CN111767028A - Cognitive resource management architecture and cognitive resource calling method - Google Patents
Cognitive resource management architecture and cognitive resource calling method Download PDFInfo
- Publication number
- CN111767028A CN111767028A CN202010525437.2A CN202010525437A CN111767028A CN 111767028 A CN111767028 A CN 111767028A CN 202010525437 A CN202010525437 A CN 202010525437A CN 111767028 A CN111767028 A CN 111767028A
- Authority
- CN
- China
- Prior art keywords
- interface
- cognitive
- model
- bus
- task
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F8/00—Arrangements for software engineering
- G06F8/20—Software design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- 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
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Biomedical Technology (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Data Mining & Analysis (AREA)
- Mathematical Physics (AREA)
- Biophysics (AREA)
- Artificial Intelligence (AREA)
- Life Sciences & Earth Sciences (AREA)
- Medical Informatics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Neurology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Stored Programmes (AREA)
Abstract
The embodiment of the invention provides a cognitive resource management framework and a cognitive resource calling method. The cognitive resource management system comprises a software bus and cognitive resources, wherein the software bus is positioned on the upper layer of the cognitive resources; the software bus provides a uniform calling interface for task requirements through interface design and function integration; the cognitive resources are subjected to standardized management on a pre-training model and an intelligent algorithm, and the upper-layer bus interface is dynamically loaded and initialized through application requirements and task parameters. According to the embodiment of the invention, the cognitive resource management architecture is divided into the software bus and the cognitive resource, and the cognitive resource is accessed and updated through the software bus interface; and the standardized management and unified calling mode of cognitive resources such as a pre-training model, an intelligent algorithm, a calculation framework, mathematical knowledge and the like is realized, so that distributed access to the cognitive resources can be supported, various intelligent models and algorithms can be reused, and a good intelligent algorithm development environment is provided.
Description
Technical Field
The invention relates to the technical field of artificial intelligence, in particular to a cognitive resource management framework and a cognitive resource calling method.
Background
In the field of artificial intelligence, artificial intelligence enables a machine to learn, think and understand like a human, and the artificial intelligence covers a wide range of subject fields such as cognition and reasoning, computer vision, natural language understanding and interaction, machine learning and the like. In recent years, the theory and technology of artificial intelligence have been developed rapidly, and the intelligent algorithm is more and more diversified. Traditional machine learning algorithms include decision trees, random forests, artificial neural networks, bayesian learning, and the like. With the advent of the big data era, artificial intelligence algorithms driven by a large amount of data are widely applied to various fields in real life, such as deep learning, reinforcement learning, transfer learning, meta learning and the like. The development of intelligent algorithms has led to the widespread use of artificial intelligence techniques in many fields, including robotics, speech recognition, image recognition, natural language processing, expert systems, and the like.
The intelligent algorithm is widely applied in various fields, meanwhile, the rapid development of various intelligent computing framework technologies is brought, corresponding intelligent computing frameworks are proposed in both academic circles and industrial circles, and more hardware architectures and software platforms are supported. Currently, the mainstream computing framework is: TensorFlow, Caffe, Apache System ML, PyTorch, MXNet, CNTK, Theano, and the like. Corresponding cognitive resource management architectures are needed to be designed in response to various task requirements, wherein cognitive resources comprise various intelligent algorithms, pre-training models, intelligent computing frames and corresponding mathematical knowledge, and calling and updating of machine learning pre-training models are provided for application development. However, the design for specific tasks and application scenarios often causes the tasks and models to be tightly coupled, and resource calls are repeated, which is not beneficial to unified management and development.
From the perspective of software architecture design, the software architecture design provides a basis and specification for large-scale development and provides reusable resources. The large-scale development of software systems must have a certain foundation and follow a certain specification, which is a requirement of software engineering itself and a requirement of user development. Therefore, a resource management framework based on task and model separation needs to be designed, so that the development of a bottom model and the development of an upper application are decoupled, and the effective reuse of resources is realized.
Disclosure of Invention
The embodiment of the invention provides a cognitive resource management architecture system and a cognitive resource calling method, which are used for solving the defects that in the prior art, due to the fact that tasks and models are tightly coupled, resource calling is repeated, and management and development are not facilitated.
In one aspect, an embodiment of the present invention provides a cognitive resource management architecture, including:
the software bus is positioned at the upper layer of the cognitive resource; wherein:
the software bus provides a uniform calling interface for task requirements through interface design and function integration;
the cognitive resources are subjected to standardized management on a pre-training model and an intelligent algorithm, and dynamic loading and initialization of an upper-layer bus interface through application requirements and task parameters are realized.
Further, the software bus comprises interface design and implementation, automatic selection and generation of models and model cross-frame migration; wherein:
the interface design and implementation are used for designing a unified external access interface and a normalized internal hierarchical calling interface;
the model automatic selection and generation are used for automatically selecting a matched model for the transmitted parameters for reasoning calculation based on an automatic machine learning technology according to task requirements, and when the existing model is not matched with the current task, a new model is generated through retraining or a neural network architecture search technology;
the model cross-framework migration is used for realizing the migration and distributed deployment of the model under different intelligent computing frameworks based on an open neural network switching method.
Further, the interface design and implementation comprises an external interface and an internal interface; wherein:
the external interface is used for using a universal container on the uppermost layer of the bus and providing a uniform calling interface for application service;
the internal interface is used for designing a fat tree structure through sequential derivation relations formed among the three hierarchies, and classifying and managing the cognitive resources in a subdivision mode.
Further, the internal interface comprises a bus base class interface, a specific task function bus subclass and a resource connection layer interface; wherein:
the bus base class interface is positioned on the uppermost layer of the internal interface and is used for instantiating the cognitive bus;
the specific task function bus subclass is positioned in the internal interface middle layer and is used for deriving and generating a subclass after the data type of the cognitive resource is explicitly called;
and the resource connection layer interface is positioned at the lowest layer of the internal interface and is used for creating a running environment of the cognitive resources and loading a corresponding intelligent algorithm and model for learning training and forward calculation.
Further, the cognitive resources comprise a pre-training model library, an intelligent algorithm library, a learning framework and a mathematical knowledge base; wherein:
the pre-training model library is used for managing and storing a plurality of trained models and managing the plurality of pre-training models based on realized tasks and application scenes;
the intelligent algorithm library is used for managing a plurality of machine learning algorithms, deep learning algorithms and reinforcement learning algorithms;
the learning framework is used for integrating the existing intelligent algorithm framework, deploying a framework for supporting learning training at the background, deploying an inference framework for supporting accelerated optimization at the front end, integrating common modules of various frameworks, and developing a self-learning training framework and an accelerated optimization module;
the mathematical knowledge base comprises mathematical knowledge required in the intelligent algorithm construction process and is used for providing mathematical basis and physical knowledge support for constructing the intelligent algorithm.
On the other hand, an embodiment of the present invention provides a method for invoking a cognitive resource, including:
unified external interfaces are used in the task plug-in through calling a software bus of cognitive resource management, target detection model calling and updating requirements are initiated, and task parameters and data are transmitted;
the software bus dynamically initializes different cognitive resources through hierarchical interface transmission according to the task parameters, selects and assembles a pre-training model for calculation or initiates learning training, and realizes specific interface calling;
loading the pre-training model or the intelligent algorithm and the dependent operating environment in an internal interface, initiating reasoning calculation or training requirements, and completing a target task;
and managing the pre-training model, the intelligent algorithm and the intelligent computing framework by the cognitive resource according to different target tasks.
According to the cognitive resource management framework and the cognitive resource calling method provided by the embodiment of the invention, the cognitive resource management framework is divided into the software bus and the cognitive resource, and the cognitive resource is accessed and updated through the software bus interface; and the standardized management and unified calling mode of cognitive resources such as a pre-training model, an intelligent algorithm, a calculation framework, mathematical knowledge and the like is realized, so that distributed access to the cognitive resources can be supported, various intelligent models and algorithms can be reused, and a good intelligent algorithm development environment is provided.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below, and it is obvious that the drawings in the following description are some embodiments of the present invention, and for those skilled in the art, other drawings can be obtained according to these drawings without creative efforts.
Fig. 1 is a general design diagram of a cognitive resource management architecture according to an embodiment of the present invention;
fig. 2 is a software bus interface design structure diagram of cognitive resource management provided in the embodiment of the present invention;
fig. 3 is a flowchart of a cognitive resource calling method according to an embodiment of the present invention;
fig. 4 is a schematic view of a whole flow of cognitive resource invocation based on a target detection task according to an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are some, but not all, embodiments of the present invention. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Aiming at the problems in the prior art, the embodiment of the invention provides a cognitive resource management architecture design scheme for realizing task and model separation, which separates various cognitive resources such as intelligent algorithms, pre-training models, calculation frames, mathematical knowledge and the like from specific application development, and provides calling and updating of a machine learning pre-training model through unified resource management and environment support architecture. On one hand, the method provides a calling requirement for calling during operation calculation, namely, when a task is executed, corresponding cognitive resources, trained neural network models, reinforcement learning models and the like can be called through an interface, and meanwhile, corresponding operation environments are loaded. On the other hand, online learning and model updating support is provided, namely when the existing pre-training model cannot meet the task requirements, training can be performed on a new data set based on an algorithm model and an intelligent computing frame provided by a cognitive resource management frame, and parameters or a network model are updated.
Fig. 1 is a general design diagram of a cognitive resource management architecture according to an embodiment of the present invention, as shown in fig. 1, including:
the software bus is positioned at the upper layer of the cognitive resource; wherein:
the software bus provides a uniform calling interface for task requirements through interface design and function integration;
the cognitive resources are subjected to standardized management on a pre-training model and an intelligent algorithm, and dynamic loading and initialization of an upper-layer bus interface through application requirements and task parameters are realized.
Specifically, the cognitive resource management architecture design based on task and algorithm separation provided by the embodiment of the invention is divided into a software bus and a cognitive resource.
A software bus of the cognitive resource management architecture provides a unified calling interface of various cognitive resources for meeting specific task requirements through interface design and function integration, and can integrate rich intelligent model application range and powerful functional modules and interfaces, such as automatic selection and generation of models, a model cross-frame migration technology and the like, so that the adaptivity of the cognitive resources and heterogeneous frame migration are enhanced.
The cognitive resource part of the cognitive resource management architecture mainly comprises the following steps: the system comprises a pre-training model base, an intelligent algorithm base, a learning framework and a mathematical knowledge base. The pre-training model and the intelligent algorithm are managed in a standardized naming mode, and therefore the upper-layer bus interface can be loaded and initialized dynamically through application requirements and task parameters. And the initialized cognitive resources exist in the form of data members in the bus, so that the data stream is efficiently processed.
According to the embodiment of the invention, the cognitive resource management architecture is divided into the software bus and the cognitive resource, and the cognitive resource is accessed and updated through the software bus interface; and the standardized management and unified calling mode of cognitive resources such as a pre-training model, an intelligent algorithm, a calculation framework, mathematical knowledge and the like is realized, so that distributed access to the cognitive resources can be supported, various intelligent models and algorithms can be reused, and a good intelligent algorithm development environment is provided.
Based on the above embodiment, the software bus comprises interface design and implementation, automatic selection and generation of models, and model cross-frame migration; wherein:
the interface design and implementation are used for designing a unified external access interface and a normalized internal hierarchical calling interface;
the model automatic selection and generation are used for automatically selecting a matched model for the transmitted parameters for reasoning calculation based on an automatic machine learning technology according to task requirements, and when the existing model is not matched with the current task, a new model is generated through retraining or a neural network architecture search technology;
the model cross-framework migration is used for realizing the migration and distributed deployment of the model under different intelligent computing frameworks based on an open neural network switching method.
The interface design and implementation comprises an external interface and an internal interface; wherein:
the external interface is used for using a universal container on the uppermost layer of the bus and providing a uniform calling interface for application service;
the internal interface is used for designing a fat tree structure through sequential derivation relations formed among the three hierarchies, and classifying and managing the cognitive resources in a subdivision mode.
The internal interface comprises a bus base class interface, a specific task function bus subclass and a resource connection layer interface; wherein:
the bus base class interface is positioned on the uppermost layer of the internal interface and is used for instantiating the cognitive bus;
the specific task function bus subclass is positioned in the internal interface middle layer and is used for deriving and generating a subclass after the data type of the cognitive resource is explicitly called;
and the resource connection layer interface is positioned at the lowest layer of the internal interface and is used for creating a running environment of the cognitive resources and loading a corresponding intelligent algorithm and model for learning training and forward calculation.
Specifically, as shown in fig. 2, the software bus of the cognitive resource management architecture specifically includes: the method comprises the following functions of interface design and realization, automatic selection and generation of models, model cross-frame migration and the like.
The interface design and implementation provide an external access interface with unified design and a normalized internal hierarchical calling interface, and meet the requirement that various task development layers realize the access and the update of cognitive resources through interface calling, wherein the external access interface and the internal interface are included.
And an external interface, wherein a universal container (boost) is used at the uppermost layer of the bus, a unified calling interface (registration: SoftBus) is provided for application services, the external complexity of the interface is reduced, and the unified management of the external interface is realized.
The internal interface is designed into a fat tree structure and comprises three levels of a bus base class (SoftBusBase), a specific task function bus subclass (SoftBusXX) and a resource connection layer interface (XXimplementation), and the three levels are in sequential derivation relationship. In this way, a fat tree structure is formed, and cognitive resources are classified and managed in a subdivided form. The SoftBusBase at the uppermost layer of the internal interface is used as a bus base class to instantiate the cognitive bus, and the softbusXX at the middle layer is a subclass derived after the data type of the cognitive resource is explicitly called. And the lowest resource link layer creates the operating environment of the cognitive resources and loads a corresponding intelligent algorithm and a corresponding model for learning training and forward calculation. Through the design and division of labor of three layers, the cognitive resources are called by the application layer through the bus, and the compatibility and the expandability of the internal interface are achieved.
And (3) automatic selection and generation of the model, namely, based on an automatic machine learning technology, automatically selecting a matched model according to the transmitted parameters for reasoning and calculation according to the task requirements, and generating a new model through retraining or a neural network architecture search technology when the existing model is not matched with the current task.
The model cross-frame migration technology is based on an open neural network switching method, and migration and distributed deployment of the model under different intelligent computing frames are realized.
Based on any one of the above embodiments, the cognitive resources include a pre-training model library, an intelligent algorithm library, a learning framework and a mathematical knowledge base; wherein:
the pre-training model library is used for managing and storing a plurality of trained models and managing the plurality of pre-training models based on realized tasks and application scenes;
the intelligent algorithm library is used for managing a plurality of machine learning algorithms, deep learning algorithms and reinforcement learning algorithms;
the learning framework is used for integrating the existing intelligent algorithm framework, deploying a framework for supporting learning training at the background, deploying an inference framework for supporting accelerated optimization at the front end, integrating common modules of various frameworks, and developing a self-learning training framework and an accelerated optimization module;
the mathematical knowledge base comprises mathematical knowledge required in the intelligent algorithm construction process and is used for providing mathematical basis and physical knowledge support for constructing the intelligent algorithm.
Specifically, the cognitive resource part of the cognitive resource management architecture mainly includes: the system comprises a pre-training model base, an intelligent algorithm base, a learning framework and a mathematical knowledge base.
The pre-training model library is mainly used for managing and storing various trained models and managing various pre-training models based on realized tasks and application scenes, so that the pre-training models can be quickly selected and matched under a specific task scene.
An intelligent algorithm library: the method mainly manages algorithms of various machine learning, deep learning, reinforcement learning and the like;
a learning framework: on one hand, the existing main intelligent algorithm framework is integrated, and a background deployment framework supporting learning training, such as: pytorch, Tensorflow, Darknet, Keras, Caffe, OpenAI Gy, the front-end deploys inference frameworks TensorRT, NCNN supporting accelerated optimization; on the other hand, common modules of various frames are integrated, and a learning training frame and a corresponding accelerated optimization module are developed.
A mathematical knowledge base: the method mainly comprises mathematical and physical knowledge required in the construction process of the intelligent algorithm, and provides mathematical basis and physical knowledge support for the construction of the intelligent algorithm.
Fig. 3 is a flowchart of a cognitive resource calling method according to an embodiment of the present invention, and as shown in fig. 3, the method includes:
s1, initiating a target detection model calling and updating requirement by calling a software bus of cognitive resource management to unify external interfaces in the task plug-in, and transmitting task parameters and data;
s2, the software bus dynamically initializes different cognitive resources through the hierarchical interface transmission according to the task parameters, selects and assembles a pre-training model to calculate or initiate learning training, and realizes specific interface calling;
s3, loading the pre-training model or intelligent algorithm and the dependent operating environment in the internal interface, initiating reasoning calculation or training requirements, and completing the target task;
and S4, managing the pre-training model, the intelligent algorithm and the intelligent computing framework by the cognitive resource according to different target tasks.
Specifically, as shown in fig. 4, the cognitive resource management architecture based on task and model separation is used to implement the cognitive resource call facing the target detection task as an example:
and the task plug-in layer initiates the calling and updating requirements of the target detection model and transmits corresponding task parameters and data by calling a software bus unified external interface (software bus) for cognitive resource management in the task plug-in.
And the software bus dynamically initializes different cognitive resources through hierarchical interface transmission according to task parameters, selects a pre-trained model to be assembled for calculation or initiates learning training, and realizes calling by a specific interface:
internal interface: aiming at a target detection task, three layers of internal interfaces are SoftBusBase, SoftBusDetection and DetectionTFImpl, and the layers are in a sequential derivation relationship. Cognitive resources of different tasks can derive subclass interfaces (softbus detection) for corresponding resource calling through derivation of a base class (softbus), meanwhile, resource connection layers (various different instances) of the class are built, corresponding pre-training models or intelligent algorithms and dependent running environments are loaded, inference calculation or learning training requirements are initiated, and target detection tasks are completed. For example, loading a pre-training model fast _ rcnn and carrying out target detection on the image by using a support frame Tensorflow required by forward calculation, wherein the corresponding resource link layer interface is DetectionTFImp.
And the lowest layer is a cognitive resource management layer, and the pre-training model, the intelligent algorithm and the intelligent computing framework are managed according to different tasks. For example, a target detection task has a plurality of pre-training models such as mobile _ ssd, mask _ rcnn, fast _ rcnn and the like for selection, wherein the fast _ rcnn pre-training model and the computing framework Tensorflow are loaded and initialized to process a target image data stream and return results such as the type, the confidence degree, the recognition frame and the like of target detection.
The above-described embodiments of the apparatus are merely illustrative, and the units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of the present embodiment. One of ordinary skill in the art can understand and implement it without inventive effort.
Through the above description of the embodiments, those skilled in the art will clearly understand that each embodiment can be implemented by software plus a necessary general hardware platform, and certainly can also be implemented by hardware. With this understanding in mind, the above-described technical solutions may be embodied in the form of a software product, which can be stored in a computer-readable storage medium such as ROM/RAM, magnetic disk, optical disk, etc., and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device, etc.) to execute the methods described in the embodiments or some parts of the embodiments.
Finally, it should be noted that: the above examples are only intended to illustrate the technical solution of the present invention, but not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; and such modifications or substitutions do not depart from the spirit and scope of the corresponding technical solutions of the embodiments of the present invention.
Claims (6)
1. A cognitive resource management architecture, comprising: the software bus is positioned at the upper layer of the cognitive resource; wherein:
the software bus provides a uniform calling interface for task requirements through interface design and function integration;
the cognitive resources are subjected to standardized management on a pre-training model and an intelligent algorithm, and dynamic loading and initialization of an upper-layer bus interface through application requirements and task parameters are realized.
2. The cognitive resource management architecture of claim 1, wherein the software bus comprises interface design and implementation, automatic selection and generation of models, and model cross-framework migration; wherein:
the interface design and implementation are used for designing a unified external access interface and a normalized internal hierarchical calling interface;
the model automatic selection and generation are used for automatically selecting a matched model for the transmitted parameters for reasoning calculation based on an automatic machine learning technology according to task requirements, and when the existing model is not matched with the current task, a new model is generated through retraining or a neural network architecture search technology;
the model cross-framework migration is used for realizing the migration and distributed deployment of the model under different intelligent computing frameworks based on an open neural network switching method.
3. The cognitive resource management architecture of claim 2, wherein the interface design and implementation includes an external interface and an internal interface; wherein:
the external interface is used for using a universal container on the uppermost layer of the bus and providing a uniform calling interface for application service;
the internal interface is used for designing a fat tree structure through sequential derivation relations formed among the three hierarchies, and classifying and managing the cognitive resources in a subdivision mode.
4. The cognitive resource management architecture of claim 3, wherein the internal interfaces comprise a bus base class interface, a task specific function bus subclass, and a resource connection layer interface; wherein:
the bus base class interface is positioned on the uppermost layer of the internal interface and is used for instantiating the cognitive bus;
the specific task function bus subclass is positioned in the internal interface middle layer and is used for deriving and generating a subclass after the data type of the cognitive resource is explicitly called;
and the resource connection layer interface is positioned at the lowest layer of the internal interface and is used for creating a running environment of the cognitive resources and loading a corresponding intelligent algorithm and model for learning training and forward calculation.
5. The cognitive resource management architecture of claim 1, wherein the cognitive resources comprise a pre-training model library, an intelligent algorithm library, a learning framework, and a mathematical knowledge base; wherein:
the pre-training model library is used for managing and storing a plurality of trained models and managing the plurality of pre-training models based on realized tasks and application scenes;
the intelligent algorithm library is used for managing a plurality of machine learning algorithms, deep learning algorithms and reinforcement learning algorithms;
the learning framework is used for integrating the existing intelligent algorithm framework, deploying a framework for supporting learning training at the background, deploying an inference framework for supporting accelerated optimization at the front end, integrating common modules of various frameworks, and developing a self-learning training framework and an accelerated optimization module;
the mathematical knowledge base comprises mathematical knowledge required in the intelligent algorithm construction process and is used for providing mathematical basis and physical knowledge support for constructing the intelligent algorithm.
6. A cognitive resource calling method based on the architecture of any one of claims 1 to 5, comprising:
unified external interfaces are adopted in the task plug-in unit by calling a software bus for cognitive resource management, model calling and updating requirements are initiated, and task parameters and data are transmitted;
the software bus dynamically initializes different cognitive resources through hierarchical interface transmission according to the task parameters, selects and assembles a pre-training model for calculation or initiates learning training, and realizes specific interface calling;
loading the pre-training model or the intelligent algorithm and the dependent operating environment in an internal interface, initiating reasoning calculation or training requirements, and completing a task;
and managing the pre-training model, the intelligent algorithm and the intelligent computing framework by the cognitive resource according to different tasks.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010525437.2A CN111767028B (en) | 2020-06-10 | 2020-06-10 | Cognitive resource management architecture and cognitive resource calling method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010525437.2A CN111767028B (en) | 2020-06-10 | 2020-06-10 | Cognitive resource management architecture and cognitive resource calling method |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN111767028A true CN111767028A (en) | 2020-10-13 |
| CN111767028B CN111767028B (en) | 2023-09-19 |
Family
ID=72720609
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010525437.2A Active CN111767028B (en) | 2020-06-10 | 2020-06-10 | Cognitive resource management architecture and cognitive resource calling method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111767028B (en) |
Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102624870A (en) * | 2012-02-01 | 2012-08-01 | 北京航空航天大学 | Intelligent optimization algorithm based cloud manufacturing computing resource reconfigurable collocation method |
| CN102722368A (en) * | 2012-05-21 | 2012-10-10 | 北京广利核***工程有限公司 | Plug-in software designing method based on document tree and message pump |
| US20160140452A1 (en) * | 2014-11-17 | 2016-05-19 | Optimitive S.L.U. | Methods and systems using a composition of autonomous self-learning software components for performing complex real time data-processing tasks |
| CN107851031A (en) * | 2015-05-08 | 2018-03-27 | 佛罗乔有限责任公司 | Data find node |
| CN108881446A (en) * | 2018-06-22 | 2018-11-23 | 深源恒际科技有限公司 | A kind of artificial intelligence plateform system based on deep learning |
| CN109271147A (en) * | 2018-09-13 | 2019-01-25 | 北京车联天下信息技术有限公司 | A kind of intelligence cockpit system layer design method and intelligent cockpit system |
| CN109347950A (en) * | 2018-10-17 | 2019-02-15 | 南京邮电大学 | A kind of Internet of Things intelligence s ervice system and its implementation based on Kaa Project |
| CN110490334A (en) * | 2019-08-27 | 2019-11-22 | 上海丙晟科技有限公司 | A kind of machine learning of low latency is the generation method of service |
| CN111176822A (en) * | 2020-01-02 | 2020-05-19 | 中国航空工业集团公司西安航空计算技术研究所 | OODA multi-task intelligent application-oriented embedded public software operation method |
-
2020
- 2020-06-10 CN CN202010525437.2A patent/CN111767028B/en active Active
Patent Citations (9)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN102624870A (en) * | 2012-02-01 | 2012-08-01 | 北京航空航天大学 | Intelligent optimization algorithm based cloud manufacturing computing resource reconfigurable collocation method |
| CN102722368A (en) * | 2012-05-21 | 2012-10-10 | 北京广利核***工程有限公司 | Plug-in software designing method based on document tree and message pump |
| US20160140452A1 (en) * | 2014-11-17 | 2016-05-19 | Optimitive S.L.U. | Methods and systems using a composition of autonomous self-learning software components for performing complex real time data-processing tasks |
| CN107851031A (en) * | 2015-05-08 | 2018-03-27 | 佛罗乔有限责任公司 | Data find node |
| CN108881446A (en) * | 2018-06-22 | 2018-11-23 | 深源恒际科技有限公司 | A kind of artificial intelligence plateform system based on deep learning |
| CN109271147A (en) * | 2018-09-13 | 2019-01-25 | 北京车联天下信息技术有限公司 | A kind of intelligence cockpit system layer design method and intelligent cockpit system |
| CN109347950A (en) * | 2018-10-17 | 2019-02-15 | 南京邮电大学 | A kind of Internet of Things intelligence s ervice system and its implementation based on Kaa Project |
| CN110490334A (en) * | 2019-08-27 | 2019-11-22 | 上海丙晟科技有限公司 | A kind of machine learning of low latency is the generation method of service |
| CN111176822A (en) * | 2020-01-02 | 2020-05-19 | 中国航空工业集团公司西安航空计算技术研究所 | OODA multi-task intelligent application-oriented embedded public software operation method |
Non-Patent Citations (5)
| Title |
|---|
| ZIYAN WU ET AL.: "A data-driven approach of performance evaluation for cache server groups in content delivery network", JOURNAL OF PARALLEL AND DISTRIBUTED COMPUTING, vol. 119 * |
| 姚民康: "多模型激励式的工业过程仿真支撑***的设计与实现", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 01 * |
| 姜皓: "Java基础---接口的派生", Retrieved from the Internet <URL:https://blog.csdn.net/jianghao233/article/details/101476434> * |
| 王之元: "并行计算可扩展性分析与优化——能耗、可靠性与计算性能", 《中国优秀博士学位论文全文数据库信息科技辑》, no. 07 * |
| 王昕睿: "基于机器学习的房产智能自动评估模型的研究与***实现", 《中国优秀硕士学位论文全文数据库信息科技辑》, no. 08 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN111767028B (en) | 2023-09-19 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| US11762635B2 (en) | Artificial intelligence engine with enhanced computing hardware throughput | |
| US11868896B2 (en) | Interface for working with simulations on premises | |
| US11120299B2 (en) | Installation and operation of different processes of an AI engine adapted to different configurations of hardware located on-premises and in hybrid environments | |
| US11841789B2 (en) | Visual aids for debugging | |
| WO2018236674A1 (en) | For hiearchical decomposition deep reinforcement learning for an artificial intelligence model | |
| CN112272234B (en) | Platform management system and method for realizing edge cloud cooperation intelligent service | |
| Aguilar et al. | A methodology to specify multiagent systems | |
| CN115202868A (en) | Autonomous controllable heterogeneous intelligent computing service platform and intelligent scene matching method | |
| CN113821648B (en) | Robot task processing method and system based on ontology knowledge representation | |
| Abbas | Realizing the NOSHAPE MAS organizational model: an operational view | |
| CN111767028A (en) | Cognitive resource management architecture and cognitive resource calling method | |
| CN110083350B (en) | Micro-service self-adaptive evolution method based on RMAE (remote Markov experience) in cloud computing environment | |
| Firth et al. | Robotopias: mapping utopian perspectives on new industrial technology | |
| Xu et al. | The frame problem, the relevance problem, and a package solution to both | |
| Jennings | Joint intentions as a model of multi-agent cooperation | |
| CN115565104A (en) | Action prediction method and related equipment thereof | |
| CN115373761A (en) | AI-based dynamic loading method for scene resources in real-time cloud rendering | |
| CN110674935B (en) | Method for transplanting intelligent algorithm to airborne embedded platform and intelligent computing platform | |
| Schmitt et al. | Programming electronic institutions with utopia | |
| CN113672603A (en) | Multi-source heterogeneous electric power big data automatic label implementation method and system | |
| CN112817581A (en) | Lightweight intelligent service construction and operation support method | |
| Harel et al. | Categorizing methods for integrating machine learning with executable specifications | |
| Jennings | Joint intentions as a model of multi-agent cooperation in complex dynamic environments | |
| CN117934997B (en) | Large language model system and method for generating camera case sample | |
| CN116362109B (en) | Intelligent unmanned system and method based on digital twinning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |