Model forgetting learning method and device, computer equipment and storage medium

Abstract

The embodiment of the invention provides a model forgetting learning method and device, computer equipment and a storage medium, and belongs to the technical field of artificial intelligence. Comprising the steps of determining target data needing to be deleted by an object and a target data category corresponding to the target data; acquiring sample data corresponding to the object, wherein the sample data comprises a plurality of sub-sample data of which the forgetting probabilities are ranked from low to high; determining a data category corresponding to each piece of sub-sample data, and determining the similarity between the target data category and the data category corresponding to each piece of sub-sample data; target sub-sample data with the highest similarity is determined from the multiple pieces of sub-sample data, and target model parameters of a target model corresponding to the target sub-sample data are determined; and obtaining new sample data, and performing forgetting learning training on the target model based on the target model parameters and the new sample data to obtain a target model after forgetting learning training. The forgetting learning training is carried out from the target model, so that the forgetting learning training efficiency of the model is improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a model forgetting learning method, device, computer equipment and storage medium. Background Art

[0002] With the development of artificial intelligence technology, in some application scenarios, artificial intelligence models can be trained based on user preference data, and then personalized services can be provided to the user through the trained artificial intelligence models, such as personalized content recommendations.

[0003] However, after using personalized services for a period of time, users may choose to reset some preference data. In this case, this preference data needs to be cleared, and the previously trained artificial intelligence model also needs to continue to provide personalized services. Therefore, the previously trained artificial intelligence model needs to be trained through forgetting learning.

[0004] In the related technology, the model parameters of the previously trained artificial intelligence model are directly reset to make the artificial intelligence model return to its initial state, and then the corresponding training data is used to retrain the artificial intelligence model in the initial state to achieve forgetting learning of the artificial intelligence model. However, this will waste a lot of computing power and time, resulting in low training efficiency of the artificial intelligence model. Summary of the invention

[0005] The main purpose of the embodiments of the present application is to propose a model forgetting learning method, apparatus, computer equipment and storage medium, which can improve the training efficiency of forgetting learning of artificial intelligence models.

[0006] To achieve the above-mentioned purpose, a first aspect of an embodiment of the present application proposes a model forgetting learning method, the method comprising:

[0007] Determine the target data to be deleted and the target data category corresponding to the target data;

[0008] Acquire sample data corresponding to the object, wherein the sample data includes a plurality of sub-sample data arranged in ascending order of forgetting probability;

[0009] Determine the data category corresponding to each sub-sample data, and determine the similarity between the target data category and the data category corresponding to each sub-sample data;

[0010] Determine the target sub-sample data with the highest similarity among the multiple sub-sample data, and determine the target model parameters of the target model corresponding to the target sub-sample data;

[0011] New sample data is acquired, and forgetting learning training is performed on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training.

[0012] To achieve the above-mentioned purpose, a second aspect of an embodiment of the present application proposes a model forgetting learning device, the device comprising:

[0013] A first determination module, used to determine target data to be deleted from an object and a target data category corresponding to the target data;

[0014] An acquisition module, used for acquiring sample data corresponding to the object, wherein the sample data includes a plurality of sub-sample data arranged in ascending order of forgetting probability;

[0015] A second determination module, used to determine the data category corresponding to each sub-sample data, and determine the similarity between the target data category and the data category corresponding to each sub-sample data;

[0016] A comparison module, used to determine the target sub-sample data with the highest similarity among the multiple sub-sample data, and determine the target model parameters of the target model corresponding to the target sub-sample data;

[0017] The training module is used to obtain new sample data and perform forgetting learning training on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training.

[0018] In some embodiments, the model forgetting learning device further includes a filtering module, wherein the filtering module is used to obtain feature data corresponding to the object before obtaining sample data corresponding to the object, and to segment the feature data to obtain a plurality of sub-feature data;

[0019] The information entropy corresponding to each sub-feature data is determined, and sub-sample data having the information entropy higher than a preset information entropy is screened out from the plurality of sub-feature data.

[0020] The acquisition module is further used to determine the forgetting probability corresponding to each sub-sample data;

[0021] The sample data corresponding to the object is obtained according to each sub-sample data arranged from low to high in the forgetting probability.

[0022] In some embodiments, the model forgetting learning device further includes a pre-training module, wherein the pre-training module is used to obtain a basic model before determining the target sub-sample data with the highest similarity among the multiple sub-sample data, and select the sub-sample data that is first arranged in the sample data as the current sub-sample data;

[0023] Iteratively training the basic model according to the current sub-sample data to obtain an iterative model, and saving model parameters of the iterative model;

[0024] Selecting sub-sample data arranged after the current sub-sample data from the sample data, and determining the sub-sample data arranged after the current sub-sample data as the current sub-sample data;

[0025] Iteratively train the iterative model according to the current sub-sample data to obtain an updated iterative model, and save the model parameters of the updated iterative model, and return to execute the selection of sub-sample data arranged after the current sub-sample data in the sample data until the iterative model corresponding to the last sub-sample data arranged in the sample data and the model parameters of the iterative model are obtained.

[0026] In some implementations, the comparison module is used to determine the iterative model corresponding to the target sub-sample data from among all the saved iterative models;

[0027] The iterative model corresponding to the target sub-sample data is determined as the target model, and the model parameters of the iterative model corresponding to the target sub-sample data are determined as the target model parameters of the target model.

[0028] In some embodiments, the second determination module further includes a determination submodule and an input submodule, the determination submodule being configured to determine a pre-trained single classification model corresponding to the target data category;

[0029] The input submodule is used to input each sub-sample data into the pre-trained single classification model, and output the similarity between the target data category and the data category corresponding to each sub-sample data.

[0030] In some embodiments, the input submodule is used to input each sub-sample data into each pre-trained single-classification model when there are multiple pre-trained single-classification models, and obtain the similarity between the target data category output by each pre-trained single-classification model and the data category corresponding to each sub-sample data.

[0031] The comparison module is used to compare the similarity between the target data category output by each pre-trained single classification model and the data category corresponding to each sub-sample data to determine the target sub-sample data with the highest similarity.

[0032] In some implementations, a training module is used to obtain a label value corresponding to the new sample data;

[0033] Inputting the new sample data into the target model, and outputting the predicted value of the new sample data;

[0034] A loss value between the label value and the predicted value is determined, and the target model parameters are adjusted based on the loss value to obtain a target model after forgetting learning training.

[0035] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes a computer-readable storage medium, which stores multiple instructions, and the instructions are suitable for a processor to load to execute the model forgetting learning method provided in the embodiment of the present application.

[0036] To achieve the above-mentioned purpose, a computer device includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the model forgetting learning method of the embodiment of the present application when executing the computer program.

[0037] The model forgetting learning method, device, computer equipment and storage medium proposed in the present application, by determining the target data to be deleted by the object and the target data category corresponding to the target data; obtaining the sample data corresponding to the object, the sample data contains multiple sub-sample data arranged from low to high probability of forgetting; determining the data category corresponding to each sub-sample data, and determining the similarity between the target data category and the data category corresponding to each sub-sample data; determining the target sub-sample data with the highest similarity among the multiple sub-sample data, and determining the target model parameters of the target model corresponding to the target sub-sample data; obtaining new sample data, and performing forgetting learning training on the target model based on the target model parameters and the new sample data, to obtain the target model after forgetting learning training. In this way, by determining the target data to be deleted by the object, and then determining the target sub-sample data with the highest similarity to the target data among the multiple sub-sample data with the forgetting probability from low to high, and then starting the forgetting learning training from the target model corresponding to the target sub-sample data, compared with the related art of re-training the model from the beginning for forgetting learning, the present application directly starts the forgetting learning training from the target model, which greatly improves the efficiency of the forgetting learning training of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] Figure 1 It is a scenario diagram of the model forgetting learning method provided in the embodiment of the present application;

[0039] Figure 2 It is a flowchart of the model forgetting learning method provided in the embodiment of the present application;

[0040] Figure 3 It is a schematic diagram of the process of training the basic model provided in an embodiment of the present application;

[0041] Figure 4is another flow chart of the model forgetting learning method provided in an embodiment of the present application;

[0042] Figure 5 is a schematic diagram of the structure of a model forgetting learning device provided in an embodiment of the present application;

[0043] Figure 6 It is a schematic diagram of the hardware structure of the computer device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0045] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0046] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0047] First, some nouns involved in this application are analyzed:

[0048] Artificial intelligence (AI) is a new technical science that studies and develops theories, methods, technologies and application systems for simulating, extending and expanding human intelligence. AI is a branch of computer science that attempts to understand the essence of intelligence and produce a new type of intelligent machine that can respond in a similar way to human intelligence. Research in this field includes robotics, language recognition, image recognition, natural language processing and expert systems. AI can simulate the information process of human consciousness and thinking. AI is also a theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0049] Forget Learning: As the requirements for data privacy protection become increasingly stringent, forget learning can be used to remove the influence of specific data from a trained model. When certain data is determined to be sensitive information or needs to be removed, forget learning can help the model adjust parameters so that these specific data no longer affect the output of the model, thereby protecting data privacy.

[0050] For example, in the medical field, if an AI diagnostic model trained on a large amount of patient data needs to have its data deleted to meet privacy regulations, forgetting learning can ensure that the model can still accurately diagnose other patients after deleting this data, without leaking any information about the deleted data.

[0051] Secondly, describe the technical problems existing in related technologies:

[0052] With the development of artificial intelligence technology, in some application scenarios, artificial intelligence models can be trained based on user preference data, and then personalized services can be provided to the user through the trained artificial intelligence models, such as personalized content recommendations.

[0053] However, after using personalized services for a period of time, users may choose to reset some preference data. In this case, this preference data needs to be cleared, and the previously trained artificial intelligence model also needs to continue to provide personalized services. Therefore, the previously trained artificial intelligence model needs to be trained through forgetting learning.

[0054] In the related technology, the model parameters of the previously trained artificial intelligence model are directly reset to make the artificial intelligence model return to its initial state, and then the corresponding training data is used to retrain the artificial intelligence model in the initial state to achieve forgetting learning of the artificial intelligence model. However, this will waste a lot of computing power and time, resulting in low training efficiency of the artificial intelligence model.

[0055] In order to solve the above technical problems, in this application, the target data that needs to be deleted is determined, and then the target sub-sample data with the highest similarity to the target data is determined from multiple sub-sample data with forgetting probabilities from low to high, and then forgetting learning training is started from the target model corresponding to the target sub-sample data. Compared with the related art of re-training the model from scratch in forgetting learning, the forgetting learning training is directly started from the target model in this application, which greatly improves the efficiency of the forgetting learning training of the model.

[0056] Specifically, the embodiments of the present application provide a model forgetting learning method, apparatus, computer equipment and storage medium, which can improve the training efficiency of forgetting learning of artificial intelligence models.

[0057] See also Figure 1 , Figure 1 Schematic diagram of the model forgetting learning method provided in the embodiment of the present application. The schematic diagram of the scenario can be used to roughly understand the model forgetting learning method in the embodiment of the present application.

[0058] Among them, when the subject uses some services, he or she will authorize personal data to be provided to the artificial intelligence model. The artificial intelligence model can be trained based on the personal data to provide personalized services to the subject, such as personalized content recommendations, personalized product recommendations, etc.

[0059] After obtaining the personal data, the personal data of each object can be determined as the characteristic data of the object, and then the characteristic data of each object can be segmented to obtain multiple sub-characteristic data. For example, the data categories corresponding to different data in the characteristic data can be determined first. For example, some data in the characteristic data is browsing data, some data is click data, and some data is search data, etc., so there are multiple data categories. The characteristic data can be segmented according to the data category to obtain multiple sub-characteristic data.

[0060] After obtaining multiple sub-feature data of the object, some sub-feature data with large amounts of information can be selected as sub-sample data for subsequent model training. For example, the information entropy of each sub-feature data can be determined. The larger the information entropy, the greater the amount of information in the sub-feature data. Sub-sample data with information entropy greater than the preset information entropy can be selected from multiple sub-feature data based on the information entropy.

[0061] Then determine the forgetting probability corresponding to each sub-sample data, and get the sample data corresponding to the object according to each sub-sample data arranged from low to high according to the forgetting probability. For example, the more relevant the sub-sample data is to the object, the greater the forgetting probability of the sub-sample data. Take the browsing data category and the click data category as an example. The sub-sample data corresponding to the click data category is more relevant to the object behavior, so the forgetting probability of the sub-sample data corresponding to the click data category is higher. Compared with the click data category, the browsing data category has a lower correlation with the object behavior, so the forgetting probability of the sub-sample data corresponding to the browsing data category is lower.

[0062] After obtaining the sample data, the basic model can be trained from the first sub-sample data to iterate the basic model and obtain an iterative model. Then the iterative model is trained with the second sub-sample data to obtain an updated iterative model. Similarly, the model can be trained in sequence with sub-sample data with high to low forgetting probabilities, thereby continuously iterating the model. In each iteration, the model parameters of the iterative model corresponding to each sub-sample data are saved.

[0063] It is understandable that the object can delete some target data according to its own needs, and the user can delete the click data. At this time, the trained model needs to be trained for forgetting learning. Specifically, by determining the target data that the object needs to delete and the target data category corresponding to the target data; obtaining the sample data corresponding to the object, the sample data contains multiple sub-sample data arranged from low to high probability of forgetting; determining the data category corresponding to each sub-sample data, and determining the similarity between the target data category and the data category corresponding to each sub-sample data; determining the target sub-sample data with the highest similarity among the multiple sub-sample data, and determining the target model parameters of the target model corresponding to the target sub-sample data; obtaining new sample data, and performing forgetting learning training on the target model based on the target model parameters and the new sample data, to obtain the target model after forgetting learning training. In this way, by determining the target data that needs to be deleted from the object, and then determining the target sub-sample data with the highest similarity to the target data from multiple sub-sample data with forgetting probabilities from low to high, and then starting the forgetting learning training from the target model corresponding to the target sub-sample data, compared with the related art of re-training the model from scratch, the present application directly starts the forgetting learning training from the target model, which greatly improves the efficiency of the forgetting learning training of the model.

[0064] The model forgetting learning method, apparatus, computer equipment and storage medium provided in the embodiments of the present application are specifically illustrated through the following embodiments. First, the model forgetting learning method in the embodiments of the present application is described.

[0065] The embodiments of the present application can acquire and process relevant data based on artificial intelligence technology. Among them, artificial intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0066] AI basic technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics, etc. AI software technologies mainly include computer vision technology, robotics technology, biometrics technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0067] The model forgetting learning method provided in the embodiment of the present application relates to the field of artificial intelligence technology. The model forgetting learning method provided in the embodiment of the present application can be applied to the terminal, can also be applied to the server side, and can also be software running in the terminal or the server side. In some embodiments, the terminal can be a smart phone, a tablet computer, a laptop computer, a desktop computer, etc.; the server side can be configured as an independent physical server, or it can be configured as a server cluster or distributed system composed of multiple physical servers, and can also be configured as a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN, and big data and artificial intelligence platforms; the software can be an application that implements the model forgetting learning method, etc., but is not limited to the above forms.

[0068] The present application can be used in many general or special computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer computer devices, network PCs, minicomputers, mainframe computers, distributed computing environments including any of the above systems or devices, etc. The present application can be described in the general context of computer executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. The present application can also be practiced in distributed computing environments, in which tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media including storage devices.

[0069] It should be noted that in each specific implementation of the present application, when it comes to the need to perform relevant processing based on data related to user identity or characteristics such as user information, user behavior data, user historical data, and user location information, the user's permission or consent will be obtained first, and the collection, use, and processing of these data will comply with relevant laws, regulations, and standards. In addition, when the embodiment of the present application needs to obtain the user's sensitive personal information, the user's separate permission or consent will be obtained through a pop-up window or by jumping to a confirmation page. After clearly obtaining the user's separate permission or consent, the necessary user-related data for the normal operation of the embodiment of the present application will be obtained.

[0070] See also Figure 2 , Figure 2 : is a flow chart of the model forgetting learning method provided in the embodiment of the present application. The model forgetting learning method may include the following steps:

[0071] Step 110: Determine the target data to be deleted and the target data category corresponding to the target data;

[0072] Step 120: Obtain sample data corresponding to the object, where the sample data includes a plurality of sub-sample data arranged in ascending order of forgetting probability;

[0073] Step 130: determine the data category corresponding to each sub-sample data, and determine the similarity between the target data category and the data category corresponding to each sub-sample data;

[0074] Step 140: Determine the target sub-sample data with the highest similarity among the multiple sub-sample data, and determine the target model parameters of the target model corresponding to the target sub-sample data;

[0075] Step 150: Acquire new sample data, and perform forgetting learning training on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training.

[0076] It is understandable that in the embodiment of the present application, the purpose is to implement forgetting learning training for the trained model, that is, to retrain the trained model. Before introducing steps 110 to 150 in detail, it is first described how to perform the first stage of training for the base model, that is, the training process before forgetting learning training for the target model.

[0077] In some implementations, before obtaining sample data corresponding to the object, the process includes:

[0078] (1.1) Obtaining feature data corresponding to the object, and segmenting the feature data to obtain multiple sub-feature data;

[0079] (1.2) Determine the information entropy corresponding to each sub-feature data, and select sub-sample data with information entropy higher than a preset information entropy from multiple sub-feature data.

[0080] Among them, when the subject uses some services, he or she will authorize personal data to be provided to the artificial intelligence model. The artificial intelligence model can be trained based on the personal data to provide personalized services to the subject, such as personalized content recommendations, personalized product recommendations, etc.

[0081] After obtaining the personal data, the personal data of each object can be determined as the characteristic data of the object, and then the characteristic data of each object can be segmented to obtain multiple sub-characteristic data. For example, the data categories corresponding to different data in the characteristic data can be determined first. For example, some data in the characteristic data is browsing data, some data is click data, and some data is search data, etc., so there are multiple data categories. The characteristic data can be segmented according to the data category to obtain multiple sub-characteristic data.

[0082] After obtaining multiple sub-feature data of the object, some sub-feature data with large amounts of information can be selected as sub-sample data for subsequent model training. For example, the information entropy of each sub-feature data can be determined. The larger the information entropy, the greater the amount of information in the sub-feature data. Sub-sample data with information entropy greater than the preset information entropy can be selected from multiple sub-feature data based on the information entropy.

[0083] Among them, the information entropy of the sub-feature data can be determined by determining the joint entropy of the sub-feature data, and then the joint entropy is determined as the information entropy of the sub-feature data. The larger the information entropy, the greater the amount of information in the sub-feature data, and the smaller the information entropy, the smaller the amount of information in the sub-feature data. The specific calculation method of the joint entropy is as follows:

[0084] Among them, H(S ′ ) is the joint entropy of the sub-feature data, i.e., information entropy, x 1 ,…,x n are the values ​​of different parameters in the sub-feature data, and P is the probability of these parameters appearing at the same time.

[0085] Then determine the forgetting probability corresponding to each sub-sample data, and arrange each sub-sample data from low to high according to the forgetting probability to obtain the sample data corresponding to the object. The sample data is used to train the basic model to achieve the update and iteration of the basic model, thereby obtaining the iterative model, which is used to provide personalized services for the object.

[0086] For example, the more relevant the sub-sample data is to the object, the greater the probability of forgetting the sub-sample data. For example, the browsing data category and the click data category, the sub-sample data corresponding to the click data category is more relevant to the object behavior, so the probability of forgetting the sub-sample data corresponding to the click data category is higher. Compared with the click data category, the browsing data category has a lower correlation with the object behavior, so the probability of forgetting the sub-sample data corresponding to the browsing data category is lower. Multiple sub-sample data can be arranged from low to high according to the probability of forgetting, and then sub-sample data can be selected from low to high starting from the first word sample data to iteratively train the previous model, thereby achieving continuous iteration of the model.

[0087] See also Figure 3 , Figure 3 is a flow chart of training a basic model provided in an embodiment of the present application. In some embodiments, iterative training of the basic model may include the following steps:

[0088] Step 201: Obtain a basic model, and select the first sub-sample data in the sample data as the current sub-sample data;

[0089] Step 202: iteratively train the basic model according to the current sub-sample data to obtain an iterative model, and save the model parameters of the iterative model;

[0090] Step 203: Select sub-sample data arranged after the current sub-sample data from the sample data, and determine the sub-sample data arranged after the current sub-sample data as the current sub-sample data;

[0091] Step 204, iteratively train the iterative model according to the current sub-sample data to obtain an updated iterative model, and save the model parameters of the updated iterative model, and return to execute the sub-sample data arranged after the current sub-sample data in the sample data until the iterative model and model parameters of the iterative model corresponding to the last sub-sample data arranged in the sample data are obtained.

[0092] Steps 201 to 204 are described in detail below. These steps occur at least before step 140.

[0093] In step 201, a basic model is obtained, and the first sub-sample data in the sample data is selected as the current sub-sample data.

[0094] Among them, the basic model can be understood as a basic model used to provide personalized services for objects. It has a certain model structure and model functions. However, for different objects, the basic model needs to be trained to obtain a model suitable for personalized services for the object.

[0095] In the first training, the first sub-sample data in the sample data can be selected as the current sub-sample data, and the basic model is trained using the sub-sample data. Among them, the first sub-sample data has the lowest probability of being forgotten, and it is considered that it has the lowest degree of relevance to the object.

[0096] In step 202, the basic model is iteratively trained according to the current sub-sample data to obtain an iterative model, and the model parameters of the iterative model are saved.

[0097] Then, the basic model is iteratively trained according to the current sub-sample data to obtain an iterative model. This training can be understood as the first iteration of the basic model, and the model parameters of the iterative model of the first iteration are saved.

[0098] In step 203, sub-sample data arranged after the current sub-sample data is selected from the sample data, and the sub-sample data arranged after the current sub-sample data is determined as the current sub-sample data.

[0099] For example, if the current sub-sample data is the first sub-sample data, then the sub-sample data after the current sub-sample data is the second sub-sample data. Similarly, if the current sub-sample data is the Mth sub-sample data, then the sub-sample data after the current sub-sample data is the M+1th sub-sample data.

[0100] In step 204, the iterative model is iteratively trained according to the current sub-sample data to obtain an updated iterative model, and the model parameters of the updated iterative model are saved. The execution returns to select the sub-sample data arranged after the current sub-sample data in the sample data until the iterative model and model parameters of the iterative model corresponding to the last sub-sample data arranged in the sample data are obtained.

[0101] After updating the current sub-sample data, the iterative training of the latest iterative model is continued according to the updated current sub-sample data, that is, the iterative training of the iterative model corresponding to the previous sub-sample data of the current sub-sample data is continued, so as to realize the iterative training of the iterative model of the previous iteration through the current sub-sample data, thereby updating the iterative model and saving the model parameters of the updated iterative model.

[0102] Then, the execution returns to select the sub-sample data arranged after the current sub-sample data in the sample data, that is, returns to step 203, and then continues to train the iteration model of the previous iteration through the current sub-sample data, so as to obtain an updated iteration model. And so on, until the last sub-sample data arranged continues to iteratively train the iteration model of the previous iteration, so as to obtain the final iteration model and the model parameters of the iteration model, at this time, it is considered that the basic model training corresponding to the object is completed, and the iteration model obtained by training is suitable for personalized services for the object.

[0103] It should be noted that after each iterative model is obtained, the model parameters of the iterative model will be saved, so that among the multiple sub-sample data arranged from low to high according to the forgetting probability, each sub-sample data corresponds to an iterative model, and the model parameters of each iterative model are saved.

[0104] It can be seen from step 201 to step 204 that the iteration of the model in the present application is performed by sequentially training and iterating multiple sub-sample data with forgetting probabilities from high to low, so that each sub-sample data corresponds to an iterative model, and when the model is subsequently trained for forgetting learning, the corresponding target model can be found, and the forgetting learning training can be started from the target model. The knowledge learned before the target model can be preserved, and the knowledge learned after the target model can be forgotten. This ensures the generalization ability of the target model and the efficiency of the forgetting learning training of the target model.

[0105] The above content is the process of training the basic model in this application to iterate the basic model, obtain the iterative model and provide personalized services for the corresponding objects.

[0106] The following will describe step 110 to step 150 in detail. The process of forgetting learning training of the iterative model will be described in detail later.

[0107] In step 110, target data to be deleted and a target data category corresponding to the target data are determined.

[0108] When the object needs to stop using certain personalized services during the process of using personalized services, the corresponding target data will be deleted. The target data that the object needs to delete and the target data category corresponding to the target data can be determined according to the deletion instruction input by the object.

[0109] For example, if the target data is browsing history data, then the target data category of the target data may be determined to be a browsing data category.

[0110] In some implementations, the target data may also be input into a pre-trained classification model to output a target category vector corresponding to the target data, where the target category vector may represent the target data category of the target data.

[0111] In step 120, sample data corresponding to the object is obtained, and the sample data includes a plurality of sub-sample data arranged from low to high probability of forgetting.

[0112] It should be noted that step 120 may occur before iterating the basic model, that is, before steps 201 to 204. That is to say, the iterative training process of the basic model requires the use of multiple sub-sample data arranged from low to high forgetting probabilities, and in the subsequent forgetting learning training process of a target model in the iterative model, multiple sub-sample data arranged from low to high forgetting probabilities are also used.

[0113] In some implementations, obtaining sample data corresponding to an object includes:

[0114] (1.1) Determine the forgetting probability corresponding to each sub-sample data;

[0115] (1.2) According to each sub-sample data arranged from low to high according to the forgetting probability, the sample data corresponding to the object is obtained.

[0116] The forgetting probability of each sub-sample data can be determined by determining the correlation between each sub-sample data and the object. The correlation can be preset based on experience, or by obtaining the historical behavior data of the object, and then determining which types of historical behavior data have changed less in a period of time based on the historical behavior data, and such historical behavior data are considered to have a lower probability of forgetting, and determining which types of historical behavior data have changed more in a period of time based on the historical behavior data, and such historical behavior data are considered to have a higher probability of forgetting.

[0117] For example, the more relevant the sub-sample data is to the object, the greater the probability of forgetting the sub-sample data. For example, the browsing data category and the click data category, the sub-sample data corresponding to the click data category is more relevant to the object behavior, so the probability of forgetting the sub-sample data corresponding to the click data category is higher. Compared with the click data category, the browsing data category has a lower correlation with the object behavior, so the probability of forgetting the sub-sample data corresponding to the browsing data category is lower. Then, in this way, each sub-sample data can be arranged from low to high according to the forgetting probability to obtain the sample data corresponding to the object.

[0118] In step 130, the data category corresponding to each sub-sample data is determined, and the similarity between the target data category and the data category corresponding to each sub-sample data is determined.

[0119] Each sub-sample data has a corresponding data category, such as a browse data category, a click data category, and a search data category, etc. The data category of each sub-sample data can be directly determined, and then the similarity between the target data category and the data category corresponding to each sub-sample data can be determined.

[0120] In some embodiments, each sub-sample data may be input into a pre-trained classification model, and then a data category vector corresponding to each sub-sample data is output, and then the similarity between the data category vector corresponding to each sub-sample data and the target category vector corresponding to the target data is determined. For example, the cosine distance between the data category vector corresponding to each sub-sample data and the target category vector corresponding to the target data may be determined, and the cosine distance is determined as the similarity between the target data category and the data category, thereby obtaining the similarity between the target data category and the data category corresponding to each sub-sample data.

[0121] In some implementations, determining the similarity between the target data category and the data category corresponding to each sub-sample data includes:

[0122] (1.1) Determine the pre-trained single classification model corresponding to the target data category;

[0123] (1.2) Input each sub-sample data into the pre-trained single classification model and output the similarity between the target data category and the data category corresponding to each sub-sample data.

[0124] Among them, at least one pre-trained single classification model can be pre-trained for the target data, and each pre-trained single classification model can be used to distinguish whether other data belongs to the target data category, thereby outputting a probability value. When the probability value is greater than a preset probability value, it is considered that the other data belongs to the target data category.

[0125] Each sub-sample data is input into the pre-trained single classification model, and the probability value of whether each sub-sample data belongs to the target data category is output. The probability value can be considered as the similarity between the target data category and the data category corresponding to each sub-sample data.

[0126] The advantage of this is that, compared with a multi-classification model, a single-classification model can more accurately determine the similarity between the data category of the sub-sample data and the target data category of the target data, which can improve the accuracy of determining the similarity.

[0127] In some implementations, each sub-sample data is input into a pre-trained single classification model, and the similarity between the target data category and the data category corresponding to each sub-sample data is output, including:

[0128] (1.2.1) When there are multiple pre-trained single-classification models, each sub-sample data is input into each pre-trained single-classification model to obtain the similarity between the target data category output by each pre-trained single-classification model and the data category corresponding to each sub-sample data.

[0129] Among them, if there are multiple pre-trained single-classification models, each sub-sample data is input into each pre-trained single-classification model, then each sub-sample data will correspond to multiple similarities, that is, the similarity between the target data category and the data category corresponding to each sub-sample data, and this similarity value has multiple. Since each pre-trained single-classification model is different, the advantage of doing so is that the similarity between the target data category and the data category corresponding to each sub-sample data can be evaluated from different dimensions, thereby improving the accuracy of determining the similarity.

[0130] In step 140, target sub-sample data with the highest similarity is determined from the plurality of sub-sample data, and target model parameters of the target model corresponding to the target sub-sample data are determined.

[0131] Wherein, no matter the number of pre-trained single-classification models is one or more, the similarity between the target data category output by each pre-trained single-classification model and the data category corresponding to each sub-sample data is compared to determine the target sub-sample data with the highest similarity.

[0132] The sub-sample data with the highest similarity is directly determined as the target sub-sample data, and it is considered that the target sub-sample data has a high degree of similarity with the target data. The advantage of this is that the target sub-sample data that is most similar to the target data can be determined to the greatest extent.

[0133] In some implementations, determining a target model parameter of a target model corresponding to the target subsample data includes:

[0134] (1.1) Determine the iterative model corresponding to the target sub-sample data from all the saved iterative models;

[0135] (1.2) The iterative model corresponding to the target sub-sample data is determined as the target model, and the model parameters of the iterative model corresponding to the target sub-sample data are determined as the target model parameters of the target model.

[0136] It can be understood that after determining the target sub-sample data, the iterative model corresponding to the target sub-sample data is determined among all the saved iterative models. The iterative model corresponding to the target sub-sample data has a high degree of match with the target data and can provide personalized services related to the target data. The iterative model corresponding to the target sub-sample data is determined as the target model. At this time, it is necessary to continue to perform forgetting learning training on the target model.

[0137] In this way, the sub-sample data arranged before the target sub-sample data do not need to be trained again. The knowledge learned by the iterative model of the sub-sample data before the target sub-sample data is preserved, and the knowledge learned by the iterative model corresponding to the sub-sample data arranged after the target sub-sample data needs to be forgotten. Because the subsequent iterative models are based on the target model iteration, the forgetting learning training is started based on the target model. This avoids retraining the model from scratch and improves the efficiency of the model's forgetting learning training.

[0138] The model parameters of the iterative model corresponding to the target sub-sample data may be determined as the target model parameters of the target model, and the target model parameters may be subsequently adjusted to achieve forgetting learning training of the target model.

[0139] In step 150, new sample data is obtained, and forgetting learning training is performed on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training.

[0140] After determining the target model, new sample data can be obtained. The new sample data can be the data regenerated when the object uses the corresponding service, such as regenerated browsing data or click data, etc. Then, forgetting learning training is performed on the target model based on the new sample data to adjust the target model parameters of the target model, thereby obtaining the target model after forgetting learning training.

[0141] In some implementations, performing forgetting learning training on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training includes:

[0142] (1.1) Obtain the label value corresponding to the new sample data;

[0143] (1.2) Input the new sample data into the target model and output the predicted value of the new sample data;

[0144] (1.3) Determine the loss value between the label value and the predicted value, and adjust the target model parameters based on the loss value to obtain the target model after forgetting learning training.

[0145] Among them, for the new sample data, the label value corresponding to the new sample data can be determined, and the label value can be the recommendation value of the recommended content related to the new sample data. Then the new sample data is input into the target model, and the predicted value of the new sample data is output. The predicted value can be a recommendation value generated by the target model based on the new sample data, which can be used to recommend the corresponding recommended content.

[0146] Finally, the loss value between the label value and the predicted value is determined, and the target model parameters are adjusted based on the loss value to obtain the target model after forgetting learning training. For example, when the loss value is greater than the preset loss value, it means that the prediction of the target model is not accurate enough. At this time, the target model parameters need to be adjusted based on the loss value to obtain the adjusted target model.

[0147] From the above content, it can be seen that in the present application, when the model is trained for forgetting learning, it is not retrained from the basic model, nor is it trained from scratch. Instead, the target sub-sample data that is most similar to the target data and its corresponding target model are determined, and forgetting learning training is started from the target model. This saves the retraining of sub-sample data with a lower probability of forgetting than the target sub-sample data, thereby saving the training process.

[0148] As can be seen from the above, in the present application, by determining the target data that needs to be deleted by the object and the target data category corresponding to the target data; obtaining the sample data corresponding to the object, the sample data contains multiple sub-sample data arranged from low to high probability of forgetting; determining the data category corresponding to each sub-sample data, and determining the similarity between the target data category and the data category corresponding to each sub-sample data; determining the target sub-sample data with the highest similarity among multiple sub-sample data, and determining the target model parameters of the target model corresponding to the target sub-sample data; obtaining new sample data, and performing forgetting learning training on the target model based on the target model parameters and the new sample data, to obtain the target model after forgetting learning training. In this way, by determining the target data that needs to be deleted by the object, and then determining the target sub-sample data with the highest similarity to the target data among multiple sub-sample data with a low to high probability of forgetting, and then starting the forgetting learning training from the target model corresponding to the target sub-sample data, compared with the related art of re-training the model from the beginning for forgetting learning, the present application directly starts the forgetting learning training from the target model, which greatly improves the efficiency of the forgetting learning training of the model.

[0149] See also Figure 4 , Figure 4 is another flow chart of the model forgetting learning method provided in the embodiment of the present application. The model forgetting learning method may include the following steps:

[0150] Step 301: Obtain sample data corresponding to the object, where the sample data includes a plurality of sub-sample data arranged from low to high probability of forgetting;

[0151] Step 302: Obtain a basic model, and select the first sub-sample data in the sample data as the current sub-sample data;

[0152] Step 303: iteratively train the basic model according to the current sub-sample data to obtain an iterative model, and save the model parameters of the iterative model;

[0153] Step 304: Select sub-sample data arranged after the current sub-sample data from the sample data, and determine the sub-sample data arranged after the current sub-sample data as the current sub-sample data;

[0154] Step 305: iteratively train the iterative model according to the current sub-sample data to obtain an updated iterative model, save the model parameters of the updated iterative model, and return to execute the sub-sample data arranged after the current sub-sample data in the sample data until the iterative model and model parameters of the iterative model corresponding to the last sub-sample data arranged in the sample data are obtained;

[0155] Step 306: Determine the target data to be deleted and the target data category corresponding to the target data;

[0156] Step 307: determine the pre-trained single classification model corresponding to the target data category;

[0157] Step 308: input each sub-sample data into the pre-trained single classification model, output the similarity between the target data category and the data category corresponding to each sub-sample data, and determine the target sub-sample data with the highest similarity among the multiple sub-sample data;

[0158] Step 309: Determine the iterative model corresponding to the target sub-sample data from all the saved iterative models;

[0159] Step 310: determine the iterative model corresponding to the target sub-sample data as the target model, and determine the model parameters of the iterative model corresponding to the target sub-sample data as the target model parameters of the target model;

[0160] Step 311: obtain new sample data, and perform forgetting learning training on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training.

[0161] In the above embodiments, the description of each embodiment has its own focus. For the part that is not described in detail in a certain embodiment, please refer to the above-mentioned model forgetting learning method, which will not be repeated here.

[0162] See also Figure 5 , the embodiment of the present application also provides a model forgetting learning device, which can implement the above-mentioned model forgetting learning method. In the embodiment of the present application, the term "module" or "unit" refers to a computer program or a part of a computer program with a predetermined function, and works with other related parts to achieve a predetermined goal, and can be implemented in whole or in part by using software, hardware (such as processing circuits or memories) or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be part of an overall module or unit that includes the function of the module or unit.

[0163] The model forgetting learning device includes:

[0164] A first determination module 410 is used to determine target data to be deleted and a target data category corresponding to the target data;

[0165] An acquisition module 420 is used to acquire sample data corresponding to the object, where the sample data includes a plurality of sub-sample data arranged in ascending order of forgetting probability;

[0166] A second determination module 430 is used to determine the data category corresponding to each sub-sample data, and determine the similarity between the target data category and the data category corresponding to each sub-sample data;

[0167] A comparison module 440 is used to determine the target sub-sample data with the highest similarity among the multiple sub-sample data, and determine the target model parameters of the target model corresponding to the target sub-sample data;

[0168] The training module 450 is used to obtain new sample data, and perform forgetting learning training on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training.

[0169] In some embodiments, the model forgetting learning device further includes a filtering module 460, which is used to obtain feature data corresponding to the object before obtaining sample data corresponding to the object, and to segment the feature data to obtain a plurality of sub-feature data;

[0170] The information entropy corresponding to each sub-feature data is determined, and sub-sample data having an information entropy higher than a preset information entropy is screened out from the multiple sub-feature data.

[0171] The acquisition module 420 is further used to determine the forgetting probability corresponding to each sub-sample data;

[0172] According to each sub-sample data arranged from low to high according to the forgetting probability, the sample data corresponding to the object is obtained.

[0173] In some embodiments, the model forgetting learning device further includes a pre-training module 470, which is used to obtain a basic model before determining the target sub-sample data with the highest similarity among the multiple sub-sample data, and select the first sub-sample data in the sample data as the current sub-sample data;

[0174] Iteratively train the basic model according to the current sub-sample data to obtain an iterative model, and save the model parameters of the iterative model;

[0175] Selecting sub-sample data arranged after the current sub-sample data from the sample data, and determining the sub-sample data arranged after the current sub-sample data as the current sub-sample data;

[0176] Iteratively train the iterative model according to the current sub-sample data to obtain an updated iterative model, save the model parameters of the updated iterative model, and return to execute the sub-sample data arranged after the current sub-sample data in the sample data until the iterative model and model parameters of the iterative model corresponding to the last sub-sample data arranged in the sample data are obtained.

[0177] In some implementations, the comparison module 440 is used to determine the iterative model corresponding to the target sub-sample data from among all the saved iterative models;

[0178] The iterative model corresponding to the target sub-sample data is determined as the target model, and the model parameters of the iterative model corresponding to the target sub-sample data are determined as the target model parameters of the target model.

[0179] In some embodiments, the second determination module 430 further includes a determination submodule 431 and an input submodule 431, wherein the determination submodule 432 is configured to determine a pre-trained single classification model corresponding to the target data category;

[0180] The input submodule 432 is used to input each sub-sample data into the pre-trained single classification model, and output the similarity between the target data category and the data category corresponding to each sub-sample data.

[0181] In some embodiments, the input submodule 432 is used to input each sub-sample data into each pre-trained single-classification model when there are multiple pre-trained single-classification models, and obtain the similarity between the target data category output by each pre-trained single-classification model and the data category corresponding to each sub-sample data.

[0182] The comparison module 440 is used to compare the similarity between the target data category output by each pre-trained single classification model and the data category corresponding to each sub-sample data, so as to determine the target sub-sample data with the highest similarity.

[0183] In some embodiments, the training module 450 is used to obtain label values ​​corresponding to new sample data;

[0184] Input the new sample data into the target model and output the predicted value of the new sample data;

[0185] The loss value between the label value and the predicted value is determined, and the target model parameters are adjusted based on the loss value to obtain the target model after forgetting learning training.

[0186] In the above embodiments, the description of each embodiment has its own focus. For the part that is not described in detail in a certain embodiment, please refer to the above-mentioned model forgetting learning method, which will not be repeated here.

[0187] As can be seen from the above, in an embodiment of the present application, the first determination module 410 determines the target data that needs to be deleted from the object and the target data category corresponding to the target data; the acquisition module 420 acquires the sample data corresponding to the object, and the sample data includes multiple sub-sample data arranged in ascending order of forgetting probability; the second determination module 430 determines the data category corresponding to each sub-sample data, and determines the similarity between the target data category and the data category corresponding to each sub-sample data; the comparison module 440 determines the target sub-sample data with the highest similarity among multiple sub-sample data, and determines the target model parameters of the target model corresponding to the target sub-sample data; the training module 450 acquires new sample data, and performs forgetting learning training on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training. In this way, by determining the target data that needs to be deleted from the object, and then determining the target sub-sample data with the highest similarity to the target data from multiple sub-sample data with forgetting probabilities from low to high, and then starting the forgetting learning training from the target model corresponding to the target sub-sample data, compared with the related art of re-training the model from scratch, the present application directly starts the forgetting learning training from the target model, which greatly improves the efficiency of the forgetting learning training of the model.

[0188] The embodiment of the present application also provides a computer device, the computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the above-mentioned model forgetting learning method when executing the computer program. The computer device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0189] See also Figure 6 , Figure 6 The hardware structure of a computer device according to another embodiment is shown, and the computer device includes:

[0190] The processor 501 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application;

[0191] The memory 502 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 502 can store an operating system and other applications. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 502, and the processor 501 calls and executes the model forgetting learning method of the embodiment of this application;

[0192] Input / output interface 503, used to implement information input and output;

[0193] Communication interface 504, used to realize communication interaction between the device and other devices, which can be realized by wired mode (such as USB, network cable, etc.) or wireless mode (such as mobile network, WIFI, Bluetooth, etc.);

[0194] A bus 505 that transmits information between the various components of the device (e.g., the processor 501, the memory 502, the input / output interface 503, and the communication interface 504);

[0195] The processor 501 , the memory 502 , the input / output interface 503 and the communication interface 504 are connected to each other in communication within the device via the bus 505 .

[0196] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program, and when the computer program is executed by a processor, the above-mentioned model forgetting learning method is implemented.

[0197] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0198] The model forgetting learning method, device, computer equipment and storage medium proposed in the present application, by determining the target data to be deleted by the object and the target data category corresponding to the target data; obtaining the sample data corresponding to the object, the sample data contains multiple sub-sample data arranged from low to high probability of forgetting; determining the data category corresponding to each sub-sample data, and determining the similarity between the target data category and the data category corresponding to each sub-sample data; determining the target sub-sample data with the highest similarity among the multiple sub-sample data, and determining the target model parameters of the target model corresponding to the target sub-sample data; obtaining new sample data, and performing forgetting learning training on the target model based on the target model parameters and the new sample data, to obtain the target model after forgetting learning training. In this way, by determining the target data to be deleted by the object, and then determining the target sub-sample data with the highest similarity to the target data among the multiple sub-sample data with the forgetting probability from low to high, and then starting the forgetting learning training from the target model corresponding to the target sub-sample data, compared with the related art of re-training the model from the beginning for forgetting learning, the present application directly starts the forgetting learning training from the target model, which greatly improves the efficiency of the forgetting learning training of the model.

[0199] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0200] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0201] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0202] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0203] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0204] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0205] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0206] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0207] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0208] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a storage medium and includes multiple instructions for 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 of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM), random access memory (RAM), disk or optical disk, and other media that can store programs.

[0209] The preferred embodiments of the present application are described above with reference to the accompanying drawings, but the scope of the rights of the present application is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present application should be within the scope of the rights of the present application.

Claims

1. A model forgetting learning method, characterized in that: include: Determine the target data to be deleted and the target data category corresponding to the target data; Acquire sample data corresponding to the object, wherein the sample data includes a plurality of sub-sample data arranged in ascending order of forgetting probability; Determine the data category corresponding to each sub-sample data, and determine the similarity between the target data category and the data category corresponding to each sub-sample data; Determine the target sub-sample data with the highest similarity among the multiple sub-sample data, and determine the target model parameters of the target model corresponding to the target sub-sample data; New sample data is acquired, and forgetting learning training is performed on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training.

2. The model forgetting learning method according to claim 1, characterized in that: Before obtaining the sample data corresponding to the object, the method further includes: Acquire feature data corresponding to the object, and segment the feature data to obtain a plurality of sub-feature data; Determine the information entropy corresponding to each sub-feature data, and screen out sub-sample data whose information entropy is higher than a preset information entropy from the multiple sub-feature data; The obtaining of sample data corresponding to the object includes: Determine the forgetting probability corresponding to each sub-sample data; The sample data corresponding to the object is obtained according to each sub-sample data arranged from low to high in the forgetting probability.

3. The model forgetting learning method according to claim 1, characterized in that: Before determining the target sub-sample data with the highest similarity among the plurality of sub-sample data, the method further includes: Obtain a basic model, and select the first sub-sample data in the sample data as the current sub-sample data; Iteratively training the basic model according to the current sub-sample data to obtain an iterative model, and saving model parameters of the iterative model; Selecting sub-sample data arranged after the current sub-sample data from the sample data, and determining the sub-sample data arranged after the current sub-sample data as the current sub-sample data; Iteratively train the iterative model according to the current sub-sample data to obtain an updated iterative model, and save the model parameters of the updated iterative model, and return to execute the selection of sub-sample data arranged after the current sub-sample data in the sample data until the iterative model corresponding to the last sub-sample data arranged in the sample data and the model parameters of the iterative model are obtained.

4. The model forgetting learning method according to claim 3, characterized in that: The determining the target model parameters of the target model corresponding to the target sub-sample data includes: Determining the iterative model corresponding to the target sub-sample data from among all the saved iterative models; The iterative model corresponding to the target sub-sample data is determined as the target model, and the model parameters of the iterative model corresponding to the target sub-sample data are determined as the target model parameters of the target model.

5. The model forgetting learning method according to claim 1, characterized in that: The determining the similarity between the target data category and the data category corresponding to each sub-sample data includes: Determine a pre-trained single classification model corresponding to the target data category; Each sub-sample data is input into the pre-trained single classification model, and the similarity between the target data category and the data category corresponding to each sub-sample data is output.

6. The model forgetting learning method according to claim 5, characterized in that: The step of inputting each sub-sample data into the pre-trained single classification model and outputting the similarity between the target data category and the data category corresponding to each sub-sample data includes: When there are multiple pre-trained single-classification models, each sub-sample data is input into each pre-trained single-classification model, and the similarity between the target data category output by each pre-trained single-classification model and the data category corresponding to each sub-sample data is obtained; The step of determining the target sub-sample data with the highest similarity among the plurality of sub-sample data includes: The similarity between the target data category output by each pre-trained single classification model and the data category corresponding to each sub-sample data is compared to determine the target sub-sample data with the highest similarity.

7. The model forgetting learning method according to claim 1, characterized in that: The performing forgetting learning training on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training includes: Obtaining a label value corresponding to the new sample data; Inputting the new sample data into the target model, and outputting the predicted value of the new sample data; A loss value between the label value and the predicted value is determined, and the target model parameters are adjusted based on the loss value to obtain a target model after forgetting learning training.

8. A model forgetting learning device, characterized in that: include: A first determination module, used to determine target data to be deleted from an object and a target data category corresponding to the target data; An acquisition module, used for acquiring sample data corresponding to the object, wherein the sample data includes a plurality of sub-sample data arranged in ascending order of forgetting probability; A second determination module, used to determine the data category corresponding to each sub-sample data, and determine the similarity between the target data category and the data category corresponding to each sub-sample data; A comparison module, used to determine the target sub-sample data with the highest similarity among the multiple sub-sample data, and determine the target model parameters of the target model corresponding to the target sub-sample data; The training module is used to obtain new sample data and perform forgetting learning training on the target model based on the target model parameters and the new sample data to obtain the target model after forgetting learning training.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, which are suitable for loading by a processor to execute the model forgetting learning method described in any one of claims 1 to 7.

10. A computer 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, the model forgetting learning method described in any one of claims 1 to 7 is implemented.

Images