1 (en) Method for implementing humanlike artificial general intelligence
Yue A world-self model towards understanding intelligence
Olivier Explorations of the ‘transhuman’dimension of artificial intelligence
Ogiso et al. Expression of Emotion in Robots Using a Flow ofArtificial Consciousness
KR102661537B1 (en) Device, system, and method for providing language learning in virtual space

Legal Events

Date Code Title Description
121 Ep: the epo has been informed by wipo that ep was designated in this application

Ref document number: 20935418

Country of ref document: EP

Kind code of ref document: A1

NENP Non-entry into the national phase

Ref country code: DE

122 Ep: pct application non-entry in european phase

Ref document number: 20935418

Country of ref document: EP

Kind code of ref document: A1