Method and apparatus for material property prediction based on modular deep learning

Abstract

The application provides a material attribute prediction method and device based on modular deep learning, which is applied to a cloud material knowledge sharing platform, the cloud material knowledge sharing platform is loaded with a material knowledge base, and the material knowledge base comprises: a plurality of parameter modules, one parameter module corresponds to one heterogeneous material task, and the heterogeneous material task is used for realizing material attribute prediction; the method comprises the following steps: receiving a target attribute prediction problem of a target material uploaded by a user; querying the material knowledge base based on the target attribute prediction problem, determining a target parameter module corresponding to the target attribute prediction problem, and returning the target parameter module to the user; wherein the target parameter module is used for enabling the user to train a model based on the target parameter module, and performing attribute prediction of the target material based on the trained model. Through the method provided by the application, the most relevant module to a given downstream task is selected from the material knowledge base, and the downstream material attribute prediction task is effectively adapted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of material big data, and in particular to a material attribute prediction method and device based on modular deep learning. BACKGROUND

[0002] Existing deep learning-based material knowledge sharing solutions can be divided into two categories: (1) pre-training a neural network on a certain specific material task and fine-tuning the network on a downstream task; (2) pre-training a neural network skeleton on all source tasks and directly integrating all networks for downstream task fine-tuning. The disadvantages of the prior art are: (1) lack of flexibility. The existing solution shares knowledge from the same source task for each downstream task, which cannot decouple and filter the upstream heterogeneous material knowledge, thus easily causing negative transfer of material knowledge; (2) parameter inefficiency. The existing solution needs to copy the complete pre-training network to realize knowledge sharing. Due to the over-parameterization characteristics of the neural network, the existing solution uses a large space cost to store material knowledge, which is not conducive to the transmission and sharing of knowledge.

[0003] How to effectively realize the knowledge decoupling and sharing between different material attributes is a technical problem to be solved at present. SUMMARY

[0004] The present application provides a material attribute prediction method and device based on modular deep learning to solve the defects in the prior art.

[0005] The present application provides a material attribute prediction method based on modular deep learning, applied to a cloud material knowledge sharing platform, wherein the cloud material knowledge sharing platform loads a material knowledge base, and the material knowledge base comprises: a plurality of parameter modules, one parameter module corresponding to one heterogeneous material task, and the heterogeneous material task being used to realize material attribute prediction.

[0006] The method comprises the following steps:

[0007] Receiving a target attribute prediction problem of a target material uploaded by a user;

[0008] Based on the target attribute prediction problem, querying in the material knowledge base to determine a corresponding target parameter module, and returning the target parameter module to the user; wherein the target parameter module is used to enable the user to train a model based on the target parameter module and perform attribute prediction of the target material based on the trained model.

[0009] According to the material attribute prediction method based on modular deep learning provided by the present application, before the step of receiving a target attribute prediction problem of a target material uploaded by a user, the method further comprises:

[0010] Obtain a pre-training material representation skeleton; wherein the pre-training material representation skeleton is used to mount a parameter module;

[0011] For a plurality of preset heterogeneous material tasks, a corresponding parameter module is designed for each heterogeneous material task and mounted on the pre-training material representation skeleton; wherein the parameter module is used to store the heterogeneous knowledge of the corresponding heterogeneous material task;

[0012] Each parameter module corresponding to each heterogeneous material task is stored to obtain the material knowledge base.

[0013] According to the material attribute prediction method based on modular deep learning provided by the application, for a plurality of preset heterogeneous material tasks, a corresponding parameter module is designed for each heterogeneous material task and mounted on the pre-training material representation skeleton, which comprises:

[0014] For each heterogeneous material task, an initial parameter module corresponding to the heterogeneous material task is generated, and the initial parameter module is mounted on the pre-training material representation skeleton;

[0015] The heterogeneous knowledge of the heterogeneous material task is routed into the corresponding initial parameter module through a routing network to obtain the parameter module corresponding to the heterogeneous material task.

[0016] According to the material attribute prediction method based on modular deep learning provided by the application, the parameter module corresponding to each heterogeneous material task in the plurality of heterogeneous material tasks is stored to obtain the material knowledge base, which comprises:

[0017] The parameter module is jointly trained;

[0018] In the case where the training meets the convergence condition, the trained parameter module is stored to obtain the material knowledge base.

[0019] According to the material attribute prediction method based on modular deep learning provided by the application, the target parameter module corresponding to the target attribute prediction problem is determined by querying the material knowledge base based on the target attribute prediction problem, which comprises:

[0020] Based on the target attribute prediction problem, a plurality of target lightweight parameter sub-modules are obtained by querying the material knowledge base through an adaptive modular assembly AMA algorithm; wherein the combination of the plurality of target lightweight parameter sub-modules is a module combination with the lowest generalization error;

[0021] The plurality of target lightweight parameter sub-modules are assembled and spliced to obtain the target parameter module corresponding to the target attribute prediction problem.

[0022] According to the material attribute prediction method based on modular deep learning provided by the application, the target attribute prediction problem is predicted based on the target attribute prediction problem, and a plurality of target lightweight parameter sub-modules are obtained by querying the material knowledge base through an adaptive modular assembly AMA algorithm.

[0023] The prediction algorithm based on label propagation is calculated to obtain the prediction performance of each parameter module on the target heterogeneous material task corresponding to the target attribute prediction problem.

[0024] Based on the prediction performance of each parameter module, the plurality of target lightweight parameter sub-modules are obtained through modular weight optimization based on the target algorithm, wherein the target algorithm is an optimization algorithm without training.

[0025] According to the material attribute prediction method based on modular deep learning provided by the application, the plurality of target lightweight parameter sub-modules are assembled and spliced to obtain the target parameter module corresponding to the target attribute prediction problem, including:

[0026] The plurality of target lightweight parameter sub-modules are assembled and spliced to the pre-trained material representation skeleton to generate a deep learning model of the target heterogeneous material task corresponding to the target attribute prediction problem.

[0027] The deep learning model is jointly fine-tuned to perform target attribute prediction parameter module based on the jointly fine-tuned deep learning model.

[0028] The application also provides a material attribute prediction device based on modular deep learning, which is applied to a cloud material knowledge sharing platform, and the cloud material knowledge sharing platform loads a material knowledge base, and the material knowledge base includes a plurality of parameter modules, one parameter module corresponding to one heterogeneous material task, and the heterogeneous material task is used for realizing material attribute prediction.

[0029] The application also provides a material attribute prediction device based on modular deep learning, which is applied to a cloud material knowledge sharing platform, and the cloud material knowledge sharing platform loads a material knowledge base, and the material knowledge base includes a plurality of parameter modules, one parameter module corresponding to one heterogeneous material task, and the heterogeneous material task is used for realizing material attribute prediction.

[0030] The receiving module is used for receiving the target attribute prediction problem of the target material uploaded by the user.

[0031] The determining module is used for querying the corresponding target parameter module in the material knowledge base based on the target attribute prediction problem, and returning the target parameter module to the user, wherein the target parameter module is used for enabling the user to train a model based on the target parameter module, and performing attribute prediction of the target material based on the trained model.

[0032] The application further provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, and the processor implements the material attribute prediction method based on modular deep learning according to any one of the above when executing the computer program.

[0033] The application further provides a non-transitory computer-readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the material attribute prediction method based on modular deep learning according to any one of the above.

[0034] The application further provides a computer program product, which includes a computer program, and the computer program is executed by a processor to implement the material attribute prediction method based on modular deep learning according to any one of the above.

[0035] The application provides a material attribute prediction method and device based on modular deep learning, which is applied to a cloud material knowledge sharing platform, the cloud material knowledge sharing platform loads a material knowledge base, the material knowledge base includes: a plurality of parameter modules, one parameter module corresponds to one heterogeneous material task, and the heterogeneous material task is used for realizing material attribute prediction; a target attribute prediction problem of a target material uploaded by a user is received; a corresponding target parameter module is determined based on the target attribute prediction problem in the material knowledge base, and the target parameter module is returned to the user; and the target parameter module is used for enabling the user to train a model based on the target parameter module and performing attribute prediction of the target material based on the trained model. Therefore, the application trains a series of parameter modules containing decoupled material knowledge for data-rich tasks, establishes a material knowledge base, selects the most relevant parameter module from the material knowledge base for a given downstream task, is high in efficiency, and trains a model based on the selected parameter module and fine-tunes the model, effectively adapting to a downstream material attribute prediction task. BRIEF DESCRIPTION OF DRAWINGS

[0036] In order to more clearly illustrate the technical solutions of the present application or prior art, the following will briefly introduce the drawings needed in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor.

[0037] Figure 1 is a flowchart of the material attribute prediction method based on modular deep learning provided by the application.

[0038] Figure 2 is a complete flowchart of the material attribute prediction method based on modular deep learning provided by the application.

[0039] Figure 3 is a structural schematic diagram of a material property prediction device based on modular deep learning provided by the present application.

[0040] Figure 4 is a structural schematic diagram of an electronic device provided by the present application. DETAILED DESCRIPTION

[0041] To make the objectives, technical solutions, and advantages of the present application clearer, the technical solutions in the present application will be described clearly and completely below with reference to the drawings in the present application. Obviously, the described embodiments are some, but not all, of the embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by a person of ordinary skill in the art without creative work fall within the protection scope of the present application.

[0042] Before the method of the present application is described in detail, the terms involved in the present application are first explained and described schematically.

[0043] Modular deep learning: a deep learning method that decomposes complex models into independent sub-modules.

[0044] Pre-training: training a machine learning model on a large-scale dataset to learn a general representation independent of specific tasks.

[0045] Fine-tuning: improving the performance of a pre-trained model by continuing training on a specific task dataset.

[0046] Label propagation: a semi-supervised learning algorithm that propagates labels from labeled data to unlabeled data for classification.

[0047] Genetic algorithm: an optimization algorithm that simulates the natural evolution process, and iteratively optimizes through selection, crossover, and mutation operations.

[0048] The present application is described below. Figures 1-4 A material property prediction method and device based on modular deep learning are described.

[0049] Figure 1 is a flowchart of a material property prediction method based on modular deep learning provided by the present application, as Figure 1 shown, applied to a cloud-based material knowledge sharing platform, the cloud-based material knowledge sharing platform loaded with a material knowledge base, the material knowledge base including: a plurality of parameter modules, one set of parameter modules corresponding to one heterogeneous material task, the heterogeneous material task used to realize the prediction of the properties of materials;

[0050] The method includes the following:

[0051] Step 100, receiving a target attribute prediction problem of a target material uploaded by a user.

[0052] It should be noted that the execution subject of the embodiment of the present application is a cloud material knowledge sharing platform. The platform is used for storing, managing and indexing the uploaded parameter module, which can be a lightweight parameter module, and provides search, screening and download functions, so that users can efficiently find the required parameter module.

[0053] Figure 2 is the complete flowchart of the material attribute prediction method based on modular deep learning provided by the present application. As shown in Figure 2 The method provided by the embodiment of the present application is divided into two parts: modular knowledge decoupling (Modular Knowledge Decoupling for Source Tasks) and adaptive module assembly (Adaptive Module Assembly for Target Tasks).

[0054] For upstream heterogeneous material tasks, the embodiment uses modular deep learning technology to realize knowledge decoupling and storage. Specifically, a general pre-trained material representation backbone is selected, and a certain number of parameter modules are mounted for each upstream task. The heterogeneous knowledge of the upstream task is adaptively routed to different parameter modules through a routing network. It should be noted that a group of parameter modules corresponds to a heterogeneous material task, and each task corresponds to more than one parameter module. After training convergence, these modules containing task knowledge are stored in the material knowledge base.

[0055] Before step 100, the construction process of the material knowledge base provided by the embodiment is first described, which specifically includes:

[0056] Step S100, obtaining a pre-trained material representation backbone; wherein the pre-trained material representation backbone is used to mount parameter modules.

[0057] It should be noted that heterogeneous materials refer to materials prepared by two or more materials with different crystal structures, compositions or morphological characteristics, which have new structures and properties. The pre-trained material representation backbone is the basic part of the deep learning model, which is designed to extract and express the basic features of the material. The pre-trained material representation backbone in the embodiment is used to mount a plurality of parameter modules corresponding to the respective upstream tasks.

[0058] Step S200, for a plurality of preset heterogeneous material tasks, a corresponding parameter module is designed and mounted for each heterogeneous material task on the pre-trained material representation backbone; wherein the parameter module is used to store the heterogeneous knowledge of the corresponding heterogeneous material task.

[0059] Step S200 specifically comprises:

[0060] Step S210, for each heterogeneous material task, an initial parameter module corresponding to the task is generated and mounted on the pre-trained material representation skeleton.

[0061] Specifically, an initial parameter module is generated for each heterogeneous material task and mounted on the pre-trained material representation skeleton, and these parameter modules are used to learn and store the relevant knowledge of the respective task.

[0062] Step S220, the heterogeneous knowledge of the heterogeneous material task is routed into the corresponding initial parameter module through the routing network, to obtain the parameter module corresponding to the heterogeneous material task.

[0063] Specifically, the heterogeneous knowledge of the task is adaptively routed into the corresponding initial parameter module according to the specific requirements of the task, and parameter learning and knowledge storage are performed, so as to obtain a parameter module customized for each task.

[0064] Step S300, the parameter module corresponding to each heterogeneous material task is stored, to obtain the material knowledge base.

[0065] Step S300 specifically comprises:

[0066] Step S310, the parameter modules are jointly trained.

[0067] It should be noted that the pre-trained material representation skeleton needs to be frozen during training, and only the parameter modules are trained, so as to optimize the performance of the entire model on multiple tasks.

[0068] Step S320, in the case where the training meets the convergence condition, the trained parameter modules are stored, to obtain the material knowledge base.

[0069] Specifically, after the training reaches the convergence condition, the trained parameter modules corresponding to each task and their related information are separated from the model and stored, to construct and update the material knowledge base.

[0070] Step 200, based on the target attribute prediction problem, the corresponding target parameter module is determined in the material knowledge base, and the target parameter module is returned to the user; wherein the target parameter module is used to enable the user to train a model based on the target parameter module, and to perform attribute prediction of the target material based on the trained model.

[0071] Specifically, continuing to refer to Figure 2For the assembly of downstream modules, the embodiment designs an AMA (Adaptive Modular Assembly) algorithm without training to select the most relevant parameter modules for each downstream task. The AMA algorithm first uses a label propagation-based prediction algorithm to estimate the prediction performance of each module on the downstream task. Then, the AMA algorithm uses a modular weight optimization based on a genetic algorithm, and adaptively selects the module combination that minimizes the generalization error. After the module assembly, these modules are spliced onto the pre-trained material representation skeleton and jointly fine-tuned. The following will be described in detail.

[0072] Step 200 queries the material knowledge base based on the target property prediction problem to determine the corresponding target parameter module, including:

[0073] Step 210, based on the target property prediction problem, obtains a plurality of target lightweight parameter sub-modules through an adaptive modular assembly AMA algorithm in the material knowledge base; wherein the combination of the plurality of target lightweight parameter sub-modules is: the module combination that minimizes the generalization error.

[0074] Step 210 specifically includes:

[0075] Step 211, based on a label propagation-based prediction algorithm, calculates the prediction performance of each parameter module on the target heterogeneous material task corresponding to the target property prediction problem.

[0076] Specifically, the label propagation algorithm is applied to simulate the performance of the module on the target task according to the label information of the target property prediction problem, and to calculate the prediction performance score of each module.

[0077] Step 212, based on the prediction performance of each parameter module, obtains the plurality of target lightweight parameter sub-modules through a modular weight optimization based on a target algorithm; wherein the target algorithm is a training-free optimization algorithm.

[0078] Specifically, the target algorithm is a training-free optimization algorithm. Taking a genetic algorithm as an example, a modular weight optimization model is constructed based on a genetic algorithm framework, the prediction performance score is taken as the optimization target, and through iterative search and optimization, the combination of the plurality of target lightweight parameter sub-modules that minimizes the overall generalization error is determined.

[0079] Step 220, assembling and splicing the plurality of target lightweight parameter sub-modules to obtain the target parameter module corresponding to the target property prediction problem.

[0080] Step 220 specifically includes:

[0081] Step 221, assembling the plurality of target lightweight parameter submodules, and splicing the assembled plurality of target lightweight parameter submodules to the pre-trained material representation skeleton to generate a deep learning model of a target heterogeneous material task corresponding to the target attribute prediction problem.

[0082] Specifically, the assembled module is combined and spliced to the pre-trained material representation skeleton to construct a deep learning model for the target attribute prediction problem. This model integrates the general feature extraction capability of the pre-trained material representation skeleton and the specific task knowledge of the target parameter module.

[0083] Step 222, jointly fine-tuning the deep learning model to perform target attribute prediction parameter module based on the jointly fine-tuned deep learning model.

[0084] Specifically, the constructed deep learning model is jointly fine-tuned, and the model parameters are further optimized through training on the target data set to improve the accuracy and generalization ability of the model on the target attribute prediction problem. Finally, the jointly fine-tuned deep learning model parameter module is obtained for subsequent material attribute prediction tasks.

[0085] The parameter module of the embodiment of the application is uploaded to the cloud material knowledge sharing platform. When facing a material attribute prediction problem to be solved, the cloud material knowledge sharing platform determines the most relevant knowledge module for the user to download, thereby realizing effective knowledge transfer and accurate attribute prediction.

[0086] The above is a step-by-step description of the material attribute prediction method based on modular deep learning provided by the application. As can be seen from the above description of the steps, according to the material attribute prediction method based on modular deep learning provided by the application, the cloud material knowledge sharing platform is applied, the cloud material knowledge sharing platform loads a material knowledge base, the material knowledge base includes: a plurality of parameter modules, a group of parameter modules corresponding to a heterogeneous material task, the heterogeneous material task is used to realize the attribute prediction of the material; by receiving the target attribute prediction problem of the target material uploaded by the user; based on the target attribute prediction problem, determine the corresponding target parameter module in the material knowledge base, and return the target parameter module to the user; wherein the target parameter module is used to train a model based on the target parameter module, and perform attribute prediction of the target material based on the trained model. As can be seen, the application trains a series of parameter modules containing decoupled material knowledge for data-rich tasks, establishes a material knowledge base, selects the most relevant parameter module for a given downstream task from the material knowledge base, which is efficient; train the model based on the selected parameter module and fine-tune it to effectively adapt to the downstream material attribute prediction task.

[0087] The modular deep learning-based material property prediction device provided by the present application is described below. The modular deep learning-based material property prediction device described below can be referred to in correspondence with the modular deep learning-based material property prediction method described above.

[0088] Figure 3 FIG. 1 is a structural schematic diagram of the modular deep learning-based material property prediction device provided by the present application. As shown in FIG. 1, the modular deep learning-based material property prediction device provided by the present application is applied to a cloud material knowledge sharing platform. The cloud material knowledge sharing platform is loaded with a material knowledge base. The material knowledge base includes a plurality of parameter modules. One parameter module corresponds to one heterogeneous material task. The heterogeneous material task is used to realize the prediction of the properties of materials. Figure 3

[0089] The device includes:

[0090] The receiving module 301 is configured to receive a target property prediction problem of a target material uploaded by a user.

[0091] The determining module 302 is configured to query the material knowledge base based on the target property prediction problem, determine a target parameter module corresponding to the target property prediction problem, and return the target parameter module to the user. The target parameter module is used to enable the user to train a model based on the target parameter module and perform the prediction of the properties of the target material based on the trained model.

[0092] The modular deep learning-based material property prediction device provided by the present application is applied to a cloud material knowledge sharing platform. The cloud material knowledge sharing platform is loaded with a material knowledge base. The material knowledge base includes a plurality of parameter modules. One parameter module corresponds to one heterogeneous material task. The heterogeneous material task is used to realize the prediction of the properties of materials. The target property prediction problem of a target material uploaded by a user is received. The material knowledge base is queried based on the target property prediction problem. A target parameter module corresponding to the target property prediction problem is determined. The target parameter module is returned to the user. The target parameter module is used to enable the user to train a model based on the target parameter module and perform the prediction of the properties of the target material based on the trained model. Therefore, the present application trains a series of parameter modules containing decoupled material knowledge for data-rich tasks, establishes a material knowledge base, selects the parameter module most relevant to a given downstream task from the material knowledge base, and has high efficiency. The model is trained based on the selected parameter module and fine-tuned, which effectively adapts to the downstream material property prediction task.

[0093] Based on the above embodiment, in the present embodiment, the device further includes a library generation module, which is specifically configured to:

[0094] ​The pre-training material representation skeleton is used to mount a parameter module before the receiving user uploads a target attribute prediction problem of a target material; wherein the pre-training material representation skeleton is used to mount a parameter module;

[0095] For a plurality of preset heterogeneous material tasks, a corresponding parameter module is designed for each heterogeneous material task and mounted on the pre-training material representation skeleton; wherein the parameter module is used to store the heterogeneous knowledge of the corresponding heterogeneous material task;

[0096] Each parameter module corresponding to each heterogeneous material task is stored to obtain the material knowledge base.

[0097] Based on the above embodiment, in this embodiment, the library generation module is specifically used for:

[0098] For each heterogeneous material task, an initial parameter module corresponding to the heterogeneous material task is generated, and the initial parameter module is mounted on the pre-training material representation skeleton;

[0099] The heterogeneous knowledge of the heterogeneous material task is routed into the corresponding initial parameter module through a routing network to obtain the parameter module corresponding to the heterogeneous material task.

[0100] Based on the above embodiment, in this embodiment, the library generation module is specifically used for:

[0101] The parameter module is jointly trained;

[0102] In the case of training that meets the convergence condition, the trained parameter module is stored to obtain the material knowledge base.

[0103] Based on the above embodiment, in this embodiment, the determination module 302 is specifically used for:

[0104] Based on the target attribute prediction problem, a plurality of target lightweight parameter sub-modules are obtained by querying the material knowledge base through an adaptive modular assembly AMA algorithm; wherein the combination of the plurality of target lightweight parameter sub-modules is a module combination that minimizes the generalization error;

[0105] The plurality of target lightweight parameter sub-modules are assembled and spliced to obtain a target parameter module corresponding to the target attribute prediction problem.

[0106] Based on the above embodiment, in this embodiment, the determination module 302 is specifically used for:

[0107] Based on a label propagation prediction algorithm, the prediction performance of each parameter module on the target heterogeneous material task corresponding to the target attribute prediction problem is calculated;

[0108] Based on the prediction performance of each parameter module, the plurality of target lightweight parameter sub-modules are obtained through modular weight optimization based on a target algorithm; wherein the target algorithm is an optimization algorithm without training.

[0109] Based on the above embodiments, in this embodiment, the determination module 302 is specifically configured to:

[0110] The plurality of target lightweight parameter sub-modules are assembled and spliced onto the pre-training material representation skeleton to generate a deep learning model of a target heterogeneous material task corresponding to the target attribute prediction problem;

[0111] The deep learning model is jointly fine-tuned to perform target attribute prediction parameter module based on the jointly fine-tuned deep learning model.

[0112] Figure 4 An example of an entity structure diagram of an electronic device is shown in Figure 4 As shown, the electronic device can include a processor (processor) 410, a communications interface (Communications Interface) 420, a memory (memory) 430 and a communication bus 440, wherein the processor 410, the communication interface 420, the memory 430 complete the communication among each other through the communication bus 440. The processor 410 can call the logical instructions in the memory 430 to execute the material attribute prediction method based on modular deep learning, which is applied to a cloud material knowledge sharing platform loaded with a material knowledge base, wherein the material knowledge base includes a plurality of parameter modules, and one parameter module corresponds to one heterogeneous material task, and the heterogeneous material task is used to realize the attribute prediction of the material.

[0113] The method comprises:

[0114] Receiving a target attribute prediction problem of a target material uploaded by a user;

[0115] Based on the target attribute prediction problem, the corresponding target parameter module is determined in the material knowledge base, and the target parameter module is returned to the user; wherein the target parameter module is used to enable the user to train a model based on the target parameter module, and to perform attribute prediction of the target material based on the trained model.

[0116] Further, the logic instructions in the memory 430 described above can be implemented in the form of software functional units and sold or used as standalone products, and can be stored in a computer readable storage medium. Based on such understanding, the technical solutions of the present application essentially or the parts that make contributions to the prior art or parts of the technical solutions can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present application. The aforementioned storage medium includes: a U disk, a mobile hard disk, a read-only memory (ROM, Read-Only Memory), a random access memory (RAM, Random Access Memory), a magnetic disk or an optical disk, and various media that can store program codes.

[0117] In another aspect, the present application also provides a computer program product, which comprises a computer program, the computer program can be stored on a non-transitory computer readable storage medium, and the computer program can be executed by a processor to enable a computer to execute the material attribute prediction method based on modular deep learning provided by the above-mentioned methods, applied to a cloud material knowledge sharing platform loaded with a material knowledge base, the material knowledge base comprising: a plurality of parameter modules, one parameter module corresponding to one heterogeneous material task, the heterogeneous material task being used to realize the attribute prediction of the material;

[0118] The method comprises:

[0119] receiving a target attribute prediction problem of a target material uploaded by a user;

[0120] querying in the material knowledge base based on the target attribute prediction problem, determining a corresponding target parameter module, and returning the target parameter module to the user; wherein the target parameter module is used to enable the user to train a model based on the target parameter module and perform attribute prediction of the target material based on the trained model.

[0121] In yet another aspect, the present application also provides a non-transitory computer readable storage medium having a computer program stored thereon, the computer program being executed by a processor to implement the material attribute prediction method based on modular deep learning provided by the above-mentioned methods, applied to a cloud material knowledge sharing platform loaded with a material knowledge base, the material knowledge base comprising: a plurality of parameter modules, one parameter module corresponding to one heterogeneous material task, the heterogeneous material task being used to realize the attribute prediction of the material;

[0122] The method comprises:

[0123] receiving a target attribute prediction problem of a target material uploaded by a user;

[0124] querying in the material knowledge base based on the target attribute prediction problem, determining a corresponding target parameter module, and returning the target parameter module to the user; wherein the target parameter module is used to enable the user to train a model based on the target parameter module, and to perform attribute prediction of the target material based on the trained model.

[0125] The device embodiments described above are merely illustrative, wherein the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e., can be located in one place, or can be distributed on multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment scheme according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0126] From the above description of the embodiments, those skilled in the art can clearly understand that the embodiments can be realized by means of software plus a necessary general hardware platform, and of course can also be realized by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part that contributes to the prior art, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions to make a computer device (which can be a personal computer, a server, or a network device, etc.) execute the methods described in each embodiment or some part of the embodiments.

[0127] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, and not to limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A method for predicting material properties based on modular deep learning, characterized in that, The material knowledge sharing platform is applied to a cloud-based material knowledge sharing platform. The cloud-based material knowledge sharing platform is loaded with a material knowledge base, which includes multiple sets of parameter modules. Each set of parameter modules corresponds to a heterogeneous material task. The heterogeneous material task is used to predict the properties of the material. The method includes: The problem of predicting the target attributes of target materials uploaded by users; Based on the target attribute prediction problem, a query is performed in the material knowledge base to determine the corresponding target parameter module, and the target parameter module is returned to the user; wherein, the target parameter module is used to: enable the user to train a model based on the target parameter module, and perform attribute prediction of the target material based on the trained model; before receiving the target attribute prediction problem of the target material uploaded by the user, the method further includes: Obtain a pre-trained material representation skeleton; wherein the pre-trained material representation skeleton is used to mount parameter modules; For multiple pre-defined heterogeneous material tasks, a corresponding parameter module is designed and attached to the pre-trained material representation skeleton for each heterogeneous material task; wherein, the parameter module is used to store the heterogeneous knowledge of the corresponding heterogeneous material task; The parameter modules corresponding to each heterogeneous material task are stored to obtain the material knowledge base.

2. The material property prediction method based on modular deep learning according to claim 1, characterized in that, For multiple preset heterogeneous material tasks, the method involves designing and attaching corresponding parameter modules to the pre-trained material representation skeleton for each heterogeneous material task, including: For each heterogeneous material task, a corresponding initial parameter module is generated, and the initial parameter module is attached to the pre-trained material representation skeleton; The heterogeneous knowledge of the heterogeneous material task is routed to the corresponding initial parameter module through a routing network, thereby obtaining the parameter module corresponding to the heterogeneous material task.

3. The material property prediction method based on modular deep learning according to claim 1, characterized in that, The step of storing the parameter modules corresponding to each of the multiple heterogeneous material tasks to obtain the material knowledge base includes: Joint training of the parameter modules; If the training meets the convergence condition, the trained parameter modules are stored to obtain the material knowledge base.

4. The material property prediction method based on modular deep learning according to claim 1, characterized in that, The module for determining the corresponding target parameters by querying the material knowledge base based on the target attribute prediction problem includes: Based on the target property prediction problem, multiple target lightweight parameter sub-modules are obtained by querying the material knowledge base through the adaptive modular assembly (AMA) algorithm; wherein, the combination of the multiple target lightweight parameter sub-modules is the module combination that minimizes the generalization error. The multiple lightweight parameter sub-modules are assembled and spliced ​​together to obtain the target parameter module corresponding to the target attribute prediction problem.

5. The material property prediction method based on modular deep learning according to claim 4, characterized in that, The problem of predicting the target properties involves querying the material knowledge base using an adaptive modular assembly (AMA) algorithm to obtain multiple target lightweight parameter sub-modules, including: The prediction algorithm based on label propagation is used to calculate the prediction performance of each parameter module on the target heterogeneous material task corresponding to the target property prediction problem. Based on the prediction performance of each parameter module, the multiple target lightweight parameter sub-modules are obtained through modular weight optimization based on the target algorithm; wherein, the target algorithm is an optimization algorithm that does not require training.

6. The material property prediction method based on modular deep learning according to claim 4, characterized in that, The assembly and splicing of the multiple lightweight target parameter sub-modules to obtain the target parameter module corresponding to the target attribute prediction problem includes: The multiple target lightweight parameter sub-modules are assembled, and the assembled multiple target lightweight parameter sub-modules are spliced ​​onto the pre-trained material representation skeleton to generate a deep learning model for the target heterogeneous material task corresponding to the target attribute prediction problem; The deep learning model is jointly fine-tuned to perform a target attribute prediction parameter module based on the jointly fine-tuned deep learning model.

7. A material property prediction device based on modular deep learning, characterized in that, The material knowledge sharing platform is applied to a cloud-based material knowledge sharing platform. The cloud-based material knowledge sharing platform is loaded with a material knowledge base, which includes multiple sets of parameter modules. Each set of parameter modules corresponds to a heterogeneous material task. The heterogeneous material task is used to predict the properties of the material. The device includes: The receiving module is used to receive the target attribute prediction problem of the target material uploaded by the user; The determination module is used to query the material knowledge base based on the target property prediction problem, determine the corresponding target parameter module, and return the target parameter module to the user; wherein, the target parameter module is used to enable the user to train a model based on the target parameter module, and to perform property prediction of the target material based on the trained model; The device also includes a library generation module, specifically used for: Before receiving the target attribute prediction problem of the target material uploaded by the user, a pre-trained material representation skeleton is obtained; wherein, the pre-trained material representation skeleton is used to mount the parameter module; For multiple pre-defined heterogeneous material tasks, a corresponding parameter module is designed and attached to the pre-trained material representation skeleton for each heterogeneous material task; wherein, the parameter module is used to store the heterogeneous knowledge of the corresponding heterogeneous material task; The parameter modules corresponding to each heterogeneous material task are stored to obtain the material knowledge base.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the material property prediction method based on modular deep learning as described in any one of claims 1 to 6.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the material property prediction method based on modular deep learning as described in any one of claims 1 to 6.

Images