CN115879536A - Learning cognition analysis model robustness optimization method based on causal effect - Google Patents
Learning cognition analysis model robustness optimization method based on causal effect Download PDFInfo
- Publication number
- CN115879536A CN115879536A CN202211233184.7A CN202211233184A CN115879536A CN 115879536 A CN115879536 A CN 115879536A CN 202211233184 A CN202211233184 A CN 202211233184A CN 115879536 A CN115879536 A CN 115879536A
- Authority
- CN
- China
- Prior art keywords
- cognitive
- learning
- causal
- analysis
- robustness optimization
- 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.)
- Pending
Links
Images
Classifications
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Management, Administration, Business Operations System, And Electronic Commerce (AREA)
Abstract
The invention discloses a learning cognition analysis model robustness optimization method based on causal effect, which comprises the following steps: constructing a cognitive analysis multi-type model in a hybrid learning process; constructing a learner cognitive dynamic causal network and a causal effect framework; and performing robustness optimization according to the cognition analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework, wherein the robustness optimization comprises mixed learning process cognition analysis time domain perception robustness optimization, mixed learning multi-mode cognition analysis model structure robustness optimization and cognition data distribution deviation target domain robustness optimization in a complex scene. The method can improve the robustness of the model and the analysis efficiency of the model, and can be widely applied to the technical field of computers.
Description
Technical Field
The invention relates to the technical field of computers, in particular to a learning cognition analysis model robustness optimization method based on causal effect.
Background
The mixed learning multi-modal process data contains rich learning cognition depiction information, but the factors are numerous and continuously change at any time, most of the existing mainstream knowledge tracking and cognition diagnosis methods are based on independent same distribution hypothesis and cannot adapt to mixed learning scenes with variable data distribution.
At present, the robustness of the hybrid learning cognitive modeling analysis in a complex scene is insufficient, and the model analysis effect in the complex education scene is low.
Disclosure of Invention
In view of this, the embodiment of the present invention provides a learning cognitive analysis model robustness optimization method based on a causal effect, so as to improve the robustness of the model and improve the model analysis efficiency.
In one aspect, the embodiment of the invention provides a causal effect-based learning cognitive analysis model robustness optimization method, which comprises the following steps:
constructing a cognitive analysis multi-type model in a hybrid learning process;
constructing a learner cognitive dynamic cause and effect network and a cause and effect framework;
carrying out robustness optimization according to the cognition analysis multi-type model and the learner cognition dynamic causal network and causal effect framework;
the robustness optimization comprises time domain perception robustness optimization of cognitive analysis in a hybrid learning process, structure robustness optimization of a hybrid learning multi-mode cognitive analysis model and robustness optimization of a cognitive data distribution offset target domain in a complex scene.
Optionally, the constructing a cognitive analysis multi-type model of the hybrid learning process includes:
performing modal feature extraction on the collected multi-modal learning process data; wherein the modal features include video features, audio features, and text features;
performing combined learning on the extracted multi-mode features by adopting a multi-type backbone network and a time sequence network to obtain the global representation of learning cognitive features;
and constructing different cognitive classifiers or regressors aiming at different learning cognitive calculation downstream tasks.
Optionally, the constructing a learner cognitive dynamic causal network and causal effect framework includes:
constructing a learner cognition causal graph by learning cognition influence variable detection and constructing a causal directed graph;
and (3) carrying out an effect averaging or optimizing method by utilizing a common cause and effect calculation method in the field and adopting a multipath effect calculation mode to obtain a cause and effect calculation result.
Optionally, when the robustness optimization is a hybrid learning process cognitive analysis time-domain perceptual robustness optimization, the performing robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework includes:
based on a direct total causal effect framework and a learner cognitive dynamic causal network model, a reasonable causal hypothesis of data perception uncertainty factors is provided, and learning cognitive data perception is constructed for anti-fact inference analysis;
aiming at the problem that data are partially lost in the process of hybrid learning, a causal total effect-based guidance random gating network mechanism is constructed, the intervention data generation of a potential data perception loss scene is carried out, and the optimization space search of a learning cognitive analysis model is enhanced;
aiming at the problem of irregular data perception sampling granularity under a complex situation, constructing a method for any data perception sampling granularity under a continuous time domain based on a time sequence differential neural network model;
the random gating mechanism and the irregular continuous time domain perception method are combined for use, and the cognitive analysis time domain perception optimization in the hybrid learning process is achieved.
Optionally, when the robustness optimization is a structural robustness optimization of a hybrid learning multi-modal cognitive analysis model, the performing robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework includes:
analyzing and learning the structural weight distribution characteristics of the cognitive analysis network, constructing a fluctuation suppression-gain model structural plasticity mechanism aiming at a complex cognitive analysis scene by combining a biological plasticity theory, and determining the neuron connection fluctuation description of a time sequence encoder and a decoder in a cognitive analysis model;
aiming at the data change characteristic of a learning cognitive computation task in a complex scene, determining a plasticity weight self-adaptive dynamic update rule by combining an uncertain factor causal total effect computation framework;
and (4) embedding a structure of plasticity dynamic weight according to the structural characteristics of the learning cognitive analysis model, and completing structural robustness optimization of the mixed learning multi-mode cognitive analysis model.
Optionally, when the robustness optimization is robustness optimization of a cognitive data distribution shift target domain in a complex scene, the performing robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework includes:
determining migration dynamic distribution of complex scene data of a learning cognition analysis task based on element distribution and dynamic learning cognition causal representation, and analyzing a distribution offset potential process of a pre-training data source domain and an application target domain;
obtaining a predicted value of the transferred dynamic distribution, and performing learning cognitive analysis on a target data domain and performing target independent component extraction and gain;
and performing target domain data enhancement generation according to the gain target domain independent component, performing target domain experience risk minimization optimization based on the enhanced data, and completing cognitive data distribution migration target domain learning cognitive state robustness analysis in a complex scene.
In another aspect, an embodiment of the present invention further provides a device for optimizing robustness of a learning cognitive analysis model based on a causal effect, including:
the first module is used for constructing a cognitive analysis multi-type model in a hybrid learning process;
the second module is used for constructing a learner cognitive dynamic causal network and a causal effect framework;
the third module is used for carrying out robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework;
the robustness optimization comprises time domain perception robustness optimization of cognitive analysis in a hybrid learning process, structure robustness optimization of a hybrid learning multi-mode cognitive analysis model and robustness optimization of a cognitive data distribution offset target domain in a complex scene.
Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;
the memory is used for storing programs;
the processor executes the program to implement the method as described above.
Yet another aspect of the embodiments of the present invention provides a computer-readable storage medium, which stores a program, which is executed by a processor to implement the method as described above.
The embodiment of the invention also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and the computer instructions executed by the processor cause the computer device to perform the foregoing method.
The embodiment of the invention constructs a cognitive analysis multi-type model in a hybrid learning process; constructing a learner cognitive dynamic causal network and a causal effect framework; and carrying out robustness optimization according to the cognition analysis multi-type model and the learner cognition dynamic causal network and causal effect framework, wherein the robustness optimization comprises cognition analysis time domain perception robustness optimization in a mixed learning process, mixed learning multi-mode cognition analysis model structure robustness optimization and cognition data distribution deviation target domain robustness optimization in a complex scene. The method can improve the robustness of the model and improve the analysis efficiency of the model.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required to be used in the description of the embodiments are briefly introduced below, and it is obvious that the drawings in the description below are only some embodiments of the present application, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
FIG. 1 is a flowchart illustrating the overall steps provided by an embodiment of the present invention.
Detailed Description
In order to make the objects, technical solutions and advantages of the present application more clearly understood, the present application is further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of and not restrictive on the broad application.
In view of the problems in the prior art, an aspect of the embodiments of the present invention provides a learning cognitive analysis model robustness optimization method based on causal effect, including:
constructing a cognitive analysis multi-type model in a hybrid learning process;
constructing a learner cognitive dynamic causal network and a causal effect framework;
carrying out robustness optimization according to the cognition analysis multi-type model and the learner cognition dynamic causal network and causal effect framework;
the robustness optimization comprises time domain perception robustness optimization of cognitive analysis in a hybrid learning process, structure robustness optimization of a hybrid learning multi-mode cognitive analysis model and robustness optimization of cognitive data distribution migration target domains in complex scenes.
Optionally, the constructing a cognitive analysis multi-type model of the hybrid learning process includes:
carrying out modal characteristic extraction on the collected multi-modal learning process data; wherein the modal features comprise video features, audio features, and text features;
performing combined learning on the extracted multi-modal characteristics by adopting a multi-type backbone network and a time sequence network to obtain the global representation of learning cognitive characteristics;
and constructing different cognitive classifiers or regressors aiming at different learning cognitive calculation downstream tasks.
Optionally, the constructing a learner cognitive dynamic causal network and causal effect framework includes:
constructing a learner cognition causal graph by learning cognition influence variable detection and constructing a causal directed graph;
and (3) carrying out an effect averaging or maximization method by utilizing a common causal effect calculation method in the field and adopting a multipath effect calculation mode to obtain a causal effect calculation result.
Optionally, when the robustness optimization is a hybrid learning process cognitive analysis time-domain perceptual robustness optimization, the performing robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework includes:
based on a direct general causal effect framework and a learner cognitive dynamic causal network model, providing a reasonable causal hypothesis of a data perception uncertainty factor, and constructing learning cognitive data perception for performing counterfactual inference analysis;
aiming at the problem that data are partially lost in the process of hybrid learning possibly, a causal total effect-based guided random gating network mechanism is constructed, potential data perception missing scene intervention data generation is carried out, and learning cognitive analysis model optimization space search is enhanced;
aiming at the problem of irregular data perception sampling granularity in a complex situation, constructing a method for any data perception sampling granularity in a continuous time domain based on a time sequence differential neural network model;
the random gating mechanism and the irregular continuous time domain perception method are combined for use, and the time domain perception optimization of cognitive analysis in the hybrid learning process is achieved.
Optionally, when the robustness optimization is a structural robustness optimization of a hybrid learning multi-modal cognitive analysis model, the performing robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework includes:
analyzing and learning the structural weight distribution characteristics of the cognitive analysis network, constructing a fluctuation suppression-gain model structural plasticity mechanism aiming at a complex cognitive analysis scene by combining a biological plasticity theory, and determining the neuron connection fluctuation description of a time sequence encoder and a decoder in a cognitive analysis model;
aiming at the data change characteristic of a learning cognitive computation task in a complex scene, determining a plasticity weight self-adaptive dynamic update rule by combining an uncertain factor causal total effect computation framework;
and (4) according to the structural characteristics of the learning cognitive analysis model, performing framework embedding of plasticity dynamic weights, and completing structural robustness optimization of the mixed learning multi-mode cognitive analysis model.
Optionally, when the robustness optimization is robustness optimization of a cognitive data distribution shift target domain in a complex scene, the robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework includes:
determining migration dynamic distribution of complex scene data of a learning cognition analysis task based on element distribution and dynamic learning cognition causal representation, and analyzing a distribution offset potential process of a pre-training data source domain and an application target domain;
obtaining a predicted value of the transferred dynamic distribution, and performing learning cognitive analysis on a target data domain and performing target independent component extraction and gain;
and performing target domain data enhancement generation according to the gain target domain independent component, performing target domain experience risk minimization optimization based on the enhanced data, and completing cognitive data distribution migration target domain learning cognitive state robustness analysis in a complex scene.
In another aspect, an embodiment of the present invention further provides a device for optimizing robustness of a learning cognitive analysis model based on a causal effect, where the device includes:
the first module is used for constructing a cognitive analysis multi-type model in a hybrid learning process;
the second module is used for constructing a learner cognitive dynamic cause-and-effect network and a cause-and-effect framework;
the third module is used for carrying out robustness optimization according to the cognition analysis multi-type model and the learner cognition dynamic causal network and causal effect framework;
the robustness optimization comprises time domain perception robustness optimization of cognitive analysis in a hybrid learning process, structure robustness optimization of a hybrid learning multi-mode cognitive analysis model and robustness optimization of a cognitive data distribution offset target domain in a complex scene.
Another aspect of the embodiments of the present invention further provides an electronic device, including a processor and a memory;
the memory is used for storing programs;
the processor executes the program to implement the method as described above.
Another aspect of the embodiments of the present invention also provides a computer-readable storage medium, which stores a program, and the program is executed by a processor to implement the method as described above.
The embodiment of the invention also discloses a computer program product or a computer program, which comprises computer instructions, and the computer instructions are stored in a computer readable storage medium. The computer instructions may be read by a processor of a computer device from a computer-readable storage medium, and the computer instructions executed by the processor cause the computer device to perform the foregoing method.
The following detailed description of the embodiments of the present invention is made with reference to the accompanying drawings:
as shown in fig. 1, aiming at the problem that the robustness of the existing hybrid learning cognitive modeling analysis in a complex scene is insufficient, the invention provides a causal effect-based cognitive analysis model robustness optimization method in the complex learning scene based on multi-modal learning process data in combination with the development of causal learning research, and solves the challenging problems of low model analysis effect and the like in the complex education scene. The method comprises the following specific steps: s1, constructing a cognitive analysis multi-type model in a hybrid learning process; s2, constructing a learner cognitive dynamic causal network and a causal effect framework; and S3, carrying out robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework, wherein the robustness optimization comprises cognitive analysis time domain perception robustness optimization in a mixed learning process, mixed learning multi-mode cognitive analysis model structure robustness optimization and cognitive data distribution deviation target domain robustness optimization in a complex scene.
The following is introduced for the specific implementation process of each step:
s1, constructing a multi-modal cognitive analysis model framework in a hybrid learning process:
the robustness analysis method provided by the invention mainly aims at a multi-mode cognitive analysis model in a hybrid learning process, and an analysis model framework mainly comprises multi-mode feature extraction, multi-mode feature joint learning, cognitive analysis downstream task construction and the like. The following is a description of the several main functions within the framework and the aspects that need to be robust optimized.
1) And multi-mode multi-feature extraction in the learning process: the method mainly comprises the steps of carrying out modal feature extraction on collected multi-modal learning process data, such as video features, audio features, text features and the like, and preparing feature data for subsequent multi-modal feature combined learning. For the extraction of different modal characteristics, different backbone networks can be used for extraction, and the invention uniformly uses a Transformer network as the backbone network and performs corresponding pre-training optimization aiming at the different modal characteristics. Because the acquired data relates to space-time conversion, the sampling of the data has a larger problem in the aspects of sampling time resolution and loss, and aiming at the problem, the invention provides a design for optimizing and enhancing the cognitive analysis time domain perception robustness in the mixed learning process, which is further elaborated later.
2) And multi-modal feature joint learning: the method mainly performs combined learning on the extracted multi-modal features to obtain a global representation of learning cognitive features, and mainly adopts a multi-type backbone network + time sequence network architecture. Due to the fact that the collected time sequence data are various in types, structured or unstructured, and have graph structures or non-graph structures, corresponding structure 'bias' or 'bias' exists due to the structural characteristics of the used models when corresponding features are extracted. Different network modules are also used in multi-feature contact learning, and the similarity exists. Aiming at the problem, the invention provides a robustness optimization and enhancement design oriented to a mixed multi-modal cognitive analysis model structure, which will be further elaborated later.
3) And (3) cognitive analysis downstream task construction: different learning cognition classifiers/regressions need to be constructed for different learning cognition calculation downstream tasks. Although many similarities exist in the learning cognitive tasks, application analysis can be performed by adopting a large-scale pre-training (large model, source domain) + target task fine tuning (small data, target domain) "model, the problem of source domain and target domain distribution deviation still exists, and the problem of poor adaptability of a new training model in a complex scene is caused. Aiming at the problem, the invention provides a robustness optimization and enhancement design of a cognitive data distribution deviation target domain in a complex scene, which is further elaborated later.
S2, learner cognitive causal discovery and causal effect framework design
In the invention, the robustness optimization in multiple aspects of a learning cognition analysis model is required to be carried out based on the learning cognition causal effect, so that a learner cognition causal graph and a causal effect calculation mechanism are required to be constructed.
1) Constructing a learner cognition cause and effect graph: the learner cognition causal graph construction mainly comprises two steps of learning cognition influence variable detection and causal directed graph construction. The method mainly comprises the steps of carrying out large-scale condition independence test on acquired data and obtaining potential undiscovered learning cognition influence variables. The learning cognition cause and effect graph is constructed by mainly adopting a directed graph gradient descent method, the learning cognition cause and effect graph is mainly achieved to be a directed adjacency matrix, the learning cognition cause and effect graph is fitted through error gradient back propagation, in order to increase the interpretability of the automatic generation cause and effect graph, the recognized cause and effect relationship in the current field is pre-filled before automatic graph construction, and then the learned cognition cause and effect graph is kept unchanged in the construction process. And setting a preset acceptable error loss in the process of patterning, and finishing the construction of the learning cognitive causal graph when the acceptable error is reached.
2) And designing a causal effect framework: the causal effect calculation of the invention adopts a 'multi-path calculation + result optimization' framework, namely, a causal effect calculation method commonly used in the field, such as a direct total effect, an indirect causal effect and the like, is utilized, a multi-path effect calculation mode is adopted on the causal effect calculation method and the assumption, and then an effect averaging or a most-valued method is carried out to obtain a more accurate causal effect calculation result.
S3, carrying out robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework:
1) Time domain perception robustness optimization of cognitive analysis in hybrid learning process
Due to the fact that the mixed learning environment data sensing has cross-space-time characteristics, uncertainty situations such as granularity mismatching and loss can occur in data sensing sampling due to various complex factors, and aiming at the problems, the method combines a random gating mechanism and a neural differential network technology to conduct time domain sensing robustness optimization of cognitive analysis in the mixed learning process. Firstly, based on a direct general causal effect framework and a learner cognitive dynamic causal network model, proposing a reasonable causal hypothesis of data perception uncertainty factors (such as sampling granularity and perception deficiency), constructing learning cognitive data perception for counterfactual inference analysis, and supposing that the ith cause and effect model is subjected toLack of data perception of individual learning cognitive factors (x) 0 = null), its direct total effect analysis distribution can be expressed as:
T(Y i )＝[Y d ＝i|do(X＝x)]-[Y d ＝i|do(X＝x 0 )], (1)
the first term on the right of the equation is the cognitive variable distribution corresponding to the original cause-and-effect diagram, the second term on the right of the equation represents the distribution result of the inverse fact interference, and d represents the value of the intermediate harmonic variable related to momentum.
And then, aiming at the problem that data are partially lost in the process of hybrid learning, a causal total effect-based guided random gating network mechanism is constructed, the generation of the intervention data of the scene with potential data perception loss is carried out, and the optimization space search of a learning cognitive analysis model is enhanced. The random gating network mechanism is mainly realized through a GRU time sequence neural network and random gating factors, the values of the random gating factors are calculated by adopting a common random algorithm, and the effect generated by simulated data loss is calculated by adopting equation (1).
Then, aiming at the problem of data perception irregular sampling granularity in a complex situation, a method for any data perception sampling granularity in a continuous time domain is constructed based on a time sequence differential neural network model, and the perception time sequence data in the learning process is assumed to be marked as X = ((t) 0 ,x 0 ),…,(t n ,x n ) Therein), wherein
For learning process data x j Sampling time point of (1), x j Expressed as:
wherein, denotes probability of possible loss of perceptual data, t 0 <…<t n Representing irregular time intervals. According to the time sampling granularity requirement of the cognitive analysis model and the combination of a causal total effect framework, the irregular time sequence micro-neural network model can be described as follows:
H t ＝ODENET(A*H t-1 ,f,θ f ,t begin ,t end ),t＝t 0 ,…,t n , (3)
wherein ODENET represents a time sequence micro-neural network model, A represents a time sequence self-attention mechanism under the constraint of the causal total effect, H represents the hidden state of the data in the learning process, f represents the time sequence model, and theta f Parameters can be learned for a time series model, and any time granularity rule sampling data can be generated by utilizing the model.
And finally, realizing the time domain perception optimization of cognitive analysis in the hybrid learning process by combining a random gating mechanism and an irregular continuous time domain perception method, namely connecting the random gating mechanism and an irregular time domain sampling solution mechanism in a residual error connection mode.
2) Structure robustness optimization of multi-modal cognitive analysis model through hybrid learning
Most deep learning models lack flexibility due to sampling of offline training-online usage patterns, and are particularly inefficient in dynamically changing complex scenes. The method combines a plasticity weight updating mechanism to enhance the elasticity of the structure of the multi-modal cognitive analysis model, and improves the learning cognitive analysis robustness in a complex dynamic scene.
Firstly, analyzing and learning the weight distribution characteristics of a cognitive analysis network structure, and constructing a fluctuation suppression-gain model structure plasticity mechanism aiming at a complex cognitive analysis scene by combining a biological plasticity theory, wherein in a cognitive analysis model, the neuron connection fluctuation quantity of a time sequence encoder and a decoder is described as follows:
wherein x is j (t) represents the activation amount of the postsynaptic neuron at time t, x i (t-1) is the activation amount of the anterior synaptonemal neuron at the t-1 moment,
weights can be learned for the neural network itself, combined>
Is a plastic dynamic weight, p (x) j ) And (4) stabilizing the abnormal fluctuation by combining a plasticity weight updating rule for suppressing a gain distribution probability function.
Then, aiming at the data change characteristic of the learning cognition calculation task in the complex scene, the invention provides a plasticity weight self-adaptive dynamic updating rule which can be described by an equation (5) by combining the uncertain factor causal total effect calculation framework provided by the invention,
wherein epsilon (V) is a fluctuation causal effect measurement function, is used for updating the adaptive control plasticity weight, and is dynamically calculated according to the fluctuation quantity of the connected neuron, and the calculation process is described by equation (5),
wherein, the first and the second end of the pipe are connected with each other,
alpha and gamma are control parameters, d represents intermediate harmonic variable values related to momentum, and influences of direct effects and indirect effects of changes of complex environmental factors are balanced.
And finally, embedding a plasticity dynamic weight framework according to the structural characteristics of the learning cognitive analysis model, mainly embedding the plasticity dynamic weight in a multi-mode feature combined learning stage, enhancing the elasticity of the complex environment adaptation of the analysis model, reducing errors caused by the structural bias of the model, and finishing the structural robustness optimization of the mixed learning multi-mode cognitive analysis model.
3) Cognitive data distribution deviation target domain robustness optimization in complex scene
The paradigm of large-scale pre-training (large model, source domain) + target task fine tuning (small data, target domain) is the mainstream approach of the application of the deep learning technology at present, but the complex cross-time domain hybrid learning cognitive analysis task scene data distribution is dynamic and changeable, and great challenges are brought to the application paradigm. The learning cognition analysis method combines a learning cognition dynamic causal model and a meta-distribution hypothesis to construct a data distribution migration-oriented target domain optimization mechanism, and the learning cognition analysis robustness under a complex scene is enhanced.
Firstly, based on element distribution and dynamic learning cognition causal representation, the invention provides complex scene data generation migration dynamic distribution suitable for learning cognition analysis tasks, and analyzes the potential process of distribution offset of a pre-training data source domain and an application target domain, and the dynamic distribution for controlling data generation migration is recorded as f t . F is carried out on pre-training multi-modal learning process source domain data by utilizing a non-linear independent component analysis method based on generalized contrast learning t The estimation process of (2) is described by equation (7),
wherein, the first and the second end of the pipe are connected with each other,
parameterized function set, <' > or>
Indicates that the dynamic causal source domain positive case analysis in the t-th stage is present>
Indicates a negative case analysis>
For the analysis function, is>
Indicating the expected distribution of k'.
The estimate can then be used
Approximate substitution of f t Performing learning cognitive analysis on a target data domain, and performing target independent component extraction and gain, wherein the independent component after gain estimation can be expressed as,
where D represents the dimension of the data,
the independent component set representing the no gain after the target data field extraction is used in the process>
Fitting the inverse function by using a reversible neural network.
Finally, the gain target domain obtained based on equation (8) is independent into diversity, target domain data enhancement is generated, and target domain empirical risk minimization optimization is carried out based on the enhancement data, optimization loss is described by equation (9),
where Ω (-) is a regular term used for control
Complexity, therefore generated->
Gain information can be provided by fine tuning of target domain tasks, and robustness analysis of cognitive state learning of cognitive data distribution deviation target domain under complex scene is carried out
In summary, the invention provides a causality-effect-based cognition analysis model robustness optimization method in a complex learning scene by combining the causal learning research progress and taking multi-mode learning process data as a basis aiming at the problem that the robustness of the existing mixed learning cognition modeling analysis in the complex scene is insufficient, and the challenging problems of low model analysis effect and the like in the complex education scene are solved.
In alternative embodiments, the functions/acts noted in the block diagrams may occur out of the order noted in the operational illustrations. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality/acts involved. Furthermore, the embodiments presented and described in the flow charts of the present invention are provided by way of example in order to provide a more comprehensive understanding of the technology. The disclosed methods are not limited to the operations and logic flows presented herein. Alternative embodiments are contemplated in which the order of various operations is changed, and in which sub-operations described as part of larger operations are performed independently.
Furthermore, although the present invention is described in the context of functional modules, it should be understood that, unless otherwise indicated to the contrary, one or more of the described functions and/or features may be integrated in a single physical device and/or software module, or one or more functions and/or features may be implemented in separate physical devices or software modules. It will also be understood that a detailed discussion of the actual implementation of each module is not necessary for an understanding of the present invention. Rather, the actual implementation of the various functional modules in the apparatus disclosed herein will be understood within the ordinary skill of an engineer given the nature, function, and interrelationships of the modules. Accordingly, those of ordinary skill in the art will be able to practice the invention as set forth in the claims without undue experimentation. It is also to be understood that the specific concepts disclosed are merely illustrative of and not intended to limit the scope of the invention, which is to be determined from the appended claims along with their full scope of equivalents.
The functions, if implemented in the form of software functional units and sold or used as a stand-alone product, may be stored in a computer readable storage medium. Based on such understanding, the technical solution of the present invention may be embodied in the form of a software product, which is stored in a storage medium and includes instructions for causing a computer device (which may be a personal computer, a server, or a network device) to execute all or part of the steps of the method according to the embodiments of the present invention. And the aforementioned storage medium includes: a U-disk, a removable hard disk, a Read-Only Memory (ROM), a Random Access Memory (RAM), a magnetic disk, or an optical disk, and various media capable of storing program codes.
The logic and/or steps represented in the flowcharts or otherwise described herein, e.g., an ordered listing of executable instructions that can be considered to implement logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device.
More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or more wires, a portable computer diskette (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). Further, the computer-readable medium could even be paper or another suitable medium upon which the program is printed, as the program can be electronically captured, via for instance optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It should be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, any one or combination of the following techniques, which are known in the art, may be used: a discrete logic circuit having a logic gate circuit for implementing a logic function on a data signal, an application specific integrated circuit having an appropriate combinational logic gate circuit, a Programmable Gate Array (PGA), a Field Programmable Gate Array (FPGA), or the like.
In the description of the specification, reference to the description of "one embodiment," "some embodiments," "an example," "a specific example," or "some examples" or the like means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.
While embodiments of the invention have been shown and described, it will be understood by those of ordinary skill in the art that: various changes, modifications, substitutions and alterations can be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the claims and their equivalents.
While the preferred embodiments of the present invention have been illustrated and described, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the invention as defined in the appended claims.
Claims (10)
1. A learning cognition analysis model robustness optimization method based on causal effect is characterized by comprising the following steps:
constructing a cognitive analysis multi-type model in a hybrid learning process;
constructing a learner cognitive dynamic causal network and a causal effect framework;
carrying out robustness optimization according to the cognition analysis multi-type model and the learner cognition dynamic causal network and causal effect framework;
the robustness optimization comprises time domain perception robustness optimization of cognitive analysis in a hybrid learning process, structure robustness optimization of a hybrid learning multi-mode cognitive analysis model and robustness optimization of a cognitive data distribution offset target domain in a complex scene.
2. The method for optimizing the robustness of the learning cognition analysis model based on the causal effect as claimed in claim 1, wherein the constructing the cognition analysis multi-type model of the hybrid learning process comprises:
performing modal feature extraction on the collected multi-modal learning process data; wherein the modal features include video features, audio features, and text features;
performing combined learning on the extracted multi-modal characteristics by adopting a multi-type backbone network and a time sequence network to obtain the global representation of learning cognitive characteristics;
and constructing different cognitive classifiers or regressors aiming at different learning cognitive calculation downstream tasks.
3. The causal effect-based learning cognition analysis model robustness optimization method of claim 1, wherein the constructing of learner cognitive dynamic causal network and causal effect framework comprises:
constructing a learner cognitive cause and effect graph through learning cognitive influence variable detection and cause and effect directed graph construction;
and (3) carrying out an effect averaging or optimizing method by utilizing a common cause and effect calculation method in the field and adopting a multipath effect calculation mode to obtain a cause and effect calculation result.
4. The method as claimed in claim 1, wherein when the robustness optimization is a hybrid learning process cognitive analysis time-domain perceptual robustness optimization, the robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework comprises:
based on a direct total causal effect framework and a learner cognitive dynamic causal network model, a reasonable causal hypothesis of data perception uncertainty factors is provided, and learning cognitive data perception is constructed for anti-fact inference analysis;
aiming at the problem that data are partially lost in the process of hybrid learning possibly, a causal total effect-based guided random gating network mechanism is constructed, potential data perception missing scene intervention data generation is carried out, and learning cognitive analysis model optimization space search is enhanced;
aiming at the problem of irregular data perception sampling granularity under a complex situation, constructing a method for any data perception sampling granularity under a continuous time domain based on a time sequence differential neural network model;
the random gating mechanism and the irregular continuous time domain perception method are combined for use, and the time domain perception optimization of cognitive analysis in the hybrid learning process is achieved.
5. The causal learning cognitive analysis model robustness optimization method of claim 1, wherein when the robustness optimization is a mixed learning multi-modal cognitive analysis model structure robustness optimization, the robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework comprises:
analyzing and learning the weight distribution characteristics of the cognitive analysis network structure, constructing a fluctuation suppression-gain model structure plasticity mechanism aiming at a complex cognitive analysis scene by combining a biological plasticity theory, and determining the neuron connection fluctuation description of a time sequence encoder and a decoder in a cognitive analysis model;
aiming at the data change characteristic of a learning cognitive computation task in a complex scene, determining a plasticity weight self-adaptive dynamic update rule by combining an uncertain factor causal total effect computation framework;
and (4) embedding a structure of plasticity dynamic weight according to the structural characteristics of the learning cognitive analysis model, and completing structural robustness optimization of the mixed learning multi-mode cognitive analysis model.
6. The method as claimed in claim 1, wherein when the robustness optimization is robustness optimization of a cognitive data distribution shift target domain in a complex scene, the robustness optimization according to the cognitive analysis multi-type model and the learner cognitive dynamic causal network and causal effect framework comprises:
determining migration dynamic distribution of complex scene data of a learning cognition analysis task based on element distribution and dynamic learning cognition causal representation, and analyzing a distribution offset potential process of a pre-training data source domain and an application target domain;
obtaining a predicted value of the transferred dynamic distribution, and performing learning cognitive analysis on a target data domain and performing target independent component extraction and gain;
and performing target domain data enhancement generation according to the gain target domain independent component, performing target domain experience risk minimization optimization based on the enhanced data, and completing cognitive state robustness analysis of cognitive data distribution deviation target domain learning under a complex scene.
7. A learning cognition analysis model robustness optimization device based on causal effect is characterized by comprising the following steps:
the first module is used for constructing a cognitive analysis multi-type model in a hybrid learning process;
the second module is used for constructing a learner cognitive dynamic cause-and-effect network and a cause-and-effect framework;
the third module is used for carrying out robustness optimization according to the cognition analysis multi-type model and the learner cognition dynamic causal network and causal effect framework;
the robustness optimization comprises time domain perception robustness optimization of cognitive analysis in a hybrid learning process, structure robustness optimization of a hybrid learning multi-mode cognitive analysis model and robustness optimization of a cognitive data distribution offset target domain in a complex scene.
8. An electronic device comprising a processor and a memory;
the memory is used for storing programs;
the processor executing the program realizes the method of any one of claims 1 to 6.
9. A computer-readable storage medium, characterized in that the storage medium stores a program, which is executed by a processor to implement the method according to any one of claims 1 to 6.
10. A computer program product comprising a computer program, characterized in that the computer program realizes the method according to any of claims 1 to 6 when executed by a processor.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211233184.7A CN115879536A (en) | 2022-10-10 | 2022-10-10 | Learning cognition analysis model robustness optimization method based on causal effect |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211233184.7A CN115879536A (en) | 2022-10-10 | 2022-10-10 | Learning cognition analysis model robustness optimization method based on causal effect |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115879536A true CN115879536A (en) | 2023-03-31 |
Family
ID=85770342
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211233184.7A Pending CN115879536A (en) | 2022-10-10 | 2022-10-10 | Learning cognition analysis model robustness optimization method based on causal effect |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115879536A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116151329A (en) * | 2023-04-23 | 2023-05-23 | 山东师范大学 | Student knowledge state tracking method and system based on inverse fact graph learning |
-
2022
- 2022-10-10 CN CN202211233184.7A patent/CN115879536A/en active Pending
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN116151329A (en) * | 2023-04-23 | 2023-05-23 | 山东师范大学 | Student knowledge state tracking method and system based on inverse fact graph learning |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Duan et al. | Curriculum deepsdf | |
| US11610131B2 (en) | Ensembling of neural network models | |
| KR102532749B1 (en) | Method and apparatus for hierarchical learning of neural networks based on weak supervised learning | |
| US10832138B2 (en) | Method and apparatus for extending neural network | |
| US20220004935A1 (en) | Ensemble learning for deep feature defect detection | |
| JP7325414B2 (en) | Training a First Neural Network Model and a Second Neural Network Model | |
| CN111105008A (en) | Model training method, data recognition method and data recognition device | |
| CN111489365B (en) | Training method of neural network, image processing method and device | |
| JP2016157426A (en) | Neural network training method and apparatus, and recognition method and apparatus | |
| WO2021138092A1 (en) | Artificial neural network architectures based on synaptic connectivity graphs | |
| CN112116090A (en) | Neural network structure searching method and device, computer equipment and storage medium | |
| US11669056B2 (en) | Generation of a control system for a target system | |
| TWI831016B (en) | Machine learning method, machine learning system and non-transitory computer-readable storage medium | |
| CN115879536A (en) | Learning cognition analysis model robustness optimization method based on causal effect | |
| CN114298299A (en) | Model training method, device, equipment and storage medium based on course learning | |
| Júnior et al. | Regional models: A new approach for nonlinear system identification via clustering of the self-organizing map | |
| CN110781978A (en) | Feature processing method and system for machine learning | |
| CN110633728A (en) | Financial signal mining method and system based on Monte Carlo search algorithm | |
| Shan et al. | Trimmed data-driven evolutionary optimization using selective surrogate ensembles | |
| US20240119363A1 (en) | System and process for deconfounded imitation learning | |
| US20230162029A1 (en) | Interactive qualitative-quantitative live labeling for deep learning artificial intelligence | |
| Huisman et al. | Understanding transfer learning and gradient-based meta-learning techniques | |
| Lau | Evaluation of machine learning techniques on reducing computational expense for numerical weather prediction | |
| CN117669650A (en) | Training method and device for reinforcement learning model | |
| Chen et al. | Fast optimal structures generator for parameterized quantum circuits |
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 |