Modular parametric visual program induction method, apparatus, medium, and product

Abstract

The application relates to the field of visual program generation, and provides a modular parametric visual program induction method, equipment, medium and product. An optimized modular parametric model is obtained by constructing a modular parametric model and training under the cooperation of a hierarchical Monte Carlo tree search algorithm, the to-be-processed data is input into the optimized modular parametric model, the hierarchical Monte Carlo tree search algorithm is combined to process the to-be-processed data, and a result program expression is output. The modular parametric visual program induction method can be used for a complex visual scene and has strong generalization capability. Through the combination of the hierarchical Monte Carlo tree search algorithm, the model training data is augmented in the training stage of the modular parametric model, the training efficiency of the model is improved, and in the test stage, the hierarchical Monte Carlo tree search algorithm is used as an efficient search algorithm, so that the problem of program search space explosion caused by complex program expressions is solved.

Description

Technical Field

[0001] This application relates to the field of visual program generation, and more specifically, to a modular parametric visual program induction method, device, medium, and product. Background Technology

[0002] Using programming languages ​​to describe visual content has many advantages. On the one hand, it simulates the process of humans describing the world, using machine language to programmatically describe the visual content seen. On the other hand, the generated program content can be used by automated design tools (such as CAD) to automatically generate the required models.

[0003] However, current technologies for visual content procedural generation are all based on non-parametric procedural generation, which cannot be applied to complex visual scenes. Specifically, on the one hand, complex visual scenes require more complex visual modeling models; on the other hand, complex visual scenes require more complex procedural descriptions, which lead to an explosion in the procedural search space. How to simultaneously solve the problems of search space explosion, visual procedural induction, and parametric procedural generation is an urgent issue to be addressed. Summary of the Invention

[0004] This application provides a modular parametric visual program induction method, device, medium, and product, aiming to simultaneously solve the problems of search space explosion, visual program induction, and parametric program generation.

[0005] The first aspect of this application provides a modular parametric visual program induction method, including: A modular parametric model is constructed, which consists of multiple parametric sub-models. Different parametric sub-models contain different types of parameters, which are used to describe various attributes of the data. An augmented training dataset is generated based on the hierarchical Monte Carlo tree search algorithm and the basic training dataset. The modular parametric model is trained and optimized based on the training dataset to obtain an optimized modular parametric model, which is composed of optimized parametric sub-models.

[0006] Optionally, a modular parametric visual procedure induction method further includes: Input the data to be processed into the optimized modular parametric model; The optimized modular parametric model, combined with the hierarchical Monte Carlo tree search algorithm, processes the data to be processed and outputs the results in a programmatic expression.

[0007] Optionally, a modular parametric model can be constructed, including: Based on different types of metafunctions in the target domain, the parametric sub-model is constructed, and the parametric sub-model is defined as a sub-neural network corresponding to the different types of parameters; The modular parametric model is a collection of the parametric sub-models corresponding to each of the different types of parameters.

[0008] Optionally, the optimized modular parametric model, in conjunction with the hierarchical Monte Carlo tree search algorithm, processes the data to be processed, and outputs a programmatic expression of the results, including: The optimized modular parametric model uses different types of optimized parametric sub-models to express the data to be processed in a program. Based on the hierarchical Monte Carlo tree search algorithm, the optimized modular parametric model searches for program expressions that conform to the data to be processed as the target program expression; The optimized modular parametric model integrates the target program expression into a result program expression for output.

[0009] Optionally, the sub-neural network is specifically:

[0010] Where f is the metafunction. For the parameters, For the sub-neural network, It is the combination of the current input state and the target output state. This indicates that the sub-neural network is in The parameter prediction below, at this time the corresponding metafunction is ; The set of parameterized sub-models is specifically as follows: ,in, This represents the set of metafunctions within the target domain. Represents the set of metafunctions. Individual functions.

[0011] Optionally, after constructing the modular parametric model, an augmented training dataset is generated based on the hierarchical Monte Carlo tree search algorithm and the basic training dataset, including: The parameterized sub-model provides search guidance for the hierarchical Monte Carlo tree search algorithm based on the corresponding different types of parameters; The hierarchical Monte Carlo tree search algorithm augments the basic training dataset based on the search guide to obtain the augmented training dataset. The hierarchical Monte Carlo tree search algorithm augments the combination of the parameterized sub-models to obtain augmented combinations of parameterized sub-models.

[0012] Optionally, the hierarchical Monte Carlo tree search algorithm augments the basic training dataset based on the search guidance to obtain the augmented training dataset, specifically as follows:

[0013] in, For input and output are Time Program The probability of; For input and output The corresponding function probability; For input and output The corresponding parameter probability; The distance metric function corresponding to the target domain is used in the construction The metafunction that can reduce the target distance during time filtering ; Used to select the option that minimizes the distance to the target when constructing the sub-neural network. .

[0014] Optionally, after generating the training dataset, the modular parametric model is trained and optimized based on the training dataset to obtain an optimized modular parametric model, including: The hierarchical Monte Carlo tree search algorithm searches for suitable combinations of augmented parameterized sub-models sequentially based on the search guidelines. The augmented training dataset is input into the appropriate combination of augmented parametric sub-models to perform modeling and calculate the loss function value; Based on the loss function value, the parameters corresponding to the parametric sub-models in the suitable augmented parametric sub-model combination are optimized to obtain the optimized parametric sub-models; The optimized parameterized sub-models are combined into the optimized modular parameterized model.

[0015] A second aspect of this application provides an electronic device, including a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps in a modular parametric visual program induction method proposed in this application.

[0016] A third aspect of this application provides a computer-readable storage medium storing a computer program / instructions thereon, which, when executed by a processor, implements the steps in a modular parameterized visual program induction method proposed in this application.

[0017] A fourth aspect of this application provides a computer program product, including a computer program / instructions that, when executed by a processor, implement the steps in a modular parametric visual program induction method proposed in this application.

[0018] Beneficial effects: This application provides a modular parametric visual program induction method, device, medium, and product. By constructing a modular parametric model and training it with the aid of a hierarchical Monte Carlo tree search algorithm, an optimized modular parametric model is obtained. The data to be processed is input into the optimized modular parametric model, and the hierarchical Monte Carlo tree search algorithm is used to process the data, outputting a programmatic expression of the result. This method has the following advantages: (1) By constructing and training an optimized modular parametric model, the different types of parameters and parameter functions contained therein can be used for complex visual scenes, and the efficient expression of different types of parameters can handle various models in various scenarios to generate appropriate program expressions, with strong generalization ability.

[0019] (2) By combining hierarchical Monte Carlo tree search algorithm, the training data of the model is augmented during the training phase of the modular parametric model, thereby improving the training efficiency of the model. During the testing phase, it serves as an efficient search algorithm, solving the problem of the program search space explosion caused by complex program expressions. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart of a modular parametric model training method proposed in one embodiment of this application; Figure 2 This is a flowchart illustrating a modular, parametric visual program induction method proposed in one embodiment of this application. Detailed Implementation

[0022] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0023] In related technologies, techniques for generating visual content are all based on non-parametric generation, which cannot be applied to complex visual scenes. Specifically, on the one hand, complex visual scenes require more complex visual modeling models; on the other hand, complex visual scenes require more complex program descriptions, which lead to an explosion in the program search space. How to simultaneously solve the problems of search space explosion, visual program induction, and parametric program generation is an urgent issue to be addressed.

[0024] In view of this, this application proposes a modular parametric visual program induction method. By constructing a modular parametric model and training it with the help of a hierarchical Monte Carlo tree search algorithm, an optimized modular parametric model is obtained. The data to be processed is input into the optimized modular parametric model, and the hierarchical Monte Carlo tree search algorithm is used to process the data to be processed, and a suitable result program expression is output.

[0025] First, the modular parametric model in the modular parametric vision program inductive method is constructed and trained. Figure 1 A flowchart of a modular parametric model training method is shown, such as... Figure 1 As shown, the specific steps of the modular parametric model training method provided in this application embodiment are as follows: S101. Construct a modular parametric model.

[0026] In practice, several different types of parameters are first determined. These parameters describe various attributes of the data. The different types of parameters cover as many fields and directions as possible to characterize the possible attributes of various types of data to be processed. For example, parameters describing the radius of the bottom of a cup, parameters describing the height of a bottle, and parameters describing the area of ​​a box, etc.

[0027] Subsequently, for each of the different types of parameters and different meta-functions in the target domain (e.g., different drawing instructions in the CAD domain), a parametric sub-model corresponding to each of the different types of parameters is constructed. The parametric sub-model is defined as a sub-neural network corresponding to the different types of parameters. Each parametric sub-model contains a specific type of sub-function, and each sub-function corresponds to a type of parameter, allowing different parametric sub-models to correspond to a specific data attribute, thus enabling each parametric sub-model to operate independently.

[0028] The sub-neural network is specifically as follows:

[0029] Where f is the metafunction. For the parameters, For the sub-neural network, It is the combination of the current input state and the target output state. This indicates that the sub-neural network is in The parameter prediction below, at this time the corresponding metafunction is ; The set of parameterized sub-models is specifically as follows: ,in, This represents the set of metafunctions within the target domain. Represents the set of metafunctions. Individual functions.

[0030] The modular parametric model is composed of parametric sub-models corresponding to each of the different types of parameters. The modular parametric model is a collection of parametric sub-models for the target domain, consisting of multiple parametric sub-models, each containing different types of parameters.

[0031] S102. Based on the hierarchical Monte Carlo tree search algorithm and the basic dataset, generate an augmented training dataset.

[0032] In specific implementation, the parameterized sub-model provides search guidance for the hierarchical Monte Carlo tree search algorithm based on the corresponding different types of parameters. Specifically, the different types of parameters in the parameterized sub-model provide prior knowledge as search guidance for the hierarchical Monte Carlo tree search algorithm, thereby narrowing the search space of the hierarchical Monte Carlo tree search algorithm.

[0033] The hierarchical Monte Carlo tree search algorithm, based on the search guidelines, excludes data irrelevant to the target data in the basic training dataset, and performs a directional augmentation of the search space on the basic training dataset to obtain an augmented training dataset. Simultaneously, the hierarchical Monte Carlo tree search algorithm augments the combination of the parametric sub-models, resulting in augmented parametric sub-model combinations.

[0034] The Monte Carlo Tree Search (MCTS) algorithm is a tree-based search method that remains effective even with a large search space. It uses random numbers (or more commonly pseudo-random numbers) to solve many computational problems. It associates the problem with a certain probability model and uses a computer to perform statistical simulation or sampling to obtain an approximate solution. The Monte Carlo search tree focuses on branches that are more worthwhile to search; if a move is good, the Monte Carlo tree will expand it deeply, otherwise it will not. This application, by limiting the scenario of the original Monte Carlo tree, obtains a hierarchical Monte Carlo Tree Search algorithm that can be applied to this modular parametric model. This hierarchical Monte Carlo Tree Search algorithm uses search guidelines to exclude data irrelevant to the target data in the basic training dataset, thereby reducing the dependence of the parametric sub-model on the training data.

[0035] Specifically, in this embodiment of the application, the search space of the basic training dataset is augmented using the following method to obtain the augmented training dataset:

[0036] in, For input and output are Time Program The probability of; For input and output The corresponding function probability; For input and output The corresponding parameter probability; The distance metric function corresponding to the target domain is used in the construction The metafunction that can reduce the target distance during time filtering Used to select the option that minimizes the distance to the target when constructing the sub-neural network. .

[0037] S103. The modular parametric model is trained and optimized based on the augmented training dataset to obtain the optimized modular parametric model.

[0038] In practice, the hierarchical Monte Carlo tree search algorithm searches for suitable combinations of augmented parameterized sub-models (including suitable sub-functions and suitable function parameters) sequentially based on the search guidelines.

[0039] The augmented training dataset is input into the appropriate combination of augmented parametric sub-models for modeling and loss function value is calculated. Then, based on the loss function value, the parameters corresponding to the parametric sub-models in the appropriate combination of augmented parametric sub-models are optimized to obtain the optimized parametric sub-models.

[0040] The optimized parameterized sub-models are combined into the optimized modular parameterized model.

[0041] By constructing a hierarchical Monte Carlo search tree, the function space can be traversed efficiently. Based on the search guide, data that is irrelevant to the target data in the basic training dataset can be eliminated. This allows the hierarchical construction of the Monte Carlo search tree to narrow the search space during the search process, thereby solving the problem of modeling diverse programs and effectively addressing the additional challenges to model learning caused by the diversity of program expressions during training.

[0042] After constructing and optimizing the modular parametric model to obtain the optimized modular parametric model, the testing phase begins. Figure 2 A flowchart illustrating the use of a modular, parametric visual procedural inductive method is shown, such as... Figure 2 As shown, the specific steps are as follows: S201. Input the data to be processed into the optimized modular parametric model.

[0043] In practice, the data to be processed is input into the optimized modular parametric model. The data to be processed can be visual data such as image data and 3D model data. This application does not impose any specific restrictions on this.

[0044] S202. In the optimized modular parametric model, different types of optimized parametric sub-models are used to express the data to be processed in a program.

[0045] In practice, based on the attributes corresponding to the different types of parameters contained in the optimized parametric sub-model, an attempt is made to express these attributes of the input data to be processed in a programmatic manner. For example, if parameter 'a' is used to describe the radius of the cup's bottom surface in the data, then the parametric sub-model containing parameter 'a' will attempt to output a programmatic expression about the radius of the cup's bottom surface in the data to be processed.

[0046] S203. Based on the hierarchical Monte Carlo tree search algorithm, the optimized modular parametric model search program expression that matches the data to be processed is taken as the target program expression.

[0047] Since the attributes represented by the various parameterized sub-models contained in the optimized modular parameterized model may or may not be related to the data to be processed, in order to effectively reduce the search space and solve the problem of program diversity modeling, the hierarchical Monte Carlo tree search algorithm will efficiently traverse all program expressions obtained from the parameterized sub-models, search for program expressions related to the data to be processed, and use them as the target program expressions for the data to be processed; while program expressions unrelated to the data to be processed will be excluded from the function space.

[0048] S204. The optimized modular parametric model integrates the target program expression into the result program expression for output.

[0049] In practice, all target program expressions that match the data to be processed and efficiently searched based on the hierarchical Monte Carlo tree search algorithm are integrated into a result program expression, which is then used as the output of the algorithm.

[0050] The following section will elaborate on the above-mentioned modular parametric vision program induction method with specific examples.

[0051] The optimized modular parametric model contains parametric sub-models corresponding to different types of parameters. These different types of parameters cover as many fields and directions as possible, and are used to characterize the possible attributes of various types of data to be processed. For example, there are many different types of parametric sub-models, such as sub-model A corresponding to parameter 'a' describing the radius of the bottom of a cup, sub-model B corresponding to parameter 'b' describing the height of a bottle, sub-model C corresponding to parameter 'c' describing the area of ​​a box, sub-model D corresponding to parameter 'd' describing the exposure value, sub-model E corresponding to parameter 'e' describing the color temperature of a plant, sub-model F corresponding to parameter 'f' describing the chromaticity of a flower, and sub-model G corresponding to parameter 'g' describing the height of a wall.

[0052] Input a picture of a garden building into the optimized modular parametric model. Based on the attributes corresponding to the different types of parameters contained in the optimized parametric sub-model, attempt to programmatically express these attributes of the input landscape picture. For example, parametric sub-model A attempts to programmatically express the radius of the bottom of the cup in the landscape picture, parametric sub-model B attempts to programmatically express the height of the bottle in the landscape picture, parametric model C attempts to programmatically express the area of ​​the box in the landscape picture, parametric model D attempts to programmatically express the exposure value in the landscape picture, parametric model E attempts to programmatically express the color temperature of the plants in the landscape picture, and so on for other parametric sub-models.

[0053] The hierarchical Monte Carlo tree search algorithm combines and efficiently traverses all program expressions obtained from the parameterized sub-models, searching for program expressions related to the information in the garden architecture image, such as plant color temperature, flower chroma, exposure value, etc. The program expressions output by the parameterized sub-models are used as target program expressions; while program expressions unrelated to the information in the garden architecture image, such as cup bottom radius, bottle height, box area, etc., are excluded from the function space as invalid program expressions.

[0054] The optimized modular parametric model integrates the target program expression into a result program expression, which serves as the algorithm's output. This yields a set of result program expressions for the garden architecture image, output by the modular parametric visual program induction method. This result program expression can be directly used as a programming language in drawing software (such as CAD). Inputting this result program expression into the drawing software can automatically generate the required model or image.

[0055] This application provides a modular parametric visual program induction method, device, medium, and product. By constructing a modular parametric model and training it with the aid of a hierarchical Monte Carlo tree search algorithm, an optimized modular parametric model is obtained. The data to be processed is input into the optimized modular parametric model, and the hierarchical Monte Carlo tree search algorithm is used to process the data, outputting a programmatic expression of the result. This method has the following advantages: (1) By constructing and training an optimized modular parametric model, the different types of parameters and parameter functions contained therein can be used for complex visual scenes, and the efficient expression of different types of parameters can handle various models in various scenarios to generate appropriate program expressions, with strong generalization.

[0056] (2) By combining hierarchical Monte Carlo tree search algorithm, the training data of the model is augmented during the training phase of the modular parametric model, thereby improving the training efficiency of the model. During the testing phase, it serves as an efficient search algorithm, solving the problem of the program search space explosion caused by complex program expressions.

[0057] Based on the same inventive concept, embodiments of this application provide an electronic device, including a memory, a processor, and a computer program stored in the memory. The processor executes the computer program to implement the steps in a modular parametric visual program induction method proposed in embodiments of this application.

[0058] In another embodiment provided in this application, a computer-readable storage medium is also provided, on which a computer program / instruction is stored, which, when executed by a processor, implements the steps in a modular parameterized visual program induction method proposed in the embodiments of this application.

[0059] In another embodiment provided in this application, a computer program product is also provided, including a computer program / instructions that, when executed by a processor, implement the steps in a modular parameterized visual program induction method proposed in the embodiments of this application.

[0060] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially as a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid state disk (SSD)).

[0061] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0062] The various embodiments in this specification are described in a related manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0063] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand this application, and the content of this specification should not be construed as a limitation of this application. Furthermore, for those skilled in the art, there will be different forms of changes in the specific implementation methods and application scope based on this application. It is neither necessary nor possible to exhaustively list all implementation methods here, and obvious changes or modifications derived therefrom are still within the protection scope of this application.

Claims

1. A modular parametric visual program induction method, characterized in that, include: A modular parametric model is constructed, which consists of multiple parametric sub-models. Different parametric sub-models contain different types of parameters, which are used to describe various attributes of the data. Based on the hierarchical Monte Carlo Tree Search algorithm and the basic training dataset, an augmented training dataset is generated. Different types of parameters in the parametric sub-models provide prior knowledge as search guidance for the hierarchical Monte Carlo Tree Search algorithm, narrowing its search space. The hierarchical Monte Carlo Tree Search algorithm, based on this search guidance, eliminates data unrelated to the target data in the basic training dataset, thereby reducing the dependence of the parametric sub-models on the training data and performing targeted augmentation of the search space on the basic training dataset to obtain the augmented training dataset. The modular parametric model is then trained and optimized based on the augmented training dataset to obtain an optimized modular parametric model, which is composed of optimized parametric sub-models. The data to be processed is input into the optimized modular parametric model, wherein the data to be processed is visual data of the type such as image data and 3D model data; In the optimized modular parametric model, different types of optimized parametric sub-models are used to express the data to be processed in a program. Based on the hierarchical Monte Carlo tree search algorithm, the optimized modular parametric model search program expression that matches the data to be processed is taken as the target program expression. In order to effectively reduce the search space and model program diversity, the hierarchical Monte Carlo tree search algorithm will efficiently traverse all program expressions obtained from the parametric sub-models to search for program expressions related to the data to be processed, which will be taken as the target program expression of the data to be processed. Program expressions that are not related to the data to be processed will be excluded from the function space. The optimized modular parametric model integrates the target program expression into the result program expression, which is used as the output of the algorithm. This result program expression can be directly used as a programming language in drawing software. By inputting the result program expression into the drawing software, the required model or image can be automatically generated. After constructing the modular parametric model, an augmented training dataset is generated based on the hierarchical Monte Carlo tree search algorithm and the basic training dataset, including: The parameterized sub-model provides search guidance for the hierarchical Monte Carlo tree search algorithm based on the corresponding different types of parameters; The hierarchical Monte Carlo tree search algorithm augments the basic training dataset based on the search guidance to obtain the augmented training dataset, specifically as follows: in, For input and output are Time Program The probability of; For input and output The corresponding function probability; For input and output The corresponding parameter probability; The distance metric function corresponding to the target domain is used in the construction Time-based filtering can reduce the target distance metafunction Used to select the option that minimizes the distance to the target when constructing a sub-neural network. ; The hierarchical Monte Carlo tree search algorithm augments the combination of the parameterized sub-models to obtain augmented combinations of parameterized sub-models.

2. The modular parametric visual program induction method according to claim 1, characterized in that, Constructing a modular parametric model includes: Based on different types of metafunctions in the target domain, the parametric sub-model is constructed, and the parametric sub-model is defined as a sub-neural network corresponding to the different types of parameters; The modular parametric model is a collection of the parametric sub-models corresponding to each of the different types of parameters.

3. The modular parametric visual program induction method according to claim 1, characterized in that, The optimized modular parametric model, combined with the hierarchical Monte Carlo tree search algorithm, processes the data to be processed, and outputs a programmatic expression of the results, including: The optimized modular parametric model uses different types of optimized parametric sub-models to express the data to be processed in a program. Based on the hierarchical Monte Carlo tree search algorithm, the optimized modular parametric model searches for program expressions that conform to the data to be processed as the target program expression; The optimized modular parametric model integrates the target program expression into a result program expression for output.

4. The modular parametric visual program induction method according to claim 2, characterized in that, The sub-neural network is specifically as follows: Where f is the metafunction. For the parameters, For the sub-neural network, It is the combination of the current input state and the target output state. This indicates that the sub-neural network is in The parameter prediction below, at this time the corresponding metafunction is ; The set of parameterized sub-models is specifically as follows: ,in, This represents the set of metafunctions within the target domain. Represents the set of metafunctions. Individual functions.

5. The modular parametric visual program induction method according to claim 1, characterized in that, After generating the augmented training dataset, the modular parametric model is trained and optimized based on the augmented training dataset to obtain the optimized modular parametric model, including: The hierarchical Monte Carlo tree search algorithm searches for suitable combinations of augmented parameterized sub-models sequentially based on the search guidelines. The augmented training dataset is input into the appropriate combination of augmented parametric sub-models to perform modeling and calculate the loss function value; Based on the loss function value, the parameters corresponding to the parametric sub-models in the suitable augmented parametric sub-model combination are optimized to obtain the optimized parametric sub-models; The optimized parameterized sub-models are combined into the optimized modular parameterized model.

6. An electronic device, characterized in that, It includes a memory, a processor, and a computer program stored in the memory, wherein the processor executes the computer program to implement the steps in the modular parametric visual program induction method according to any one of claims 1-4.

7. A computer-readable storage medium having a computer program / instructions stored thereon, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps in the modular parametric visual program induction method described in any one of claims 1-4.

8. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps in the modular parametric visual program induction method described in any one of claims 1-4.

Images