WO2021151345A1 - Method and apparatus for parameter acquisition for recognition model, electronic device, and storage medium - Google Patents
Method and apparatus for parameter acquisition for recognition model, electronic device, and storage medium Download PDFInfo
- Publication number
- WO2021151345A1 WO2021151345A1 PCT/CN2020/131974 CN2020131974W WO2021151345A1 WO 2021151345 A1 WO2021151345 A1 WO 2021151345A1 CN 2020131974 W CN2020131974 W CN 2020131974W WO 2021151345 A1 WO2021151345 A1 WO 2021151345A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- standard
- data set
- recognition model
- loss function
- parameters
- Prior art date
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
- G06N3/08—Learning methods
Definitions
- This application relates to the field of data processing technology, and in particular to a method, device, electronic device, and computer-readable storage medium for acquiring parameters of an identification model.
- the training of a model often requires a large amount of labeled data. Manually labeling these data is not only inefficient in labeling, but also a large number of false labels, namely noisy labels, will appear during the labeling process. Instead, use the data with noisy labels.
- the model cannot be trained to obtain accurate model parameters.
- a method for acquiring parameters of a recognition model provided in this application includes:
- the loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
- the present application also provides a device for acquiring parameters of a recognition model, and the device includes:
- the training data acquisition module is used to acquire a training data set containing noise labels, and perform data standardization processing on the training data set to obtain a standard data set;
- a recognition model building module used to establish a recognition model based on a multi-layer deep neural network, and use the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;
- a transition matrix construction module which is used to construct the noise probability transition matrix of the standard data set
- a loss function construction module configured to construct a loss function based on the noise probability transition matrix
- the model parameter update module is used to calculate the update parameters of the standard recognition model by using the loss function, and replace the update parameters with the initialization parameters.
- This application also provides an electronic device, which includes:
- Memory storing at least one instruction
- the processor executes the instructions stored in the memory to implement the method for acquiring parameters of the recognition model as described below:
- the loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
- the present application also provides a computer-readable storage medium, including a storage data area and a storage program area, the storage data area stores created data, and the storage program area stores a computer program; wherein the computer program is executed by the processor as follows
- the parameter acquisition method of the recognition model :
- the loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
- FIG. 1 is a schematic flowchart of a method for acquiring parameters of a recognition model provided by an embodiment of this application;
- FIG. 2 is a schematic diagram of modules of an apparatus for acquiring parameters of a recognition model provided by an embodiment of the application;
- FIG. 3 is a schematic diagram of the internal structure of an electronic device that implements a method for acquiring parameters of a recognition model provided by an embodiment of the application;
- the execution subject of the method for acquiring parameters of the recognition model provided in the embodiment of the present application includes, but is not limited to, at least one of the electronic devices that can be configured to execute the method provided in the embodiment of the present application, such as a server and a terminal.
- the method for acquiring the parameters of the recognition model may be executed by software or hardware installed on the terminal device or the server device, and the software may be a blockchain platform.
- the server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.
- This application provides a method for acquiring parameters of a recognition model.
- FIG. 1 it is a schematic flowchart of a method for acquiring parameters of a recognition model provided by an embodiment of this application.
- the method can be executed by a device, and the device can be implemented by software and/or hardware.
- the method for acquiring parameters of the recognition model includes:
- the training data set containing noise labels means that there is some data in the training data set, but the preset standard label of the data does not correspond to the content of the data, that is, the preset standard label is data of the noise label.
- the training data set can be obtained from the blockchain node by using a python sentence with a data capture function, and the training data set can also be obtained from a database.
- the training data set is stored in different nodes of the blockchain, and the high data throughput of the blockchain can be used to improve the efficiency of obtaining the training data set.
- the performing data standardization processing on the training data set includes one or a combination of the following:
- the unique attribute value includes but is not limited to: data id and data number.
- the unique attribute value cannot describe the distribution law of the data, it will increase the content of the data, so that more additional computing resources are needed to process the data, and the efficiency of data processing is reduced. Therefore, the training data is used in the embodiment of the application.
- the centralized and unique data is deleted to improve the efficiency of subsequent data processing.
- the embodiment of the present application uses a high-dimensional mapping method to map the data in the training data set to a pre-built high-dimensional space, and then use the one-hot encoding technique to fill in the missing data.
- Utilizing the multi-dimensionality of high-bit space can improve the efficiency of searching for missing data in the training data set, and using one-hot encoding technology can improve the accuracy of data filling.
- the embodiment of the present application uses the following standardized algorithm to perform data normalization on the training data set:
- x is the standard data of data normalization
- S old data as the training data set
- S max is the maximum value of S old
- S min is the minimum value of the value S old.
- S max and S min are preset and used to limit the range of data in the training data set.
- the standard data set is obtained.
- the multi-layer deep neural network is:
- h (n) represents the network structure of the nth layer of the multilayer deep neural network.
- the multi-layer deep neural network After the multi-layer deep neural network is activated by the softmax function, it can output the predicted value of the joint distribution p(x, y) of the preset standard label y corresponding to the standard data x and x And get the predicted label of the standard data in the standard data set.
- the softmax function is an activation function for transforming the output result of the multi-layer deep neural network into a preset form.
- the output result of the multi-layer deep neural network is transformed into a probabilistic form (ie ).
- the output result of the multi-layer deep neural network transformed into a probabilistic form can intuitively see the difference between the preset standard label and the predicted label, and the model parameters are adjusted according to the difference, which is beneficial to improve the model’s performance. Training efficiency.
- the standard data set is input to the recognition model, and the recognition model is trained using the standard data set to obtain the initialization parameters of the recognition model, and it is determined that the recognition model including the initialization parameters is Standard recognition model.
- the method before establishing a recognition model based on a multi-layer deep neural network, the method further includes:
- Construct feature space Wherein, the feature space is used to store standard data sets;
- p(x) is the frequency of any standard data x in the standard data set in the feature space
- x) is the frequency of the preset standard label in the label space when the standard data x appears.
- the feature space and the label space are constructed, and the joint distribution of the standard data x and its corresponding label y in the label space is calculated as p(x, y).
- the standard data in the standard data set can be compared with the standard data. The relationship between the tags corresponding to the data is better displayed, and the efficiency of data processing is improved.
- the noise probability transition matrix of the standard data set can be expressed as:
- the size of c is the same as the number of standard data in the standard data set.
- the noise probability transition matrix represents the distribution of noise tags in the data.
- the element in the i-th row and j-th column in the noise probability transition matrix Q represents the probability of the occurrence of a noise label.
- the noise probability transition matrix of the standard data set is as follows:
- Q is the noise probability transition matrix
- ⁇ is any standard data in the standard data set
- ⁇ i is a preset standard label corresponding to ⁇
- ⁇ j is the noise label of ⁇ .
- the loss function includes but is not limited to: a backward loss function and an preceding loss function.
- the forward loss function is:
- QT is the transposed matrix of the noise probability transition matrix
- ⁇ is the error factor of the recognition model
- h is the multilayer deep neural network
- l ⁇ (h) is the loss value of the backward loss function
- y is the preset standard label of any standard data x in the standard data set
- Q is the noise probability transition matrix
- p(x, y) is the joint distribution of standard data x and the preset standard label y corresponding to x, Is the predicted value of p(x,y).
- the backward loss function is used to calculate the probability value that the label corresponding to the standard data x in the label space is a noisy label, that is, the probability that the preset standard label of the standard data x is wrong.
- the calculation of the update parameters of the standard recognition model by using the loss function includes:
- the gradient descent algorithm is used to calculate the update parameters of the standard recognition model.
- the loss function is a forward loss function and/or a backward loss function.
- the difference value when the difference value is within the preset threshold interval, it means that the recognition result of the standard recognition model is wrong. Then the gradient descent algorithm is used to update the parameters of the standard recognition model to improve the standard. Identify the accuracy of the model.
- the gradient descent algorithm includes, but is not limited to, a batch gradient descent algorithm, a stochastic gradient descent algorithm, and a small batch gradient descent algorithm.
- the difference value When the difference value is greater than the upper limit of the threshold interval, it may not be due to a recognition error of the standard recognition model that the difference value is greater than the upper limit of the threshold interval. In practical applications, due to the existence of the noise label, the preset standard label of the standard data is wrong, which will also cause the difference between the preset standard label and the predicted label to be greater than the upper limit of the threshold interval.
- the embodiment of the present application uses a loss function to calculate the probability value that the preset standard label of the standard data is a noise label, and when the probability value is less than the preset probability threshold, It indicates that the recognition result of the standard recognition model is wrong, and then the gradient descent algorithm is used to calculate the update parameters of the standard recognition model.
- the embodiment of the present application corrects the preset standard label of the standard data.
- the embodiment of the present application uses the update parameters to replace the initialization parameters. After the initialization parameters are replaced, the final recognition model can be obtained.
- the final recognition model can be used to recognize input data, and the input data includes But it is not limited to image data.
- the training data set after obtaining the training data set containing the noise label, the training data set is standardized to improve the efficiency of processing the training data; after obtaining the standard recognition model containing the initialization parameters, the standard data set is constructed
- the noise probability transition matrix is beneficial to improve the applicability of the loss function constructed from the noise transition matrix to the model, so that the loss function can be used to train more accurate model parameters; the loss function is constructed based on the noise probability transition matrix, and
- the loss function calculates the updated parameters of the standard recognition model, so that more accurate model parameters can be obtained, and the purpose of improving the accuracy of obtaining the model parameters is achieved. Therefore, the method for acquiring parameters of a recognition model proposed in this application can provide a method for improving the accuracy of acquiring model parameters.
- FIG. 2 it is a schematic diagram of the module of the parameter acquisition device of the identification model of the present application.
- the device 100 for acquiring the parameters of the recognition model described in this application can be installed in an electronic device.
- the parameter acquisition device of the recognition model may include a training data acquisition module 101, a recognition model construction module 102, a transition matrix construction module 103, a loss function construction module 104, and a model parameter update module 105.
- the module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.
- each module/unit is as follows:
- the training data acquisition module 101 is configured to acquire a training data set containing noise labels, and perform data standardization processing on the training data set to obtain a standard data set;
- the recognition model construction module 102 is configured to establish a recognition model based on a multi-layer deep neural network, and use the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;
- the transition matrix construction module 103 is used to construct the noise probability transition matrix of the standard data set
- the loss function construction module 104 is configured to construct a loss function based on the noise probability transition matrix
- the model parameter update module 105 is configured to use the loss function to calculate update parameters of the standard recognition model, and replace the update parameters with the initialization parameters.
- each module of the device for acquiring parameters of the recognition model is as follows:
- the training data acquisition module 101 is configured to acquire a training data set containing noise labels, and perform data standardization processing on the training data set to obtain a standard data set.
- the training data set containing noise labels means that there is some data in the training data set, but the preset standard label of the data does not correspond to the content of the data, that is, the preset standard label is data of the noise label.
- the training data set can be obtained from the blockchain node by using a python sentence with a data capture function, and the training data set can also be obtained from a database.
- the training data set is stored in different nodes of the blockchain, and the high data throughput of the blockchain can be used to improve the efficiency of obtaining the training data set.
- the training data acquisition module 101 performs data standardization processing on the training data set, including one or a combination of the following:
- the unique attribute value includes but is not limited to: data id and data number.
- the unique attribute value cannot describe the distribution law of the data, it will increase the content of the data, so that more additional computing resources are needed to process the data, and the efficiency of data processing is reduced. Therefore, the training data is used in the embodiment of the application.
- the centralized and unique data is deleted to improve the efficiency of subsequent data processing.
- high-dimensional mapping is used to map the data in the training data set to a pre-built high-dimensional space, and then the missing data is filled with data using a one-hot encoding technique.
- Utilizing the multi-dimensionality of high-bit space can improve the efficiency of searching for missing data in the training data set, and using one-hot encoding technology can improve the accuracy of data filling.
- this application uses the following standardized algorithm to normalize the training data set:
- x is the standard data of data normalization
- S old data as the training data set
- S max is the maximum value of S old
- S min is the minimum value of the value S old.
- S max and S min are preset and used to limit the range of data in the training data set.
- the standard data set is obtained.
- the recognition model construction module 102 is configured to establish a recognition model based on a multi-layer deep neural network, and use the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters.
- the multi-layer deep neural network is:
- h (n) represents the network structure of the nth layer of the multilayer deep neural network.
- the multi-layer deep neural network After the multi-layer deep neural network is activated by the softmax function, it can output the predicted value of the joint distribution p(x,y) of the preset standard label y corresponding to the standard data x and x And get the predicted label of the standard data in the standard data set.
- the softmax function is an activation function for transforming the output result of the multi-layer deep neural network into a preset form.
- the output result of the multi-layer deep neural network is transformed into a probabilistic form (ie ).
- the output result of the multi-layer deep neural network transformed into a probabilistic form can intuitively see the difference between the preset standard label and the predicted label, and the model parameters are adjusted according to the difference, which is beneficial to improve the model’s performance. Training efficiency.
- the standard data set is input to the recognition model, and the recognition model is trained using the standard data set to obtain the initialization parameters of the recognition model, and it is determined that the recognition model including the initialization parameters is Standard recognition model.
- the method before establishing a recognition model based on a multi-layer deep neural network, the method further includes:
- Construct feature space Wherein, the feature space is used to store standard data sets;
- p(x) is the frequency of any standard data x in the standard data set in the feature space
- x) is the frequency of the preset standard label in the label space when the standard data x appears.
- the feature space and the label space are constructed, and the joint distribution of the standard data x and its corresponding label y in the label space is calculated as p(x, y).
- the standard data in the standard data set can be compared with the standard data. The relationship between the tags corresponding to the data is better displayed, and the efficiency of data processing is improved.
- the transition matrix construction module 103 is used to construct the noise probability transition matrix of the standard data set.
- the noise probability transition matrix of the standard data set can be expressed as:
- the size of c is the same as the number of standard data in the standard data set.
- the noise probability transition matrix represents the distribution of noise tags in the data.
- the element in the i-th row and j-th column in the noise probability transition matrix Q represents the probability of the occurrence of a noise label.
- the noise probability transition matrix of the standard data set is as follows:
- Q is the noise probability transition matrix
- ⁇ is any standard data in the standard data set
- ⁇ i is a preset standard label corresponding to ⁇
- ⁇ j is the noise label of ⁇ .
- the loss function construction module 104 is configured to construct a loss function based on the noise probability transition matrix.
- the loss function includes but is not limited to: a backward loss function and an preceding loss function.
- the forward loss function is:
- QT is the transposed matrix of the noise probability transition matrix
- ⁇ is the error factor of the recognition model
- h is the multilayer deep neural network
- l ⁇ (h) is the loss value of the backward loss function
- y is the preset standard label of any standard data x in the standard data set
- Q is the noise probability transition matrix
- p(x, y) is the joint distribution of standard data x and the preset standard label y corresponding to x, Is the predicted value of p(x,y).
- the backward loss function is used to calculate the probability value that the label corresponding to the standard data x in the label space is a noisy label, that is, the probability that the preset standard label of the standard data x is wrong.
- the model parameter update module 105 is configured to use the loss function to calculate update parameters of the standard recognition model, and replace the update parameters with the initialization parameters.
- the model parameter update module 105 uses the loss function to calculate the update parameters of the standard recognition model, including:
- a gradient descent algorithm is used to calculate the update parameters of the standard recognition model.
- the loss function is a forward loss function and/or a backward loss function.
- the difference value when the difference value is within the preset threshold interval, it means that the recognition result of the standard recognition model is wrong. Then the gradient descent algorithm is used to update the parameters of the standard recognition model to improve the standard. Identify the accuracy of the model.
- the gradient descent algorithm includes, but is not limited to, a batch gradient descent algorithm, a stochastic gradient descent algorithm, and a small batch gradient descent algorithm.
- the difference value When the difference value is greater than the upper limit of the threshold interval, it may not be due to a recognition error of the standard recognition model that the difference value is greater than the upper limit of the threshold interval. In practical applications, due to the existence of the noise label, the preset standard label of the standard data is wrong, which will also cause the difference between the preset standard label and the predicted label to be greater than the upper limit of the threshold interval.
- the embodiment of the present application uses a loss function to calculate the probability value that the preset standard label of the standard data is a noise label, and when the probability value is less than the preset probability threshold, It indicates that the recognition result of the standard recognition model is wrong, and then the gradient descent algorithm is used to calculate the update parameters of the standard recognition model.
- the embodiment of the present application corrects the preset standard label of the standard data.
- the embodiment of the present application uses the update parameters to replace the initialization parameters. After the initialization parameters are replaced, the final recognition model can be obtained.
- the final recognition model can be used to recognize input data, and the input data includes But it is not limited to image data.
- the training data set after obtaining the training data set containing the noise label, the training data set is standardized to improve the efficiency of processing the training data; after obtaining the standard recognition model containing the initialization parameters, the standard data set is constructed
- the noise probability transition matrix is beneficial to improve the applicability of the loss function constructed from the noise transition matrix to the model, so that the loss function can be used to train more accurate model parameters; the loss function is constructed based on the noise probability transition matrix, and
- the loss function calculates the updated parameters of the standard recognition model, so that more accurate model parameters can be obtained, and the purpose of improving the accuracy of obtaining the model parameters is achieved. Therefore, the device for acquiring parameters of a recognition model proposed in this application can provide a method for improving the accuracy of acquiring model parameters.
- FIG. 3 it is a schematic diagram of the structure of an electronic device that implements the method for acquiring parameters of a recognition model according to the present application.
- the electronic device 1 may include a processor 10, a memory 11, and a bus, and may also include a computer program stored in the memory 11 and running on the processor 10, such as a parameter acquisition program 12 of an identification model.
- the memory 11 includes at least one type of readable storage medium, and the readable storage medium may be volatile or nonvolatile.
- the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc.
- the memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, for example, a mobile hard disk of the electronic device 1.
- the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (SmartMediaCard, SMC), and a secure digital (SecureDigital, SD) equipped on the electronic device 1. Card, flash card (FlashCard), etc.
- the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device.
- the memory 11 can be used not only to store application software and various data installed in the electronic device 1, such as the code of the parameter acquisition program 12 of the identification model, etc., but also to temporarily store data that has been output or will be output.
- the processor 10 may be composed of integrated circuits in some embodiments, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same function or different functions, including one or more Central Processing Unit (CPU), microprocessor, digital processing chip, graphics processor and a combination of various control chips, etc.
- the processor 10 is the control core (ControlUnit) of the electronic device, which uses various interfaces and lines to connect the various components of the entire electronic device, and runs or executes programs or modules stored in the memory 11 (for example, perform identification The parameter acquisition program of the model, etc.), and call the data stored in the memory 11 to execute various functions of the electronic device 1 and process data.
- ControlUnit the control core of the electronic device, which uses various interfaces and lines to connect the various components of the entire electronic device, and runs or executes programs or modules stored in the memory 11 (for example, perform identification The parameter acquisition program of the model, etc.), and call the data stored in the memory 11 to execute various functions of the electronic device 1 and process data.
- the bus may be a peripheral component interconnection standard (PCI) bus or an extended industry standard architecture (EISA) bus or the like.
- PCI peripheral component interconnection standard
- EISA extended industry standard architecture
- the bus can be divided into address bus, data bus, control bus and so on.
- the bus is configured to implement connection and communication between the memory 11 and at least one processor 10 and the like.
- FIG. 3 only shows an electronic device with components. Those skilled in the art can understand that the structure shown in FIG. 3 does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure. Components, or combinations of certain components, or different component arrangements.
- the electronic device 1 may also include a power source (such as a battery) for supplying power to various components.
- the power source may be logically connected to the at least one processor 10 through a power management device, thereby controlling power
- the device implements functions such as charge management, discharge management, and power consumption management.
- the power supply may also include any components such as one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, and power status indicators.
- the electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.
- the electronic device 1 may also include a network interface.
- the network interface may include a wired interface and/or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is usually used in the electronic device 1 Establish a communication connection with other electronic devices.
- the electronic device 1 may also include a user interface.
- the user interface may be a display (Display) and an input unit (such as a keyboard (Keyboard)).
- the user interface may also be a standard wired interface or a wireless interface.
- the display may be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light emitting diode) touch device, etc.
- the display can also be appropriately called a display screen or a display unit, which is used to display the information processed in the electronic device 1 and to display a visualized user interface.
- the parameter acquisition program 12 of the recognition model stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can realize:
- the loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
- the integrated module/unit of the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium.
- the computer-readable storage medium may be volatile or non-volatile.
- the computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read -OnlyMemory).
- the computer usable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function, etc.; the storage data area may store a block chain node Use the created data, etc.
- modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
- the functional modules in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit.
- the above-mentioned integrated unit may be implemented in the form of hardware, or may be implemented in the form of hardware plus software functional modules.
- the blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm.
- Blockchain essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block.
- the blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Health & Medical Sciences (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Image Analysis (AREA)
Abstract
A method for parameter acquisition for a recognition model, relating to data processing technology, and comprising: acquiring a training dataset containing a noise tag, and performing data standardization processing on the training dataset to obtain a standard dataset (S1); on the basis of a multi-layer deep neural network, constructing a recognition model, and using the standard dataset to train the recognition model to obtain a standard recognition model containing initialized parameters (S2); constructing a noise probability transition matrix for the standard dataset (S3); on the basis of the noise probability transition matrix, constructing a loss function (S4); and using the loss function to calculate updated parameters for the standard recognition model, thereby obtaining more accurate model parameters (S5). The described method further relates to blockchain technology, and the training dataset can be stored in nodes of a blockchain. Further disclosed are an apparatus for parameter acquisition for a recognition model, an electronic device, and a storage medium, able to improve the precision of acquired model parameters.
Description
本申请要求于2020年07月09日提交中国专利局、申请号为202010656659.8,发明名称为“识别模型的参数获取方法、装置、电子设备及存储介质”的中国专利申请的优先权,其全部内容通过引用结合在本申请中。This application claims the priority of the Chinese patent application filed with the Chinese Patent Office on July 9, 2020, the application number is 202010656659.8, and the invention title is "Methods, devices, electronic equipment and storage media for obtaining parameters of identification models", and its entire contents Incorporated in this application by reference.
本申请涉及数据处理技术领域,尤其涉及一种识别模型的参数获取方法、装置、电子设备及计算机可读存储介质。This application relates to the field of data processing technology, and in particular to a method, device, electronic device, and computer-readable storage medium for acquiring parameters of an identification model.
发明人意识到,随着人工智能的兴起,越来越多的技术人员利用带有标签的数据训练搭建好的模型,以获取需要的模型参数,进而利用模型参数让模型实现特定的功能。但一个模型的训练往往需要海量的带有标签的数据,人工手动对这些数据进行标记不仅标记效率低下且在标记过程中会出现大量错误的标签,即噪声标签,而利用带有噪声标签的数据对模型进行训练无法获取精确的模型参数。The inventor realized that with the rise of artificial intelligence, more and more technicians use labeled data to train and build models to obtain required model parameters, and then use the model parameters to enable the model to achieve specific functions. However, the training of a model often requires a large amount of labeled data. Manually labeling these data is not only inefficient in labeling, but also a large number of false labels, namely noisy labels, will appear during the labeling process. Instead, use the data with noisy labels. The model cannot be trained to obtain accurate model parameters.
因此如何利用这些带有噪声标签的数据来训练模型以获取更加精准的模型参数,成为了人们越来越关注的重点。Therefore, how to use these noisy label data to train the model to obtain more accurate model parameters has become the focus of more and more attention.
发明内容Summary of the invention
本申请提供的一种识别模型的参数获取方法,包括:A method for acquiring parameters of a recognition model provided in this application includes:
获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;Acquiring a training data set containing noise labels, and performing data standardization processing on the training data set to obtain a standard data set;
基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;Establishing a recognition model based on a multi-layer deep neural network, and using the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;
构建所述标准数据集的噪声概率转移矩阵;Constructing the noise probability transition matrix of the standard data set;
基于所述噪声概率转移矩阵构建损失函数;Constructing a loss function based on the noise probability transition matrix;
利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
本申请还提供一种识别模型的参数获取装置,所述装置包括:The present application also provides a device for acquiring parameters of a recognition model, and the device includes:
训练数据获取模块,用于获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;The training data acquisition module is used to acquire a training data set containing noise labels, and perform data standardization processing on the training data set to obtain a standard data set;
识别模型构建模块,用于基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;A recognition model building module, used to establish a recognition model based on a multi-layer deep neural network, and use the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;
转移矩阵构建模块,用于构建所述标准数据集的噪声概率转移矩阵;A transition matrix construction module, which is used to construct the noise probability transition matrix of the standard data set;
损失函数构建模块,用于基于所述噪声概率转移矩阵构建损失函数;A loss function construction module, configured to construct a loss function based on the noise probability transition matrix;
模型参数更新模块,用于利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The model parameter update module is used to calculate the update parameters of the standard recognition model by using the loss function, and replace the update parameters with the initialization parameters.
本申请还提供一种电子设备,所述电子设备包括:This application also provides an electronic device, which includes:
存储器,存储至少一个指令;及Memory, storing at least one instruction; and
处理器,执行所述存储器中存储的指令以实现如下所述的识别模型的参数获取方法:The processor executes the instructions stored in the memory to implement the method for acquiring parameters of the recognition model as described below:
获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;Acquiring a training data set containing noise labels, and performing data standardization processing on the training data set to obtain a standard data set;
基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;Establishing a recognition model based on a multi-layer deep neural network, and using the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;
构建所述标准数据集的噪声概率转移矩阵;Constructing the noise probability transition matrix of the standard data set;
基于所述噪声概率转移矩阵构建损失函数;Constructing a loss function based on the noise probability transition matrix;
利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
本申请还提供一种计算机可读存储介质,包括存储数据区和存储程序区,存储数据区存储创建的数据,存储程序区存储有计算机程序;其中,所述计算机程序被处理器执行时实现如下所述的识别模型的参数获取方法:The present application also provides a computer-readable storage medium, including a storage data area and a storage program area, the storage data area stores created data, and the storage program area stores a computer program; wherein the computer program is executed by the processor as follows The parameter acquisition method of the recognition model:
获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;Acquiring a training data set containing noise labels, and performing data standardization processing on the training data set to obtain a standard data set;
基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;Establishing a recognition model based on a multi-layer deep neural network, and using the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;
构建所述标准数据集的噪声概率转移矩阵;Constructing the noise probability transition matrix of the standard data set;
基于所述噪声概率转移矩阵构建损失函数;Constructing a loss function based on the noise probability transition matrix;
利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
图1为本申请一实施例提供的识别模型的参数获取方法的流程示意图;FIG. 1 is a schematic flowchart of a method for acquiring parameters of a recognition model provided by an embodiment of this application;
图2为本申请一实施例提供的识别模型的参数获取装置的模块示意图;2 is a schematic diagram of modules of an apparatus for acquiring parameters of a recognition model provided by an embodiment of the application;
图3为本申请一实施例提供的实现识别模型的参数获取方法的电子设备的内部结构示意图;3 is a schematic diagram of the internal structure of an electronic device that implements a method for acquiring parameters of a recognition model provided by an embodiment of the application;
本申请目的的实现、功能特点及优点将结合实施例,参照附图做进一步说明。The realization, functional characteristics, and advantages of the purpose of this application will be further described in conjunction with the embodiments and with reference to the accompanying drawings.
应当理解,此处所描述的具体实施例仅仅用以解释本申请,并不用于限定本申请。It should be understood that the specific embodiments described here are only used to explain the present application, and are not used to limit the present application.
本申请实施例提供的识别模型的参数获取方法的执行主体包括但不限于服务端、终端等能够被配置为执行本申请实施例提供的该方法的电子设备中的至少一种。换言之,所述识别模型的参数获取方法可以由安装在终端设备或服务端设备的软件或硬件来执行,所述软件可以是区块链平台。所述服务端包括但不限于:单台服务器、服务器集群、云端服务器或云端服务器集群等。The execution subject of the method for acquiring parameters of the recognition model provided in the embodiment of the present application includes, but is not limited to, at least one of the electronic devices that can be configured to execute the method provided in the embodiment of the present application, such as a server and a terminal. In other words, the method for acquiring the parameters of the recognition model may be executed by software or hardware installed on the terminal device or the server device, and the software may be a blockchain platform. The server includes but is not limited to: a single server, a server cluster, a cloud server or a cloud server cluster, etc.
本申请提供一种识别模型的参数获取方法。参照图1所示,为本申请一实施例提供的识别模型的参数获取方法的流程示意图。该方法可以由一个装置执行,该装置可以由软件和/或硬件实现。This application provides a method for acquiring parameters of a recognition model. Referring to FIG. 1, it is a schematic flowchart of a method for acquiring parameters of a recognition model provided by an embodiment of this application. The method can be executed by a device, and the device can be implemented by software and/or hardware.
在本实施例中,识别模型的参数获取方法包括:In this embodiment, the method for acquiring parameters of the recognition model includes:
S1、获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集。S1. Obtain a training data set containing noise labels, and perform data standardization processing on the training data set to obtain a standard data set.
本申请实施例中,所述包含噪声标签的训练数据集是指训练数据集中存在一些数据,但数据的预设标准标签与数据的内容不对应,即预设标准标签是噪声标签的数据。In the embodiment of the present application, the training data set containing noise labels means that there is some data in the training data set, but the preset standard label of the data does not correspond to the content of the data, that is, the preset standard label is data of the noise label.
本申请实施例可利用具有数据抓取功能的python语句从区块链节点中获取所述训练数据集,也可从数据库中获取所述训练数据集。In the embodiment of the present application, the training data set can be obtained from the blockchain node by using a python sentence with a data capture function, and the training data set can also be obtained from a database.
较佳地,所述训练数据集存储于区块链的不同节点中,利用区块链的高数据吞吐性,可提高获取所述训练数据集的效率。Preferably, the training data set is stored in different nodes of the blockchain, and the high data throughput of the blockchain can be used to improve the efficiency of obtaining the training data set.
具体地,所述对所述训练数据集进行数据标准化处理,包括下述一种或几种的组合:Specifically, the performing data standardization processing on the training data set includes one or a combination of the following:
去除所述训练数据集中的唯一属性值;Removing the unique attribute value in the training data set;
对所述训练数据集进行缺失值填充;对所述训练数据集进行数据归一化。Filling the training data set with missing values; and performing data normalization on the training data set.
详细地,所述唯一属性值包括但不限于:数据id,数据编号。In detail, the unique attribute value includes but is not limited to: data id and data number.
由于唯一属性值并不能刻画数据的分布规律,反而会增加数据的内容,使得对数据进行处理时需要占用更多额外的计算资源,降低数据处理的效率,因此,本申请实施例中将训练数据集中的唯一数据进行删除,提高后续数据处理的效率。Since the unique attribute value cannot describe the distribution law of the data, it will increase the content of the data, so that more additional computing resources are needed to process the data, and the efficiency of data processing is reduced. Therefore, the training data is used in the embodiment of the application. The centralized and unique data is deleted to improve the efficiency of subsequent data processing.
较佳地,本申请实施例利用高维映射的方法,将所述训练数据集中的数据映射至预构建的高维空间,再将缺失的数据利用独热编码技术进行数据填充。利用高位空间的多维度性,可提高对所述训练数据集中缺失数据查找的效率,利用独热编码技术可提高数据填充的准确率。Preferably, the embodiment of the present application uses a high-dimensional mapping method to map the data in the training data set to a pre-built high-dimensional space, and then use the one-hot encoding technique to fill in the missing data. Utilizing the multi-dimensionality of high-bit space can improve the efficiency of searching for missing data in the training data set, and using one-hot encoding technology can improve the accuracy of data filling.
具体地,本申请实施例利用如下标准化算法对所述训练数据集进行数据归一化:Specifically, the embodiment of the present application uses the following standardized algorithm to perform data normalization on the training data set:
其中,x为数据归一化后的标准数据,S
old为所述训练数据集中的数据,S
max为S
old取值的最大值,S
min为S
old取值的最小值。
Wherein, x is the standard data of data normalization, S old data as the training data set, S max is the maximum value of S old, S min is the minimum value of the value S old.
需要强调的是,S
max和S
min是预先设定的,用于限定所述训练数据集中数据的范围。
It should be emphasized that S max and S min are preset and used to limit the range of data in the training data set.
当完成所述数据标准化处理后,得到所述标准数据集。After completing the data standardization process, the standard data set is obtained.
S2、基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型。S2. Establish a recognition model based on a multi-layer deep neural network, and use the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters.
本申请实施例中,所述多层深度神经网络为:In the embodiment of the present application, the multi-layer deep neural network is:
h=(h
(n)·h
(n-1)·…·h
(1))
h=(h (n) ·h (n-1) ·…·h (1) )
其中,h
(n)表示所述多层深度神经网络第n层的网络结构。
Wherein, h (n) represents the network structure of the nth layer of the multilayer deep neural network.
当所述多层深度神经网络经过softmax函数进行激活后,可输出标准数据x与x对应的预设标准标签y的联合分布p(x,y)的预测值
以及得到对标准数据集中标准数据的预测标签。
After the multi-layer deep neural network is activated by the softmax function, it can output the predicted value of the joint distribution p(x, y) of the preset standard label y corresponding to the standard data x and x And get the predicted label of the standard data in the standard data set.
所述softmax函数是一种激活函数,用于将多层深度神经网络的输出结果变换成预设形式,本申请实施例中,将所述多层深度神经网络的输出结果变换成概率形式(即
)。变换成概率形式后的所述多层深度神经网络的输出结果可直观的看出所述预设标准标签与所述预测标签的差异,根据所述差异对模型参数进行调整,有利于提高模型的训练效率。
The softmax function is an activation function for transforming the output result of the multi-layer deep neural network into a preset form. In the embodiment of the present application, the output result of the multi-layer deep neural network is transformed into a probabilistic form (ie ). The output result of the multi-layer deep neural network transformed into a probabilistic form can intuitively see the difference between the preset standard label and the predicted label, and the model parameters are adjusted according to the difference, which is beneficial to improve the model’s performance. Training efficiency.
具体地,本申请实施例将所述标准数据集输入至所述识别模型,利用所述标准数据集对所述识别模型进行训练,得到所述识别模型的初始化参数,确定包含初始化参数识别模型为标准识别模型。Specifically, in this embodiment of the application, the standard data set is input to the recognition model, and the recognition model is trained using the standard data set to obtain the initialization parameters of the recognition model, and it is determined that the recognition model including the initialization parameters is Standard recognition model.
进一步地,本申请实施例中,在基于多层深度神经网络建立识别模型之前,所述方法还包括:Further, in the embodiment of the present application, before establishing a recognition model based on a multi-layer deep neural network, the method further includes:
构建特征空间:
其中,所述特征空间用于存储标准数据集;
Construct feature space: Wherein, the feature space is used to store standard data sets;
构建所述特征空间对应的标签空间:
y={e
i:i∈[c]},其中,e为标准数据集中标准数据的预设标准标签,[c]={1…c},为任意的c个正整数,其数量与标准数据集中数据数目相同,所述标签空间用于存储所述特征空间中标准数据对应的预设标准标签。
Construct a label space corresponding to the feature space: y={e i :i∈[c]}, where e is the preset standard label of the standard data in the standard data set, [c]={1...c}, which is any c positive integers, the number of which corresponds to the standard The number of data in the data set is the same, and the label space is used to store preset standard labels corresponding to standard data in the feature space.
另外,对于所述特征空间中存储的标准数据x与其在标签空间中对应的预设标准标签y的联合分布为p(x,y):In addition, the joint distribution of the standard data x stored in the feature space and the corresponding preset standard label y in the label space is p(x, y):
p(x,y)=p(y|x)p(x)p(x,y)=p(y|x)p(x)
其中,p(x)为标准数据集中任一标准数据x在所述特征空间出现的频率,p(y|x)为标准数据x出现时其预设标准标签在所述标签空间的频率。Wherein, p(x) is the frequency of any standard data x in the standard data set in the feature space, and p(y|x) is the frequency of the preset standard label in the label space when the standard data x appears.
本实施例中,构建所述特征空间和所述标签空间并计算标准数据x与其在标签空间中对应的标签y的联合分布为p(x,y),可将标准数据集中的标准数据与标准数据对应的标签之间的关系更好的展现出来,提高数据处理的效率。In this embodiment, the feature space and the label space are constructed, and the joint distribution of the standard data x and its corresponding label y in the label space is calculated as p(x, y). The standard data in the standard data set can be compared with the standard data. The relationship between the tags corresponding to the data is better displayed, and the efficiency of data processing is improved.
S3、构建所述标准数据集的噪声概率转移矩阵。S3. Construct a noise probability transition matrix of the standard data set.
本申请实施例中,所述标准数据集的噪声概率转移矩阵可表示为:In the embodiment of the present application, the noise probability transition matrix of the standard data set can be expressed as:
Q∈[0,1]
c×c
Q∈[0,1] c×c
其中,c的大小与标准数据集中标准数据的数量相同。Among them, the size of c is the same as the number of standard data in the standard data set.
所述噪声概率转移矩阵表示噪声标签在数据中的分布。The noise probability transition matrix represents the distribution of noise tags in the data.
具体的,所述噪声概率转移矩阵Q中第i行、第j列的元素表示出现噪声标签的概率。Specifically, the element in the i-th row and j-th column in the noise probability transition matrix Q represents the probability of the occurrence of a noise label.
详细地,本申请实施例中,所述标准数据集的噪声概率转移矩阵如下:In detail, in the embodiment of the present application, the noise probability transition matrix of the standard data set is as follows:
其中,Q为所述噪声概率转移矩阵,α为所述标准数据集中任一标准数据,β
i为α对应的预设标准标签,
为标准识别模型对α生成的预测标签,β
j为α的噪声标签。
Where Q is the noise probability transition matrix, α is any standard data in the standard data set, and β i is a preset standard label corresponding to α, Is the predicted label generated by the standard recognition model for α, and β j is the noise label of α.
S4、基于所述噪声概率转移矩阵构建损失函数。S4. Construct a loss function based on the noise probability transition matrix.
本申请实施例中,所述损失函数包括但不限于:后向损失函数和前项损失函数。In the embodiment of the present application, the loss function includes but is not limited to: a backward loss function and an preceding loss function.
具体地,所述前向损失函数为:Specifically, the forward loss function is:
其中,QT为所述噪声概率转移矩阵的转置矩阵,ψ为所述识别模型的误差因子,h为所述多层深度神经网络,
为所述前向损失函数的损失值。
Wherein, QT is the transposed matrix of the noise probability transition matrix, ψ is the error factor of the recognition model, h is the multilayer deep neural network, Is the loss value of the forward loss function.
具体地,所述后项损失函数为:Specifically, the latter loss function is:
其中,l
→(h)为所述后向损失函数的损失值,y为所述标准数据集中任一标准数据x的预设标准标签,
为所述标准识别模型对x的预测标签,Q为所述噪声概率转移矩阵,p(x,y)为标准数据x与x对应的预设标准标签y的联合分布,
为p(x,y)的预测值。
Where l → (h) is the loss value of the backward loss function, and y is the preset standard label of any standard data x in the standard data set, Is the predicted label of x by the standard recognition model, Q is the noise probability transition matrix, and p(x, y) is the joint distribution of standard data x and the preset standard label y corresponding to x, Is the predicted value of p(x,y).
所述后向损失函数用于计算所述标准数据x在标签空间中对应的标签为噪声标签的概率值,即所述标准数据x的预设标准标签出现错误的可能性。The backward loss function is used to calculate the probability value that the label corresponding to the standard data x in the label space is a noisy label, that is, the probability that the preset standard label of the standard data x is wrong.
S5、利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。S5. Calculate the update parameters of the standard recognition model by using the loss function, and replace the update parameters with the initialization parameters.
本申请实施例中,所述利用所述损失函数计算所述标准识别模型的更新参数,包括:In the embodiment of the present application, the calculation of the update parameters of the standard recognition model by using the loss function includes:
获取所述标准数据集中标准数据的预设标准标签,以及所述标准识别模型对所述标准数据集中标准数据的预测标签;Acquiring the preset standard label of the standard data in the standard data set, and the prediction label of the standard data in the standard data set by the standard recognition model;
利用损失函数计算所述预测标签与所述标准标签之间的差异值;Calculating a difference value between the predicted label and the standard label by using a loss function;
当所述差异值在预设阈值区间内时,利用梯度下降算法计算所述标准识别模型的更新参数;When the difference value is within a preset threshold interval, use a gradient descent algorithm to calculate the update parameters of the standard recognition model;
当所述差异值大于所述阈值区间的上限时,利用所述损失函数计算所述标准标签为噪声标签的概率值;When the difference value is greater than the upper limit of the threshold interval, use the loss function to calculate the probability value that the standard label is a noise label;
当所述概率值小于预设概率阈值时,利用梯度下降算法计算所述标准识别模型的更新 参数。When the probability value is less than the preset probability threshold, the gradient descent algorithm is used to calculate the update parameters of the standard recognition model.
本申请实施例中,所述损失函数为前向损失函数和/或后向损失函数。In the embodiment of the present application, the loss function is a forward loss function and/or a backward loss function.
本申请实施例中,当所述差异值在预设阈值区间内时,说明标准识别模型的识别结果出现错误,则利用梯度下降算法对所述标准识别模型的参数进行更新,以提高所述标准识别模型的精确度。In the embodiment of the present application, when the difference value is within the preset threshold interval, it means that the recognition result of the standard recognition model is wrong. Then the gradient descent algorithm is used to update the parameters of the standard recognition model to improve the standard. Identify the accuracy of the model.
本实施例中,所述梯度下降算法包括但不限于批量梯度下降算法、随机梯度下降算法和小批量梯度下降算法。In this embodiment, the gradient descent algorithm includes, but is not limited to, a batch gradient descent algorithm, a stochastic gradient descent algorithm, and a small batch gradient descent algorithm.
当所述差异值大于所述阈值区间的上限时,可能并不是由于标准识别模型的识别错误导致差异值大于所述阈值区间的上限。实际应用中,由于所述噪音标签的存在,标准数据的预设标准标签出现错误,也会导致预设标准标签和预测标签的差异值大于阈值区间的上限。因此,当所述差异值大于所述阈值区间的上限时,本申请实施例利用损失函数计算标准数据的预设标准标签为噪声标签的概率值,当所述概率值小于预设概率阈值时,说明所述标准识别模型的识别结果出现错误,则利用梯度下降算法计算所述标准识别模型的更新参数。When the difference value is greater than the upper limit of the threshold interval, it may not be due to a recognition error of the standard recognition model that the difference value is greater than the upper limit of the threshold interval. In practical applications, due to the existence of the noise label, the preset standard label of the standard data is wrong, which will also cause the difference between the preset standard label and the predicted label to be greater than the upper limit of the threshold interval. Therefore, when the difference value is greater than the upper limit of the threshold interval, the embodiment of the present application uses a loss function to calculate the probability value that the preset standard label of the standard data is a noise label, and when the probability value is less than the preset probability threshold, It indicates that the recognition result of the standard recognition model is wrong, and then the gradient descent algorithm is used to calculate the update parameters of the standard recognition model.
进一步地,当所述概率值大于或等于所述概率阈值时,本申请实施例对所述标准数据的预设标准标签进行修正。Further, when the probability value is greater than or equal to the probability threshold value, the embodiment of the present application corrects the preset standard label of the standard data.
进一步地,本申请实施例利用所述更新参数对所述初始化参数进行替换,初始化参数被替换后即可得到最终的识别模型,最终的识别模型可用于对输入数据进行识别,所述输入数据包括但不限于图像数据。Further, the embodiment of the present application uses the update parameters to replace the initialization parameters. After the initialization parameters are replaced, the final recognition model can be obtained. The final recognition model can be used to recognize input data, and the input data includes But it is not limited to image data.
本申请实施例通过获取包含噪声标签的训练数据集之后,对训练数据集进行标准化处理,提高对训练数据进行处理的效率;在得到包含初始化参数的标准识别模型之后,构建所述标准数据集的噪声概率转移矩阵,利于提高根据所述噪声转移矩阵构建出的损失函数对模型的适用性,以便于后续利用损失函数训练出更加精确的模型参数;基于所述噪声概率转移矩阵构建损失函数,利用损失函数计算标准识别模型的更新参数,从而能够得到更准确的模型参数,实现提高获取模型参数的精确性的目的。因此本申请提出的识别模型的参数获取方法,可以提供一种提高获取模型参数的精确度的方法。In the embodiment of the present application, after obtaining the training data set containing the noise label, the training data set is standardized to improve the efficiency of processing the training data; after obtaining the standard recognition model containing the initialization parameters, the standard data set is constructed The noise probability transition matrix is beneficial to improve the applicability of the loss function constructed from the noise transition matrix to the model, so that the loss function can be used to train more accurate model parameters; the loss function is constructed based on the noise probability transition matrix, and The loss function calculates the updated parameters of the standard recognition model, so that more accurate model parameters can be obtained, and the purpose of improving the accuracy of obtaining the model parameters is achieved. Therefore, the method for acquiring parameters of a recognition model proposed in this application can provide a method for improving the accuracy of acquiring model parameters.
如图2所示,是本申请识别模型的参数获取装置的模块示意图。As shown in Figure 2, it is a schematic diagram of the module of the parameter acquisition device of the identification model of the present application.
本申请所述识别模型的参数获取装置100可以安装于电子设备中。根据实现的功能,所述识别模型的参数获取装置可以包括训练数据获取模块101、识别模型构建模块102、转移矩阵构建模块103、损失函数构建模块104和模型参数更新模块105。本发所述模块也可以称之为单元,是指一种能够被电子设备处理器所执行,并且能够完成固定功能的一系列计算机程序段,其存储在电子设备的存储器中。The device 100 for acquiring the parameters of the recognition model described in this application can be installed in an electronic device. According to the realized function, the parameter acquisition device of the recognition model may include a training data acquisition module 101, a recognition model construction module 102, a transition matrix construction module 103, a loss function construction module 104, and a model parameter update module 105. The module described in the present invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can complete fixed functions, and are stored in the memory of the electronic device.
在本实施例中,关于各模块/单元的功能如下:In this embodiment, the functions of each module/unit are as follows:
所述训练数据获取模块101,用于获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;The training data acquisition module 101 is configured to acquire a training data set containing noise labels, and perform data standardization processing on the training data set to obtain a standard data set;
所述识别模型构建模块102,用于基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;The recognition model construction module 102 is configured to establish a recognition model based on a multi-layer deep neural network, and use the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;
所述转移矩阵构建模块103,用于构建所述标准数据集的噪声概率转移矩阵;The transition matrix construction module 103 is used to construct the noise probability transition matrix of the standard data set;
所述损失函数构建模块104,用于基于所述噪声概率转移矩阵构建损失函数;The loss function construction module 104 is configured to construct a loss function based on the noise probability transition matrix;
所述模型参数更新模块105,用于利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The model parameter update module 105 is configured to use the loss function to calculate update parameters of the standard recognition model, and replace the update parameters with the initialization parameters.
详细地,所述识别模型的参数获取装置各模块的具体实施方式如下:In detail, the specific implementation of each module of the device for acquiring parameters of the recognition model is as follows:
所述训练数据获取模块101,用于获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集。The training data acquisition module 101 is configured to acquire a training data set containing noise labels, and perform data standardization processing on the training data set to obtain a standard data set.
本申请实施例中,所述包含噪声标签的训练数据集是指训练数据集中存在一些数据,但数据的预设标准标签与数据的内容不对应,即预设标准标签是噪声标签的数据。In the embodiment of the present application, the training data set containing noise labels means that there is some data in the training data set, but the preset standard label of the data does not correspond to the content of the data, that is, the preset standard label is data of the noise label.
本申请实施例可利用具有数据抓取功能的python语句从区块链节点中获取所述训练数据集,也可从数据库中获取所述训练数据集。In the embodiment of the present application, the training data set can be obtained from the blockchain node by using a python sentence with a data capture function, and the training data set can also be obtained from a database.
较佳地,所述训练数据集存储于区块链的不同节点中,利用区块链的高数据吞吐性,可提高获取所述训练数据集的效率。Preferably, the training data set is stored in different nodes of the blockchain, and the high data throughput of the blockchain can be used to improve the efficiency of obtaining the training data set.
具体地,所述训练数据获取模块101对所述训练数据集进行数据标准化处理,包括下述一种或几种的组合:Specifically, the training data acquisition module 101 performs data standardization processing on the training data set, including one or a combination of the following:
去除所述训练数据集中的唯一属性值;Removing the unique attribute value in the training data set;
对所述训练数据集进行缺失值填充;Filling in missing values on the training data set;
对所述训练数据集进行数据归一化。Perform data normalization on the training data set.
详细地,所述唯一属性值包括但不限于:数据id,数据编号。In detail, the unique attribute value includes but is not limited to: data id and data number.
由于唯一属性值并不能刻画数据的分布规律,反而会增加数据的内容,使得对数据进行处理时需要占用更多额外的计算资源,降低数据处理的效率,因此,本申请实施例中将训练数据集中的唯一数据进行删除,提高后续数据处理的效率。Since the unique attribute value cannot describe the distribution law of the data, it will increase the content of the data, so that more additional computing resources are needed to process the data, and the efficiency of data processing is reduced. Therefore, the training data is used in the embodiment of the application. The centralized and unique data is deleted to improve the efficiency of subsequent data processing.
较佳地,本申请实施例利用高维映射将所述训练数据集中的数据映射至预构建的高维空间,再将缺失的数据利用独热编码技术进行数据填充。利用高位空间的多维度性,可提高对所述训练数据集中缺失数据查找的效率,利用独热编码技术可提高数据填充的准确率。Preferably, in the embodiment of the present application, high-dimensional mapping is used to map the data in the training data set to a pre-built high-dimensional space, and then the missing data is filled with data using a one-hot encoding technique. Utilizing the multi-dimensionality of high-bit space can improve the efficiency of searching for missing data in the training data set, and using one-hot encoding technology can improve the accuracy of data filling.
具体地,本申请采用如下标准化算法对所述训练数据集进行数据归一化:Specifically, this application uses the following standardized algorithm to normalize the training data set:
其中,x为数据归一化后的标准数据,S
old为所述训练数据集中的数据,S
max为S
old取值的最大值,S
min为S
old取值的最小值。
Wherein, x is the standard data of data normalization, S old data as the training data set, S max is the maximum value of S old, S min is the minimum value of the value S old.
需要强调的是,S
max和S
min是预先设定的,用于限定所述训练数据集中数据的范围。
It should be emphasized that S max and S min are preset and used to limit the range of data in the training data set.
当完成所述数据标准化处理后,得到所述标准数据集。After completing the data standardization process, the standard data set is obtained.
所述识别模型构建模块102,用于基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型。The recognition model construction module 102 is configured to establish a recognition model based on a multi-layer deep neural network, and use the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters.
本申请实施例中,所述多层深度神经网络为:In the embodiment of the present application, the multi-layer deep neural network is:
h=(h
(n)·h
(n-1)·…·h
(1))
h=(h (n) ·h (n-1) ·…·h (1) )
其中,h
(n)表示所述多层深度神经网络第n层的网络结构。
Wherein, h (n) represents the network structure of the nth layer of the multilayer deep neural network.
当所述多层深度神经网络经过softmax函数进行激活后,可输出标准数据x与x对应的预设标准标签y的联合分布p(x,y)的预测值
以及得到对标准数据集中标准数据的预测标签。
After the multi-layer deep neural network is activated by the softmax function, it can output the predicted value of the joint distribution p(x,y) of the preset standard label y corresponding to the standard data x and x And get the predicted label of the standard data in the standard data set.
所述softmax函数是一种激活函数,用于将多层深度神经网络的输出结果变换成预设形式,本申请实施例中,将所述多层深度神经网络的输出结果变换成概率形式(即
)。变换成概率形式后的所述多层深度神经网络的输出结果可直观的看出所述预设标准标签与所述预测标签的差异,根据所述差异对模型参数进行调整,有利于提高模型的训练效率。
The softmax function is an activation function for transforming the output result of the multi-layer deep neural network into a preset form. In the embodiment of the present application, the output result of the multi-layer deep neural network is transformed into a probabilistic form (ie ). The output result of the multi-layer deep neural network transformed into a probabilistic form can intuitively see the difference between the preset standard label and the predicted label, and the model parameters are adjusted according to the difference, which is beneficial to improve the model’s performance. Training efficiency.
具体地,本申请实施例将所述标准数据集输入至所述识别模型,利用所述标准数据集对所述识别模型进行训练,得到所述识别模型的初始化参数,确定包含初始化参数识别模型为标准识别模型。Specifically, in this embodiment of the application, the standard data set is input to the recognition model, and the recognition model is trained using the standard data set to obtain the initialization parameters of the recognition model, and it is determined that the recognition model including the initialization parameters is Standard recognition model.
进一步地,本申请实施例中,在基于多层深度神经网络建立识别模型之前,还包括:Further, in the embodiment of the present application, before establishing a recognition model based on a multi-layer deep neural network, the method further includes:
构建特征空间:
其中,所述特征空间用于存储标准数据集;
Construct feature space: Wherein, the feature space is used to store standard data sets;
构建所述特征空间对应的标签空间:
y={e
i:i∈[c]},其中,e为标准数据集中标准数据的预设标准标签,[c]={1…c},为任意的c个正整数,其数量与标准数据集中数据数目相同,所述标签空间用于存储所述特征空间中标准数据对应的预设标准标签。
Construct a label space corresponding to the feature space: y={e i :i∈[c]}, where e is the preset standard label of the standard data in the standard data set, [c]={1...c}, which is any c positive integers, the number of which corresponds to the standard The number of data in the data set is the same, and the label space is used to store preset standard labels corresponding to standard data in the feature space.
另外,对于所述特征空间中存储的标准数据x与其在标签空间中对应的预设标准标签y的联合分布为p(x,y):In addition, the joint distribution of the standard data x stored in the feature space and the corresponding preset standard label y in the label space is p(x, y):
p(x,y)=p(y|x)p(x)p(x,y)=p(y|x)p(x)
其中,p(x)为标准数据集中任一标准数据x在所述特征空间出现的频率,p(y|x)为标准数据x出现时其预设标准标签在所述标签空间的频率。Wherein, p(x) is the frequency of any standard data x in the standard data set in the feature space, and p(y|x) is the frequency of the preset standard label in the label space when the standard data x appears.
本实施例中,构建所述特征空间和所述标签空间并计算标准数据x与其在标签空间中对应的标签y的联合分布为p(x,y),可将标准数据集中的标准数据与标准数据对应的标签之间的关系更好的展现出来,提高数据处理的效率。In this embodiment, the feature space and the label space are constructed, and the joint distribution of the standard data x and its corresponding label y in the label space is calculated as p(x, y). The standard data in the standard data set can be compared with the standard data. The relationship between the tags corresponding to the data is better displayed, and the efficiency of data processing is improved.
所述转移矩阵构建模块103,用于构建所述标准数据集的噪声概率转移矩阵。The transition matrix construction module 103 is used to construct the noise probability transition matrix of the standard data set.
本申请实施例中,所述标准数据集的噪声概率转移矩阵可表示为:In the embodiment of the present application, the noise probability transition matrix of the standard data set can be expressed as:
Q∈[0,1]
c×c
Q∈[0,1] c×c
其中,c的大小与标准数据集中标准数据的数量相同。Among them, the size of c is the same as the number of standard data in the standard data set.
所述噪声概率转移矩阵表示噪声标签在数据中的分布。The noise probability transition matrix represents the distribution of noise tags in the data.
具体的,所述噪声概率转移矩阵Q中第i行、第j列的元素表示出现噪声标签的概率。Specifically, the element in the i-th row and j-th column in the noise probability transition matrix Q represents the probability of the occurrence of a noise label.
详细地,本申请实施例中,所述标准数据集的噪声概率转移矩阵如下:In detail, in the embodiment of the present application, the noise probability transition matrix of the standard data set is as follows:
其中,Q为所述噪声概率转移矩阵,α为所述标准数据集中任一标准数据,β
i为α对应的预设标准标签,
为标准识别模型对α生成的预测标签,β
j为α的噪声标签。
Where Q is the noise probability transition matrix, α is any standard data in the standard data set, and β i is a preset standard label corresponding to α, Is the predicted label generated by the standard recognition model for α, and β j is the noise label of α.
所述损失函数构建模块104,用于基于所述噪声概率转移矩阵构建损失函数。The loss function construction module 104 is configured to construct a loss function based on the noise probability transition matrix.
本申请实施例中,所述损失函数包括但不限于:后向损失函数和前项损失函数。In the embodiment of the present application, the loss function includes but is not limited to: a backward loss function and an preceding loss function.
具体地,所述前向损失函数为:Specifically, the forward loss function is:
其中,QT为所述噪声概率转移矩阵的转置矩阵,ψ为所述识别模型的误差因子,h为所述多层深度神经网络,
为所述前向损失函数的损失值。
Wherein, QT is the transposed matrix of the noise probability transition matrix, ψ is the error factor of the recognition model, h is the multilayer deep neural network, Is the loss value of the forward loss function.
具体地,所述后项损失函数为:Specifically, the latter loss function is:
其中,l
→(h)为所述后向损失函数的损失值,y为所述标准数据集中任一标准数据x的预设标准标签,
为所述标准识别模型对x的预测标签,Q为所述噪声概率转移矩阵,p(x,y)为标准数据x与x对应的预设标准标签y的联合分布,
为p(x,y)的预测值。
Where l → (h) is the loss value of the backward loss function, and y is the preset standard label of any standard data x in the standard data set, Is the predicted label of x by the standard recognition model, Q is the noise probability transition matrix, and p(x, y) is the joint distribution of standard data x and the preset standard label y corresponding to x, Is the predicted value of p(x,y).
所述后向损失函数用于计算所述标准数据x在标签空间中对应的标签为噪声标签的概率值,即所述标准数据x的预设标准标签出现错误的可能性。The backward loss function is used to calculate the probability value that the label corresponding to the standard data x in the label space is a noisy label, that is, the probability that the preset standard label of the standard data x is wrong.
所述模型参数更新模块105,用于利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The model parameter update module 105 is configured to use the loss function to calculate update parameters of the standard recognition model, and replace the update parameters with the initialization parameters.
本申请实施例中,所述模型参数更新模块105利用所述损失函数计算所述标准识别模型的更新参数,包括:In the embodiment of the present application, the model parameter update module 105 uses the loss function to calculate the update parameters of the standard recognition model, including:
获取所述标准数据集中标准数据的预设标准标签,以及所述标准识别模型对所述标准数据集中标准数据的预测标签;Acquiring the preset standard label of the standard data in the standard data set, and the prediction label of the standard data in the standard data set by the standard recognition model;
利用损失函数计算所述预测标签与所述标准标签之间的差异值;Calculating a difference value between the predicted label and the standard label by using a loss function;
当所述差异值在预设阈值区间内时,利用梯度下降算法计算所述标准识别模型的更新参数;When the difference value is within a preset threshold interval, use a gradient descent algorithm to calculate the update parameters of the standard recognition model;
当所述差异值大于所述阈值区间的上限时,利用所述损失函数计算所述标准标签为噪声标签的概率值;When the difference value is greater than the upper limit of the threshold interval, use the loss function to calculate the probability value that the standard label is a noise label;
当所述概率值小于预设概率阈值时,利用梯度下降算法计算所述标准识别模型的更新参数。When the probability value is less than the preset probability threshold, a gradient descent algorithm is used to calculate the update parameters of the standard recognition model.
本申请实施例中,所述损失函数为前向损失函数和/或后向损失函数。In the embodiment of the present application, the loss function is a forward loss function and/or a backward loss function.
本申请实施例中,当所述差异值在预设阈值区间内时,说明标准识别模型的识别结果出现错误,则利用梯度下降算法对所述标准识别模型的参数进行更新,以提高所述标准识别模型的精确度。In the embodiment of the present application, when the difference value is within the preset threshold interval, it means that the recognition result of the standard recognition model is wrong. Then the gradient descent algorithm is used to update the parameters of the standard recognition model to improve the standard. Identify the accuracy of the model.
本实施例中,所述梯度下降算法包括但不限于批量梯度下降算法、随机梯度下降算法和小批量梯度下降算法。In this embodiment, the gradient descent algorithm includes, but is not limited to, a batch gradient descent algorithm, a stochastic gradient descent algorithm, and a small batch gradient descent algorithm.
当所述差异值大于所述阈值区间的上限时,可能并不是由于标准识别模型的识别错误导致差异值大于所述阈值区间的上限。实际应用中,由于所述噪音标签的存在,标准数据的预设标准标签出现错误,也会导致预设标准标签和预测标签的差异值大于阈值区间的上限。因此,当所述差异值大于所述阈值区间的上限时,本申请实施例利用损失函数计算标准数据的预设标准标签为噪声标签的概率值,当所述概率值小于预设概率阈值时,说明所述标准识别模型的识别结果出现错误,则利用梯度下降算法计算所述标准识别模型的更新参数。When the difference value is greater than the upper limit of the threshold interval, it may not be due to a recognition error of the standard recognition model that the difference value is greater than the upper limit of the threshold interval. In practical applications, due to the existence of the noise label, the preset standard label of the standard data is wrong, which will also cause the difference between the preset standard label and the predicted label to be greater than the upper limit of the threshold interval. Therefore, when the difference value is greater than the upper limit of the threshold interval, the embodiment of the present application uses a loss function to calculate the probability value that the preset standard label of the standard data is a noise label, and when the probability value is less than the preset probability threshold, It indicates that the recognition result of the standard recognition model is wrong, and then the gradient descent algorithm is used to calculate the update parameters of the standard recognition model.
进一步地,当所述概率值大于或等于所述概率阈值时,本申请实施例对所述标准数据的预设标准标签进行修正。Further, when the probability value is greater than or equal to the probability threshold value, the embodiment of the present application corrects the preset standard label of the standard data.
进一步地,本申请实施例利用所述更新参数对所述初始化参数进行替换,初始化参数被替换后即可得到最终的识别模型,最终的识别模型可用于对输入数据进行识别,所述输入数据包括但不限于图像数据。Further, the embodiment of the present application uses the update parameters to replace the initialization parameters. After the initialization parameters are replaced, the final recognition model can be obtained. The final recognition model can be used to recognize input data, and the input data includes But it is not limited to image data.
本申请实施例通过获取包含噪声标签的训练数据集之后,对训练数据集进行标准化处理,提高对训练数据进行处理的效率;在得到包含初始化参数的标准识别模型之后,构建所述标准数据集的噪声概率转移矩阵,利于提高根据所述噪声转移矩阵构建出的损失函数对模型的适用性,以便于后续利用损失函数训练出更加精确的模型参数;基于所述噪声概率转移矩阵构建损失函数,利用损失函数计算标准识别模型的更新参数,从而能够得到更准确的模型参数,实现提高获取模型参数的精确性的目的。因此本申请提出的识别模型的参数获取装置,可以提供一种提高获取模型参数的精确度的方法。In the embodiment of the present application, after obtaining the training data set containing the noise label, the training data set is standardized to improve the efficiency of processing the training data; after obtaining the standard recognition model containing the initialization parameters, the standard data set is constructed The noise probability transition matrix is beneficial to improve the applicability of the loss function constructed from the noise transition matrix to the model, so that the loss function can be used to train more accurate model parameters; the loss function is constructed based on the noise probability transition matrix, and The loss function calculates the updated parameters of the standard recognition model, so that more accurate model parameters can be obtained, and the purpose of improving the accuracy of obtaining the model parameters is achieved. Therefore, the device for acquiring parameters of a recognition model proposed in this application can provide a method for improving the accuracy of acquiring model parameters.
如图3所示,是本申请实现识别模型的参数获取方法的电子设备的结构示意图。As shown in FIG. 3, it is a schematic diagram of the structure of an electronic device that implements the method for acquiring parameters of a recognition model according to the present application.
所述电子设备1可以包括处理器10、存储器11和总线,还可以包括存储在所述存储器11中并可在所述处理器10上运行的计算机程序,如识别模型的参数获取程序12。The electronic device 1 may include a processor 10, a memory 11, and a bus, and may also include a computer program stored in the memory 11 and running on the processor 10, such as a parameter acquisition program 12 of an identification model.
其中,所述存储器11至少包括一种类型的可读存储介质,所述可读存储介质可以是易失性的,也可以是非易失性的。具体的,所述可读存储介质包括闪存、移动硬盘、多媒体卡、卡型存储器(例如:SD或DX存储器等)、磁性存储器、磁盘、光盘等。所述存储器11在一些实施例中可以是电子设备1的内部存储单元,例如该电子设备1的移动硬盘。所述存储器11在另一些实施例中也可以是电子设备1的外部存储设备,例如电子设备1上配 备的插接式移动硬盘、智能存储卡(SmartMediaCard,SMC)、安全数字(SecureDigital,SD)卡、闪存卡(FlashCard)等。进一步地,所述存储器11还可以既包括电子设备1的内部存储单元也包括外部存储设备。所述存储器11不仅可以用于存储安装于电子设备1的应用软件及各类数据,例如识别模型的参数获取程序12的代码等,还可以用于暂时地存储已经输出或者将要输出的数据。The memory 11 includes at least one type of readable storage medium, and the readable storage medium may be volatile or nonvolatile. Specifically, the readable storage medium includes flash memory, mobile hard disk, multimedia card, card-type memory (for example: SD or DX memory, etc.), magnetic memory, magnetic disk, optical disk, etc. The memory 11 may be an internal storage unit of the electronic device 1 in some embodiments, for example, a mobile hard disk of the electronic device 1. In other embodiments, the memory 11 may also be an external storage device of the electronic device 1, such as a plug-in mobile hard disk, a smart memory card (SmartMediaCard, SMC), and a secure digital (SecureDigital, SD) equipped on the electronic device 1. Card, flash card (FlashCard), etc. Further, the memory 11 may also include both an internal storage unit of the electronic device 1 and an external storage device. The memory 11 can be used not only to store application software and various data installed in the electronic device 1, such as the code of the parameter acquisition program 12 of the identification model, etc., but also to temporarily store data that has been output or will be output.
所述处理器10在一些实施例中可以由集成电路组成,例如可以由单个封装的集成电路所组成,也可以是由多个相同功能或不同功能封装的集成电路所组成,包括一个或者多个中央处理器(CentralProcessingunit,CPU)、微处理器、数字处理芯片、图形处理器及各种控制芯片的组合等。所述处理器10是所述电子设备的控制核心(ControlUnit),利用各种接口和线路连接整个电子设备的各个部件,通过运行或执行存储在所述存储器11内的程序或者模块(例如执行识别模型的参数获取程序等),以及调用存储在所述存储器11内的数据,以执行电子设备1的各种功能和处理数据。The processor 10 may be composed of integrated circuits in some embodiments, for example, may be composed of a single packaged integrated circuit, or may be composed of multiple integrated circuits with the same function or different functions, including one or more Central Processing Unit (CPU), microprocessor, digital processing chip, graphics processor and a combination of various control chips, etc. The processor 10 is the control core (ControlUnit) of the electronic device, which uses various interfaces and lines to connect the various components of the entire electronic device, and runs or executes programs or modules stored in the memory 11 (for example, perform identification The parameter acquisition program of the model, etc.), and call the data stored in the memory 11 to execute various functions of the electronic device 1 and process data.
所述总线可以是外设部件互连标准(peripheralcomponentinterconnect,简称PCI)总线或扩展工业标准结构(extendedindustrystandardarchitecture,简称EISA)总线等。该总线可以分为地址总线、数据总线、控制总线等。所述总线被设置为实现所述存储器11以及至少一个处理器10等之间的连接通信。The bus may be a peripheral component interconnection standard (PCI) bus or an extended industry standard architecture (EISA) bus or the like. The bus can be divided into address bus, data bus, control bus and so on. The bus is configured to implement connection and communication between the memory 11 and at least one processor 10 and the like.
图3仅示出了具有部件的电子设备,本领域技术人员可以理解的是,图3示出的结构并不构成对所述电子设备1的限定,可以包括比图示更少或者更多的部件,或者组合某些部件,或者不同的部件布置。FIG. 3 only shows an electronic device with components. Those skilled in the art can understand that the structure shown in FIG. 3 does not constitute a limitation on the electronic device 1, and may include fewer or more components than shown in the figure. Components, or combinations of certain components, or different component arrangements.
例如,尽管未示出,所述电子设备1还可以包括给各个部件供电的电源(比如电池),优选地,电源可以通过电源管理装置与所述至少一个处理器10逻辑相连,从而通过电源管理装置实现充电管理、放电管理、以及功耗管理等功能。电源还可以包括一个或一个以上的直流或交流电源、再充电装置、电源故障检测电路、电源转换器或者逆变器、电源状态指示器等任意组件。所述电子设备1还可以包括多种传感器、蓝牙模块、Wi-Fi模块等,在此不再赘述。For example, although not shown, the electronic device 1 may also include a power source (such as a battery) for supplying power to various components. Preferably, the power source may be logically connected to the at least one processor 10 through a power management device, thereby controlling power The device implements functions such as charge management, discharge management, and power consumption management. The power supply may also include any components such as one or more DC or AC power supplies, recharging devices, power failure detection circuits, power converters or inverters, and power status indicators. The electronic device 1 may also include various sensors, Bluetooth modules, Wi-Fi modules, etc., which will not be repeated here.
进一步地,所述电子设备1还可以包括网络接口,可选地,所述网络接口可以包括有线接口和/或无线接口(如WI-FI接口、蓝牙接口等),通常用于在该电子设备1与其他电子设备之间建立通信连接。Further, the electronic device 1 may also include a network interface. Optionally, the network interface may include a wired interface and/or a wireless interface (such as a Wi-Fi interface, a Bluetooth interface, etc.), which is usually used in the electronic device 1 Establish a communication connection with other electronic devices.
可选地,该电子设备1还可以包括用户接口,用户接口可以是显示器(Display)、输入单元(比如键盘(Keyboard)),可选地,用户接口还可以是标准的有线接口、无线接口。可选地,在一些实施例中,显示器可以是LED显示器、液晶显示器、触控式液晶显示器以及OLED(OrganicLight-EmittingDiode,有机发光二极管)触摸器等。其中,显示器也可以适当的称为显示屏或显示单元,用于显示在电子设备1中处理的信息以及用于显示可视化的用户界面。Optionally, the electronic device 1 may also include a user interface. The user interface may be a display (Display) and an input unit (such as a keyboard (Keyboard)). Optionally, the user interface may also be a standard wired interface or a wireless interface. Optionally, in some embodiments, the display may be an LED display, a liquid crystal display, a touch liquid crystal display, an OLED (Organic Light-Emitting Diode, organic light emitting diode) touch device, etc. Among them, the display can also be appropriately called a display screen or a display unit, which is used to display the information processed in the electronic device 1 and to display a visualized user interface.
应该了解,所述实施例仅为说明之用,在专利申请范围上并不受此结构的限制。It should be understood that the embodiments are only for illustrative purposes, and are not limited by this structure in the scope of the patent application.
所述电子设备1中的所述存储器11存储的识别模型的参数获取程序12是多个指令的组合,在所述处理器10中运行时,可以实现:The parameter acquisition program 12 of the recognition model stored in the memory 11 in the electronic device 1 is a combination of multiple instructions. When running in the processor 10, it can realize:
获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;Acquiring a training data set containing noise labels, and performing data standardization processing on the training data set to obtain a standard data set;
基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;Establishing a recognition model based on a multi-layer deep neural network, and using the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;
构建所述标准数据集的噪声概率转移矩阵;Constructing the noise probability transition matrix of the standard data set;
基于所述噪声概率转移矩阵构建损失函数;Constructing a loss function based on the noise probability transition matrix;
利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
进一步地,所述电子设备1集成的模块/单元如果以软件功能单元的形式实现并作为独立的产品销售或使用时,可以存储在一个计算机可读取存储介质中。所述计算机可读存储介质可以是易失性的,也可以是非易失性的。具体的,所述计算机可读存储介质可以包括:能够携带所述计算机程序代码的任何实体或装置、记录介质、U盘、移动硬盘、磁碟、光盘、计算机存储器、只读存储器(ROM,Read-OnlyMemory)。Further, if the integrated module/unit of the electronic device 1 is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer readable storage medium. The computer-readable storage medium may be volatile or non-volatile. Specifically, the computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording medium, U disk, mobile hard disk, magnetic disk, optical disk, computer memory, read-only memory (ROM, Read -OnlyMemory).
进一步地,所述计算机可用存储介质可主要包括存储程序区和存储数据区,其中,存储程序区可存储操作***、至少一个功能所需的应用程序等;存储数据区可存储根据区块链节点的使用所创建的数据等。Further, the computer usable storage medium may mainly include a storage program area and a storage data area, wherein the storage program area may store an operating system, an application program required by at least one function, etc.; the storage data area may store a block chain node Use the created data, etc.
在本申请所提供的几个实施例中,应该理解到,所揭露的设备,装置和方法,可以通过其它的方式实现。例如,以上所描述的装置实施例仅仅是示意性的,例如,所述模块的划分,仅仅为一种逻辑功能划分,实际实现时可以有另外的划分方式。In the several embodiments provided in this application, it should be understood that the disclosed equipment, device, and method may be implemented in other ways. For example, the device embodiments described above are merely illustrative. For example, the division of the modules is only a logical function division, and there may be other division methods in actual implementation.
所述作为分离部件说明的模块可以是或者也可以不是物理上分开的,作为模块显示的部件可以是或者也可以不是物理单元,即可以位于一个地方,或者也可以分布到多个网络单元上。可以根据实际的需要选择其中的部分或者全部模块来实现本实施例方案的目的。The modules described as separate components may or may not be physically separated, and the components displayed as modules may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the objectives of the solutions of the embodiments.
另外,在本申请各个实施例中的各功能模块可以集成在一个处理单元中,也可以是各个单元单独物理存在,也可以两个或两个以上单元集成在一个单元中。上述集成的单元既可以采用硬件的形式实现,也可以采用硬件加软件功能模块的形式实现。In addition, the functional modules in the various embodiments of the present application may be integrated into one processing unit, or each unit may exist alone physically, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware, or may be implemented in the form of hardware plus software functional modules.
对于本领域技术人员而言,显然本申请不限于上述示范性实施例的细节,而且在不背离本申请的精神或基本特征的情况下,能够以其他的具体形式实现本申请。For those skilled in the art, it is obvious that the present application is not limited to the details of the foregoing exemplary embodiments, and the present application can be implemented in other specific forms without departing from the spirit or basic characteristics of the present application.
因此,无论从哪一点来看,均应将实施例看作是示范性的,而且是非限制性的,本申请的范围由所附权利要求而不是上述说明限定,因此旨在将落在权利要求的等同要件的含义和范围内的所有变化涵括在本申请内。不应将权利要求中的任何附关联图表记视为限制所涉及的权利要求。Therefore, no matter from which point of view, the embodiments should be regarded as exemplary and non-limiting. The scope of this application is defined by the appended claims rather than the above description, and therefore it is intended to fall into the claims. All changes in the meaning and scope of the equivalent elements of are included in this application. Any accompanying diagrams in the claims should not be regarded as limiting the claims involved.
本申请所指区块链是分布式数据存储、点对点传输、共识机制、加密算法等计算机技术的新型应用模式。区块链(Blockchain),本质上是一个去中心化的数据库,是一串使用密码学方法相关联产生的数据块,每一个数据块中包含了一批次网络交易的信息,用于验证其信息的有效性(防伪)和生成下一个区块。区块链可以包括区块链底层平台、平台产品服务层以及应用服务层等。The blockchain referred to in this application is a new application mode of computer technology such as distributed data storage, point-to-point transmission, consensus mechanism, and encryption algorithm. Blockchain, essentially a decentralized database, is a series of data blocks associated with cryptographic methods. Each data block contains a batch of network transaction information for verification. The validity of the information (anti-counterfeiting) and the generation of the next block. The blockchain can include the underlying platform of the blockchain, the platform product service layer, and the application service layer.
此外,显然“包括”一词不排除其他单元或步骤,单数不排除复数。***权利要求中陈述的多个单元或装置也可以由一个单元或装置通过软件或者硬件来实现。第二等词语用来表示名称,而并不表示任何特定的顺序。In addition, it is obvious that the word "including" does not exclude other units or steps, and the singular does not exclude the plural. Multiple units or devices stated in the system claims can also be implemented by one unit or device through software or hardware. The second class words are used to indicate names, and do not indicate any specific order.
最后应说明的是,以上实施例仅用以说明本申请的技术方案而非限制,尽管参照较佳实施例对本申请进行了详细说明,本领域的普通技术人员应当理解,可以对本申请的技术方案进行修改或等同替换,而不脱离本申请技术方案的精神和范围。Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the application and not to limit them. Although the application has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the application can be Make modifications or equivalent replacements without departing from the spirit and scope of the technical solution of the present application.
Claims (20)
- 一种识别模型的参数获取方法,其中,所述方法包括:A method for acquiring parameters of a recognition model, wherein the method includes:获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;Acquiring a training data set containing noise labels, and performing data standardization processing on the training data set to obtain a standard data set;基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;Establishing a recognition model based on a multi-layer deep neural network, and using the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;构建所述标准数据集的噪声概率转移矩阵;Constructing the noise probability transition matrix of the standard data set;基于所述噪声概率转移矩阵构建损失函数;Constructing a loss function based on the noise probability transition matrix;利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
- 如权利要求1所述的识别模型的参数获取方法,其中,所述对所述训练数据集进行数据标准化处理,包括下述一种或几种的组合:The method for acquiring parameters of a recognition model according to claim 1, wherein said performing data standardization processing on said training data set includes one or a combination of the following:去除所述训练数据集中的唯一属性值;Removing the unique attribute value in the training data set;对所述训练数据集进行缺失值填充;Filling in missing values on the training data set;对所述训练数据集进行数据归一化。Perform data normalization on the training data set.
- 如权利要求2所述的识别模型的参数获取方法,其中,所述对所述训练数据集进行数据归一化,包括:The method for acquiring parameters of a recognition model according to claim 2, wherein said performing data normalization on said training data set comprises:利用如下标准化算法对所述训练数据集进行数据归一化:Use the following standardized algorithm to normalize the training data set:其中,x为数据归一化后的标准数据,S old为所述训练数据集中的数据,S max为S old取值的最大值,S min为S old取值的最小值。 Wherein, x is the standard data of data normalization, S old data as the training data set, S max is the maximum value of S old, S min is the minimum value of the value S old.
- 如权利要求1所述的识别模型的参数获取方法,其中,所述噪声概率转移矩阵,包括:The method for acquiring parameters of a recognition model according to claim 1, wherein the noise probability transition matrix comprises:Q∈[0,1] c×c Q∈[0,1] c×c其中,c的大小与标准数据集中标准数据的数量相同。Among them, the size of c is the same as the number of standard data in the standard data set.
- 如权利要求1所述的识别模型的参数获取方法,其中,所述损失函数包括前向损失函数,所述前向损失函数为:The method for acquiring parameters of a recognition model according to claim 1, wherein the loss function includes a forward loss function, and the forward loss function is:
- 如权利要求3所述的识别模型的参数获取方法,其中,所述损失函数还包括后向损失函数,所述后项损失函数为:The method for acquiring parameters of a recognition model according to claim 3, wherein the loss function further comprises a backward loss function, and the latter loss function is:其中,l →(h)为所述后向损失函数的损失值,y为所述标准数据集中任一标准数据x的预设标准标签, 为所述标准识别模型对x的预测标签,Q为所述噪声概率转移矩阵,p(x,y)为标准数据x与x对应的预设标准标签y的联合分布, 为p(x,y)的预测值。 Where l → (h) is the loss value of the backward loss function, and y is the preset standard label of any standard data x in the standard data set, Is the predicted label of x by the standard recognition model, Q is the noise probability transition matrix, and p(x, y) is the joint distribution of standard data x and the preset standard label y corresponding to x, Is the predicted value of p(x,y).
- 如权利要求1至6中任一项所述的识别模型的参数获取方法,其中,所述利用所述损失函数计算所述标准识别模型的更新参数,包括:The method for acquiring parameters of a recognition model according to any one of claims 1 to 6, wherein said using the loss function to calculate the updated parameters of the standard recognition model comprises:获取所述标准数据集中标准数据的预设标准标签,以及所述标准识别模型对所述标准数据集中标准数据的预测标签;Acquiring the preset standard label of the standard data in the standard data set, and the prediction label of the standard data in the standard data set by the standard recognition model;利用损失函数计算所述预测标签与所述标准标签之间的差异值;Calculating a difference value between the predicted label and the standard label by using a loss function;当所述差异值在预设阈值区间内时,利用梯度下降算法计算所述标准识别模型的更新参数;When the difference value is within a preset threshold interval, use a gradient descent algorithm to calculate the update parameters of the standard recognition model;当所述差异值大于所述阈值区间的上限时,利用所述损失函数计算所述标准标签为噪声标签的概率值;When the difference value is greater than the upper limit of the threshold interval, use the loss function to calculate the probability value that the standard label is a noise label;当所述概率值小于预设概率阈值时,利用梯度下降算法计算所述标准识别模型的更新参数。When the probability value is less than the preset probability threshold, a gradient descent algorithm is used to calculate the update parameters of the standard recognition model.
- 一种识别模型的参数获取装置,其中,所述装置包括:A device for acquiring parameters of a recognition model, wherein the device includes:训练数据获取模块,用于获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;The training data acquisition module is used to acquire a training data set containing noise labels, and perform data standardization processing on the training data set to obtain a standard data set;识别模型构建模块,用于基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;A recognition model building module, used to establish a recognition model based on a multi-layer deep neural network, and use the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;转移矩阵构建模块,用于构建所述标准数据集的噪声概率转移矩阵;A transition matrix construction module, which is used to construct the noise probability transition matrix of the standard data set;损失函数构建模块,用于基于所述噪声概率转移矩阵构建损失函数;A loss function construction module, configured to construct a loss function based on the noise probability transition matrix;模型参数更新模块,用于利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The model parameter update module is used to calculate the update parameters of the standard recognition model by using the loss function, and replace the update parameters with the initialization parameters.
- 一种电子设备,其中,所述电子设备包括:An electronic device, wherein the electronic device includes:至少一个处理器;以及,At least one processor; and,与所述至少一个处理器通信连接的存储器;其中,A memory communicatively connected with the at least one processor; wherein,所述存储器存储有可被所述至少一个处理器执行的指令,所述指令被所述至少一个处理器执行,以使所述至少一个处理器能够执行如下所述的识别模型的参数获取方法:The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor, so that the at least one processor can execute the method for acquiring parameters of the recognition model as described below:获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;Acquiring a training data set containing noise labels, and performing data standardization processing on the training data set to obtain a standard data set;基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;Establishing a recognition model based on a multi-layer deep neural network, and using the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;构建所述标准数据集的噪声概率转移矩阵;Constructing the noise probability transition matrix of the standard data set;基于所述噪声概率转移矩阵构建损失函数;Constructing a loss function based on the noise probability transition matrix;利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
- 如权利要求9所述的电子设备,其中,所述对所述训练数据集进行数据标准化处理,包括下述一种或几种的组合:9. The electronic device according to claim 9, wherein said performing data standardization processing on said training data set comprises one or a combination of the following:去除所述训练数据集中的唯一属性值;Removing the unique attribute value in the training data set;对所述训练数据集进行缺失值填充;Filling in missing values on the training data set;对所述训练数据集进行数据归一化。Perform data normalization on the training data set.
- 如权利要求10所述的电子设备,其中,所述对所述训练数据集进行数据归一化,包括:11. The electronic device of claim 10, wherein said performing data normalization on said training data set comprises:利用如下标准化算法对所述训练数据集进行数据归一化:Use the following standardized algorithm to normalize the training data set:其中,x为数据归一化后的标准数据,S old为所述训练数据集中的数据,S max为S old取值的最大值,S min为S old取值的最小值。 Wherein, x is the standard data of data normalization, S old data as the training data set, S max is the maximum value of S old, S min is the minimum value of the value S old.
- 如权利要求9所述的电子设备,其中,所述噪声概率转移矩阵,包括:9. The electronic device of claim 9, wherein the noise probability transition matrix comprises:Q∈[0,1] c×c Q∈[0,1] c×c其中,c的大小与标准数据集中标准数据的数量相同。Among them, the size of c is the same as the number of standard data in the standard data set.
- 如权利要求9所述的电子设备,其中,所述损失函数包括前向损失函数,所述前向损失函数为:9. The electronic device of claim 9, wherein the loss function comprises a forward loss function, and the forward loss function is:
- 如权利要求11所述的电子设备,其中,所述损失函数还包括后向损失函数,所述后项损失函数为:11. The electronic device according to claim 11, wherein the loss function further comprises a backward loss function, and the latter loss function is:其中,l →(h)为所述后向损失函数的损失值,y为所述标准数据集中任一标准数据x的预设标准标签, 为所述标准识别模型对x的预测标签,Q为所述噪声概率转移矩阵,p(x,y)为标准数据x与x对应的预设标准标签y的联合分布, 为p(x,y)的预测值。 Where l → (h) is the loss value of the backward loss function, and y is the preset standard label of any standard data x in the standard data set, Is the predicted label of x by the standard recognition model, Q is the noise probability transition matrix, and p(x, y) is the joint distribution of standard data x and the preset standard label y corresponding to x, Is the predicted value of p(x,y).
- 一种计算机可读存储介质,包括存储数据区和存储程序区,存储数据区存储创建的数据,存储程序区存储有计算机程序;其中,所述计算机程序被处理器执行时实现如下所述的识别模型的参数获取方法:A computer-readable storage medium includes a storage data area and a storage program area. The storage data area stores created data, and the storage program area stores a computer program; wherein the computer program is executed by a processor to realize the following recognition The parameter acquisition method of the model:获取包含噪声标签的训练数据集,对所述训练数据集进行数据标准化处理,得到标准数据集;Acquiring a training data set containing noise labels, and performing data standardization processing on the training data set to obtain a standard data set;基于多层深度神经网络建立识别模型,利用所述标准数据集对所述识别模型进行训练,得到包含初始化参数的标准识别模型;Establishing a recognition model based on a multi-layer deep neural network, and using the standard data set to train the recognition model to obtain a standard recognition model including initialization parameters;构建所述标准数据集的噪声概率转移矩阵;Constructing the noise probability transition matrix of the standard data set;基于所述噪声概率转移矩阵构建损失函数;Constructing a loss function based on the noise probability transition matrix;利用所述损失函数计算所述标准识别模型的更新参数,将所述更新参数替换为所述初始化参数。The loss function is used to calculate the update parameters of the standard recognition model, and the update parameters are replaced with the initialization parameters.
- 如权利要求15所述的计算机可读存储介质,其中,所述对所述训练数据集进行数据标准化处理,包括下述一种或几种的组合:15. The computer-readable storage medium according to claim 15, wherein said performing data standardization processing on said training data set comprises one or a combination of the following:去除所述训练数据集中的唯一属性值;Removing the unique attribute value in the training data set;对所述训练数据集进行缺失值填充;Filling in missing values on the training data set;对所述训练数据集进行数据归一化。Perform data normalization on the training data set.
- 如权利要求16所述的计算机可读存储介质,其中,所述对所述训练数据集进行数据归一化,包括:15. The computer-readable storage medium of claim 16, wherein said performing data normalization on said training data set comprises:利用如下标准化算法对所述训练数据集进行数据归一化:Use the following standardized algorithm to normalize the training data set:其中,x为数据归一化后的标准数据,S old为所述训练数据集中的数据,S max为S old取值的最大值,S min为S old取值的最小值。 Wherein, x is the standard data of data normalization, S old data as the training data set, S max is the maximum value of S old, S min is the minimum value of the value S old.
- 如权利要求15所述的计算机可读存储介质,其中,所述噪声概率转移矩阵,包括:15. The computer-readable storage medium of claim 15, wherein the noise probability transition matrix comprises:Q∈[0,1] c×c Q∈[0,1] c×c其中,c的大小与标准数据集中标准数据的数量相同。Among them, the size of c is the same as the number of standard data in the standard data set.
- 如权利要求15所述的计算机可读存储介质,其中,所述损失函数包括前向损失函数,所述前向损失函数为:15. The computer-readable storage medium of claim 15, wherein the loss function comprises a forward loss function, and the forward loss function is:
- 如权利要求17所述的计算机可读存储介质,其中,所述损失函数还包括后向损失函数,所述后项损失函数为:17. The computer-readable storage medium of claim 17, wherein the loss function further comprises a backward loss function, and the latter loss function is:其中,l →(h)为所述后向损失函数的损失值,y为所述标准数据集中任一标准数据x的预设标准标签, 为所述标准识别模型对x的预测标签,Q为所述噪声概率转移矩阵,p(x,y)为标准数据x与x对应的预设标准标签y的联合分布, 为p(x,y)的预测值。 Where l → (h) is the loss value of the backward loss function, and y is the preset standard label of any standard data x in the standard data set, Is the predicted label of x by the standard recognition model, Q is the noise probability transition matrix, and p(x, y) is the joint distribution of standard data x and the preset standard label y corresponding to x, Is the predicted value of p(x,y).
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202010656659.8A CN111814962B (en) | 2020-07-09 | 2020-07-09 | Parameter acquisition method and device for identification model, electronic equipment and storage medium |
CN202010656659.8 | 2020-07-09 |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021151345A1 true WO2021151345A1 (en) | 2021-08-05 |
Family
ID=72842855
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/CN2020/131974 WO2021151345A1 (en) | 2020-07-09 | 2020-11-26 | Method and apparatus for parameter acquisition for recognition model, electronic device, and storage medium |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN111814962B (en) |
WO (1) | WO2021151345A1 (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113780473A (en) * | 2021-09-30 | 2021-12-10 | 平安科技(深圳)有限公司 | Data processing method and device based on depth model, electronic equipment and storage medium |
CN115270848A (en) * | 2022-06-17 | 2022-11-01 | 合肥心之声健康科技有限公司 | PPG and ECG automatic conversion intelligent algorithm, storage medium and computer system |
CN115860574A (en) * | 2023-02-06 | 2023-03-28 | 佰聆数据股份有限公司 | Method and device for analyzing use effect of charging equipment |
CN116908134A (en) * | 2023-09-12 | 2023-10-20 | 津海威视技术(天津)有限公司 | Semi-quantitative analysis method for plasticizer content and training method for analysis model |
CN117077016A (en) * | 2023-08-17 | 2023-11-17 | 中国自然资源航空物探遥感中心 | Supermatrix rock identification method of support vector machine based on aviation magnetic release data |
CN117349899A (en) * | 2023-12-06 | 2024-01-05 | 湖北省楚天云有限公司 | Sensitive data processing method, system and storage medium based on forgetting model |
Families Citing this family (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN111814962B (en) * | 2020-07-09 | 2024-05-10 | 平安科技(深圳)有限公司 | Parameter acquisition method and device for identification model, electronic equipment and storage medium |
CN112215238B (en) * | 2020-10-29 | 2022-06-07 | 支付宝(杭州)信息技术有限公司 | Method, system and device for constructing general feature extraction model |
CN112465141B (en) * | 2020-12-18 | 2024-06-28 | 平安科技(深圳)有限公司 | Model compression method, device, electronic equipment and medium |
CN113158911A (en) * | 2021-04-25 | 2021-07-23 | 北京华捷艾米科技有限公司 | Data generation method and device |
CN113902121B (en) * | 2021-07-15 | 2023-07-21 | 陈九廷 | Method, device, equipment and medium for verifying battery degradation estimation device |
CN113706204B (en) * | 2021-08-31 | 2024-04-05 | 中国平安财产保险股份有限公司 | Deep learning-based rights issuing method, device, equipment and storage medium |
US20230259762A1 (en) * | 2022-02-14 | 2023-08-17 | Samsung Electronics Co., Ltd. | Machine learning with instance-dependent label noise |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110543898A (en) * | 2019-08-16 | 2019-12-06 | 上海数禾信息科技有限公司 | Supervised learning method for noise label, data classification processing method and device |
CN110929733A (en) * | 2019-12-09 | 2020-03-27 | 上海眼控科技股份有限公司 | Denoising method and device, computer equipment, storage medium and model training method |
CN111079836A (en) * | 2019-12-16 | 2020-04-28 | 浙江大学 | Process data fault classification method based on pseudo label method and weak supervised learning |
CN111191726A (en) * | 2019-12-31 | 2020-05-22 | 浙江大学 | Fault classification method based on weak supervised learning multi-layer perceptron |
CN111814962A (en) * | 2020-07-09 | 2020-10-23 | 平安科技(深圳)有限公司 | Method and device for acquiring parameters of recognition model, electronic equipment and storage medium |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE102015010602A1 (en) * | 2015-08-18 | 2017-02-23 | Hochschule Aschaffenburg | Method for analyzing a data set of a time-of-flight mass spectrometry measurement and a device |
CN107563567A (en) * | 2017-09-18 | 2018-01-09 | 河海大学 | Core extreme learning machine Flood Forecasting Method based on sparse own coding |
CN109450830B (en) * | 2018-12-26 | 2021-09-21 | 重庆大学 | Channel estimation method based on deep learning in high-speed mobile environment |
-
2020
- 2020-07-09 CN CN202010656659.8A patent/CN111814962B/en active Active
- 2020-11-26 WO PCT/CN2020/131974 patent/WO2021151345A1/en active Application Filing
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN110543898A (en) * | 2019-08-16 | 2019-12-06 | 上海数禾信息科技有限公司 | Supervised learning method for noise label, data classification processing method and device |
CN110929733A (en) * | 2019-12-09 | 2020-03-27 | 上海眼控科技股份有限公司 | Denoising method and device, computer equipment, storage medium and model training method |
CN111079836A (en) * | 2019-12-16 | 2020-04-28 | 浙江大学 | Process data fault classification method based on pseudo label method and weak supervised learning |
CN111191726A (en) * | 2019-12-31 | 2020-05-22 | 浙江大学 | Fault classification method based on weak supervised learning multi-layer perceptron |
CN111814962A (en) * | 2020-07-09 | 2020-10-23 | 平安科技(深圳)有限公司 | Method and device for acquiring parameters of recognition model, electronic equipment and storage medium |
Cited By (11)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN113780473A (en) * | 2021-09-30 | 2021-12-10 | 平安科技(深圳)有限公司 | Data processing method and device based on depth model, electronic equipment and storage medium |
CN113780473B (en) * | 2021-09-30 | 2023-07-14 | 平安科技(深圳)有限公司 | Depth model-based data processing method and device, electronic equipment and storage medium |
CN115270848A (en) * | 2022-06-17 | 2022-11-01 | 合肥心之声健康科技有限公司 | PPG and ECG automatic conversion intelligent algorithm, storage medium and computer system |
CN115270848B (en) * | 2022-06-17 | 2023-09-29 | 合肥心之声健康科技有限公司 | PPG and ECG automatic conversion intelligent algorithm, storage medium and computer system |
CN115860574A (en) * | 2023-02-06 | 2023-03-28 | 佰聆数据股份有限公司 | Method and device for analyzing use effect of charging equipment |
CN117077016A (en) * | 2023-08-17 | 2023-11-17 | 中国自然资源航空物探遥感中心 | Supermatrix rock identification method of support vector machine based on aviation magnetic release data |
CN117077016B (en) * | 2023-08-17 | 2024-03-19 | 中国自然资源航空物探遥感中心 | Supermatrix rock identification method of support vector machine based on aviation magnetic release data |
CN116908134A (en) * | 2023-09-12 | 2023-10-20 | 津海威视技术(天津)有限公司 | Semi-quantitative analysis method for plasticizer content and training method for analysis model |
CN116908134B (en) * | 2023-09-12 | 2023-11-24 | 津海威视技术(天津)有限公司 | Semi-quantitative analysis method for plasticizer content and training method for analysis model |
CN117349899A (en) * | 2023-12-06 | 2024-01-05 | 湖北省楚天云有限公司 | Sensitive data processing method, system and storage medium based on forgetting model |
CN117349899B (en) * | 2023-12-06 | 2024-04-05 | 湖北省楚天云有限公司 | Sensitive data processing method, system and storage medium based on forgetting model |
Also Published As
Publication number | Publication date |
---|---|
CN111814962A (en) | 2020-10-23 |
CN111814962B (en) | 2024-05-10 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
WO2021151345A1 (en) | Method and apparatus for parameter acquisition for recognition model, electronic device, and storage medium | |
WO2021189906A1 (en) | Target detection method and apparatus based on federated learning, and device and storage medium | |
WO2021189826A1 (en) | Message generation method and apparatus, electronic device, and computer-readable storage medium | |
WO2022116424A1 (en) | Method and apparatus for training traffic flow prediction model, electronic device, and storage medium | |
WO2021208696A1 (en) | User intention analysis method, apparatus, electronic device, and computer storage medium | |
WO2022105179A1 (en) | Biological feature image recognition method and apparatus, and electronic device and readable storage medium | |
WO2021189913A1 (en) | Method and apparatus for target object segmentation in image, and electronic device and storage medium | |
CN112137591B (en) | Target object position detection method, device, equipment and medium based on video stream | |
CN111783982B (en) | Method, device, equipment and medium for acquiring attack sample | |
WO2022141858A1 (en) | Pedestrian detection method and apparatus, electronic device, and storage medium | |
WO2022227192A1 (en) | Image classification method and apparatus, and electronic device and medium | |
WO2021189827A1 (en) | Method and apparatus for recognizing blurred image, and device and computer-readable storage medium | |
WO2022194062A1 (en) | Disease label detection method and apparatus, electronic device, and storage medium | |
CN113158676A (en) | Professional entity and relationship combined extraction method and system and electronic equipment | |
CN113157739B (en) | Cross-modal retrieval method and device, electronic equipment and storage medium | |
CN113658002B (en) | Transaction result generation method and device based on decision tree, electronic equipment and medium | |
CN113434542B (en) | Data relationship identification method and device, electronic equipment and storage medium | |
CN114840684A (en) | Map construction method, device and equipment based on medical entity and storage medium | |
CN112269875A (en) | Text classification method and device, electronic equipment and storage medium | |
CN116468025A (en) | Electronic medical record structuring method and device, electronic equipment and storage medium | |
WO2022141867A1 (en) | Speech recognition method and apparatus, and electronic device and readable storage medium | |
CN112215336B (en) | Data labeling method, device, equipment and storage medium based on user behaviors | |
CN115049836A (en) | Image segmentation method, device, equipment and storage medium | |
CN115114408A (en) | Multi-modal emotion classification method, device, equipment and storage medium | |
WO2021151303A1 (en) | Named entity alignment device and apparatus, and electronic device and readable storage medium |
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: 20916345 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: 20916345 Country of ref document: EP Kind code of ref document: A1 |