Privacy protection network generation method, system and device based on Bayesian optimization

Abstract

The invention discloses a privacy protection network generation method, system and device based on Bayesian optimization, and the method comprises the steps: constructing an HEMET model, embedding a parameter quantity constraint into a Bayesian optimization collection function through index penalty to obtain an improved collection function, and constructing a proxy model through a sample set; calculating an acquisition function by using the proxy model to obtain an optimal hyper-parameter combination; constructing and verifying a new HEMET model by using the optimal hyper-parameter combination, and calculating a parameter quantity; if the parameter quantity is greater than a parameter quantity threshold, iterating a new HEMET model, and if the parameter quantity is less than or equal to the parameter quantity threshold, training and evaluating the new HEMET model to obtain the accuracy of the test set; adding the new hyper-parameter combination and accuracy to a sample set; and if the accuracy rate of the test set is greater than the accuracy rate, updating the new hyper-parameter combination by using the optimal hyper-parameter combination, and updating the HEMET model by using the new HEMET model. According to the technical scheme provided by the invention, the technical problem of single optimization target of the Bayesian optimization algorithm can be solved.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a method, system and device for generating a privacy-preserving network based on Bayesian optimization. Background Art

[0002] In terms of privacy protection on the Internet, the Bayesian optimization algorithm based on the Tree-structured ParzenEstimator (TPE) is a method that efficiently searches the hyperparameter space by modeling the distribution of high-quality and low-quality samples separately. Its function is to automatically find the best solution in complex hyperparameter combinations. Its advantages are high computational efficiency and automatic adjustment of the search range, but the optimization goal is single. Summary of the invention

[0003] The present invention provides a privacy protection network generation method, system and device based on Bayesian optimization, aiming to effectively solve the technical problem of single optimization target of Bayesian optimization algorithm in the privacy protection of Internet in the prior art.

[0004] According to a first aspect of the present invention, the present invention provides a method for generating a privacy-preserving network based on Bayesian optimization, comprising: constructing a HEMET model, and embedding a parameter quantity constraint into a Bayesian optimized acquisition function through an exponential penalty to obtain an improved acquisition function, and constructing a proxy model using a sample set; calculating the improved acquisition function using the proxy model to obtain the next possible optimal hyperparameter combination; constructing and verifying a new HEMET model using the optimal hyperparameter combination, and calculating its parameter quantity; if the parameter quantity is greater than a preset parameter quantity threshold of the HEMET model, continuing to iterate the new HEMET model, and if the parameter quantity is less than or equal to the parameter quantity threshold, training and evaluating the new HEMET model on a test set to obtain the accuracy of the test set; adding a new hyperparameter combination from the HEMET and its corresponding accuracy to the sample set; if the accuracy of the test set is greater than the accuracy of the new hyperparameter combination, updating the new hyperparameter combination using the optimal hyperparameter combination, and updating the HEMET model using the new HEMET model.

[0005] Furthermore, the improved acquisition function is: ; Among them, EI is the expected improvement function, is the weight coefficient, is the currently known optimal objective function value, is the objective function value obtained in the previous round of calculation, is the hyperparameter combination of the HEMET model, is the total number of parameters of the HEMET model, The parameter threshold of the HEMET model is set.

[0006] Furthermore, in the improved acquisition function, if ,but , the candidate solutions are filtered; if , the penalty term degenerates into a constant, maintaining the optimization logic of traditional EI, where the adjustment To achieve Pareto optimization of accuracy-parameter quantity.

[0007] Furthermore, the improved acquisition function is calculated using the proxy model to obtain a calculation formula for the next possible optimal hyperparameter combination, including: ; in, is the next possible optimal hyperparameter combination.

[0008] Furthermore, the new HEMET model is trained and evaluated on a test set to obtain an accuracy rate of the test set including: Compile a new HEMET model using the Adam optimizer , using the validation set To verify, in the test set Upper Assessment , get the accuracy of the test set .

[0009] Furthermore, it also includes: outputting the optimal hyperparameter combination and the updated new HEMET model.

[0010] According to the second aspect of the present invention, the present invention also provides a privacy-preserving network generation system based on Bayesian optimization, comprising: a model preparation module, used to construct a HEMET model, and embed parameter quantity constraints into a Bayesian optimized acquisition function through exponential penalty to obtain an improved acquisition function, and use a sample set to construct a proxy model; an acquisition function calculation module, used to use the proxy model to calculate the improved acquisition function to obtain the next possible optimal hyperparameter combination; a model construction and verification module, used to use the optimal hyperparameter combination to construct and verify a new HEMET model, and calculate its parameter quantity; a model training evaluation module, used to evaluate the model training results if the parameter quantity is large. If the parameter amount is less than or equal to the preset parameter amount threshold of the HEMET model, the new HEMET model continues to be iterated; if the parameter amount is less than or equal to the parameter amount threshold, the new HEMET model is trained and evaluated on the test set to obtain the accuracy of the test set; a sample update module is used to add the new hyperparameter combination from the HEMET and its corresponding accuracy to the sample set; an optimal solution iteration module is used to update the new hyperparameter combination with the optimal hyperparameter combination if the accuracy of the test set is greater than the accuracy of the new hyperparameter combination, and to update the HEMET model with the new HEMET model.

[0011] According to the third aspect of the present invention, the present invention also provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, it implements any one of the above-mentioned methods for generating a privacy-preserving network based on Bayesian optimization.

[0012] According to a fourth aspect of the present invention, the present invention further provides a computer-readable storage medium having a computer program stored thereon, characterized in that when the computer program is executed by a processor, it implements any one of the above-mentioned methods for generating a privacy-preserving network based on Bayesian optimization.

[0013] According to another aspect of the present invention, the present invention also provides a computer program for executing any one of the above methods for generating a privacy-preserving network based on Bayesian optimization.

[0014] Through one or more of the above embodiments of the present invention, at least the following technical effects can be achieved: In the technical solution disclosed in the present invention, the present invention performs hyperparameter tuning on the HEMET network based on the Bayesian optimization algorithm. First, an acquisition function is constructed by introducing a parameter quantity penalty term, a parameter quantity threshold is set, and then the HEMET network model is iteratively trained within the parameter quantity threshold range. A Bayesian optimization HEMET solution is proposed, thereby achieving Pareto optimization between the parameter quantity and accuracy of the model. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] The technical solutions and other beneficial effects of the present invention will be made apparent by describing in detail the specific embodiments of the present invention in conjunction with the accompanying drawings.

[0016] Figure 1 A flowchart of a method for generating a privacy-preserving network based on Bayesian optimization provided in an embodiment of the present invention; Figure 2 A network model structure diagram of a privacy-preserving network generation method based on Bayesian optimization provided in an embodiment of the present invention; Figure 3 A diagram of the internal structure of the Fire module in the network model of the privacy-preserving network generation method based on Bayesian optimization provided in an embodiment of the present invention; Figure 4 A framework diagram of a privacy protection network generation system based on Bayesian optimization provided in an embodiment of the present invention; Figure 5 A schematic block diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] The technical scheme in the embodiment of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiment of the present invention. Obviously, the described embodiment is only a part of the embodiment of the present invention, not all of the embodiments. Based on the embodiment of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0018] In the description of the present invention, it should be noted that, unless otherwise clearly specified and limited, the term "and / or" herein is merely a description of the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B may represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " herein, unless otherwise specified, generally indicates that the associated objects before and after are in an "or" relationship.

[0019] The HEMET network is a privacy-preserving mobile neural network architecture that is friendly to homomorphic encryption. It optimizes the network structure through a grid search algorithm and coefficient merging technology, which reduces inference latency while improving accuracy. However, its structural optimization may sacrifice some of the model compression advantages and significantly increase the number of model parameters. The Bayesian optimization algorithm based on the Tree-structured Parzen Estimator (TPE) (see reference [1] Bergstra J, Bardenet R, Bengio Y, et al. Algorithms for hyper-parameter optimization [J]. Advances in neural information processing systems, 2011, 24.) is a method that efficiently searches the hyperparameter space by modeling the distribution of high-quality and low-quality samples respectively. Its function is to automatically find the best solution in a complex hyperparameter combination. Its advantages are high computational efficiency and automatic adjustment of the search range, but the optimization goal is single.

[0020] Therefore, in order to solve the above problems, the embodiments of the present application provide a method, system and device for generating a privacy-preserving network based on Bayesian optimization.

[0021] Figure 1 The method for generating a privacy-preserving network based on Bayesian optimization provided by an embodiment of the present invention includes: S101, constructing a HEMET model, and embedding the parameter quantity constraint into the Bayesian optimization acquisition function through exponential penalty to obtain an improved acquisition function, and constructing a proxy model using the sample set; S102, using the proxy model to calculate the improved acquisition function to obtain the next possible optimal hyperparameter combination; S103, constructing and verifying a new HEMET model using the optimal hyperparameter combination, and calculating its parameter quantity; S104, if the parameter amount is greater than the preset parameter amount threshold of the HEMET model, continue to iterate the new HEMET model; if the parameter amount is less than or equal to the parameter amount threshold, train and evaluate the new HEMET model on the test set to obtain the accuracy of the test set; S105, adding the new hyperparameter combination from HEMET and its corresponding accuracy to the sample set; S106. If the accuracy of the test set is greater than the accuracy of the new hyperparameter combination, the new hyperparameter combination is updated using the optimal hyperparameter combination, and the HEMET model is updated using the new HEMET model.

[0022] The privacy-preserving network generation method based on Bayesian optimization provided in this embodiment performs hyperparameter tuning on the HEMET network based on the Bayesian optimization algorithm. First, an acquisition function is constructed by introducing a parameter quantity penalty term, a parameter quantity threshold is set, and then the HEMET network model is iteratively trained within the parameter quantity threshold range. A Bayesian optimization HEMET scheme is proposed, thereby achieving Pareto optimization between the parameter quantity and accuracy of the model.

[0023] In some embodiments, the sample set used in step S101 Construct the proxy model. The proxy model can be constructed according to formula (2) in Section 4 of reference [1] .

[0024] In some embodiments, the improved acquisition function is: ; Among them, EI is the expected improvement function, is the weight coefficient, is the currently known optimal objective function value, is the objective function value obtained in the previous round of calculation, is the hyperparameter combination of the HEMET model, is the total number of parameters of the HEMET model, The parameter threshold of the HEMET model is set.

[0025] In the improved acquisition function, if ,but , the candidate solutions are filtered; if , the penalty term degenerates into a constant, maintaining the optimization logic of traditional EI, where the adjustment To achieve Pareto optimization of accuracy-parameter quantity.

[0026] In some embodiments, the calculation formula for calculating the improved acquisition function using the proxy model to obtain the next possible optimal hyperparameter combination includes: ; in, is the next possible optimal hyperparameter combination.

[0027] Therefore, in step S102, based on the proxy model , calculate the improved acquisition function , and solve , get the next possible optimal hyperparameter combination .

[0028] In step S103, the optimal hyperparameter combination is used. Build the model , and calculate its parameter In this embodiment, if , indicating that the model parameter exceeds the threshold, then continue to the next round of iteration; if , then continue with the subsequent operations.

[0029] In some embodiments, the training and evaluation of the new HEMET model on a test set to obtain the accuracy of the test set includes: Compile a new HEMET model using the Adam optimizer , using the validation set To verify, in the test set Upper Assessment , get the accuracy of the test set .

[0030] In some embodiments, the privacy-preserving network generation method based on Bayesian optimization further includes: outputting the optimal hyperparameter combination and the updated new HEMET model.

[0031] Therefore, in step S105 and step S106, the new hyperparameter combination And its corresponding accuracy Add to sample collection If , indicating that the current model performance is better, then update the optimal hyperparameter combination , and update the optimal model Finally, the algorithm outputs the optimal hyperparameter combination and the optimal model .

[0032] The above obtained It is obtained after multiple iterations of optimization in the hyperparameter space, which enables the model to achieve a higher accuracy while satisfying the parameter quantity constraints; is based on Build and train the optimal model.

[0033] The embodiment of the present application also conducts experimental tests on the privacy protection network generation method based on Bayesian optimization and the existing method, as follows: A single small server was used for experimental testing. The CPU model of the device is Intel (R) Core(TM) i9-10900X, which has 10 cores, 128GB RAM memory, and is equipped with two NVIDIA GeForce RTX 3090 graphics cards. Based on the Microsoft SEAL library, the data is encrypted by the fully homomorphic encryption algorithm RNS-CKKS. To ensure that all ciphertexts can achieve a 128-bit security level in the privacy-preserving neural network, the relevant configuration of the encryption scheme is specified when the code is executed. The order of the ring polynomial 𝑁 is set to 32768, the precision of the integer part of the encrypted data is 10, and the precision of the decimal part is 51, that is, the scaling factor yes , all of these parameters together affect the security, accuracy, and efficiency of encryption operations. This experiment uses the optuna library to define the hyperparameter search space.

[0034] The network generated by the privacy-preserving network generation method based on Bayesian optimization provided in this application and the existing method is By-HEMET. The network model contains a total of 4 layers of Conv modules (where each Conv module has a structure including a convolution layer, an approximate activation layer and a batch normalization layer), 2 layers of Fire modules and 3 layers of pooling modules. The specific structure diagram of the network model is as follows Figure 2 As shown, the internal structure of the Fire module is as follows Figure 3 As shown in the figure, the Fire module mainly consists of two parts: the Squeeze layer and the Expand layer. The Squeeze layer uses a 1x1 convolution kernel to reduce the number of channels of the input feature map to reduce computational complexity; the Expand layer uses 1x1 and 3x3 convolution kernels to increase the number of channels of the feature map to extract more feature information.

[0035] For the By-HEMET model solution generated by the privacy-preserving network generation method based on Bayesian optimization in the embodiment of the present application, and the traditional HEMET model solution, a ciphertext reasoning experiment was conducted on the CIFAR10 dataset (see reference [2] David Corvoysier. 2017. SqueezeNet for CIFAR-10). The experimental results are shown in Table 1: Table 1 Comparison of model inference performance under different hyperparameter settings

[0036] From the data in Table 1, it can be seen that the optimized By-HEMET scheme provided in this embodiment reduces the total number of parameters by about 41.7% compared with the baseline model HEMET scheme. When reasoning with encrypted text, the By-HEMET scheme needs to process fewer model parameters, thereby speeding up the reasoning speed. In addition, the hyperparameter combination of the By-HEMET scheme helps the model to more accurately capture patterns and features in the data, which can make HEMET better adapt to the characteristics of the data set, thereby improving the reasoning accuracy of the privacy-preserving neural network model. The present invention has expanded the application space for privacy-preserving neural networks in practical scenarios and has important theoretical and practical application value.

[0037] See also Figure 4 The embodiment of the present application also provides a privacy protection network generation system based on Bayesian optimization, including: a model preparation module 1, an acquisition function calculation module 2, a model construction verification module 3, a model training evaluation module 4, a sample update module 5 and an optimal solution iteration module 6.

[0038] Among them, the model preparation module 1 is used to construct the HEMET model, and embed the parameter quantity constraint into the Bayesian optimization acquisition function through exponential penalty to obtain an improved acquisition function, and use the sample set to build the proxy model; the acquisition function calculation module 2 is used to use the proxy model to calculate the improved acquisition function to obtain the next possible optimal hyperparameter combination; the model construction and verification module 3 is used to use the optimal hyperparameter combination to construct and verify the new HEMET model, and calculate its parameter quantity; the model training and evaluation module 4 is used to continue to iterate the new HEMET model if the parameter quantity is greater than the preset parameter quantity threshold of the HEMET model, and if the parameter quantity is less than or equal to the parameter quantity threshold, train and evaluate the new HEMET model on the test set to obtain the accuracy of the test set; the sample update module 5 is used to add the new hyperparameter combination from the HEMET and its corresponding accuracy to the sample set; the optimal solution iteration module 6 is used to use the optimal hyperparameter combination to update the new hyperparameter combination if the accuracy of the test set is greater than the accuracy of the new hyperparameter combination, and update the HEMET model using the new HEMET model.

[0039] The privacy-preserving network generation system based on Bayesian optimization provided in this embodiment performs hyperparameter tuning on the HEMET network based on the Bayesian optimization algorithm. First, an acquisition function is constructed by introducing a parameter quantity penalty term, a parameter quantity threshold is set, and then the HEMET network model is iteratively trained within the parameter quantity threshold range. A Bayesian optimization HEMET scheme is proposed, thereby achieving Pareto optimization between the parameter quantity and accuracy of the model.

[0040] In some embodiments, the improved acquisition function is: ; Among them, EI is the expected improvement function, is the weight coefficient, is the currently known optimal objective function value, is the objective function value obtained in the previous round of calculation, is the hyperparameter combination of the HEMET model, is the total number of parameters of the HEMET model, The HEMET model parameter threshold is set.

[0041] In some embodiments, in the improved acquisition function, if ,but , the candidate solutions are filtered; if , the penalty term degenerates into a constant, maintaining the optimization logic of traditional EI, where the adjustment To achieve Pareto optimization of accuracy-parameter quantity.

[0042] In some embodiments, the calculation formula for calculating the improved acquisition function using the proxy model to obtain the next possible optimal hyperparameter combination includes: ; in, is the next possible optimal hyperparameter combination.

[0043] In some embodiments, the training and evaluation of the new HEMET model on a test set to obtain the accuracy of the test set includes: Compile a new HEMET model using the Adam optimizer , using the validation set To verify, in the test set Upper Assessment , get the accuracy of the test set .

[0044] In some embodiments, the privacy-preserving network generation system based on Bayesian optimization further includes an output module for outputting the optimal hyperparameter combination and the updated new HEMET model.

[0045] The present application embodiment provides an electronic device. Figure 5 The electronic device includes: a memory 601, a processor 602, and a computer program stored in the memory 601 and executable on the processor 602. When the processor 602 executes the computer program, the privacy protection network generation method based on Bayesian optimization described above is implemented.

[0046] Furthermore, the electronic device also includes: at least one input device 603 and at least one output device 604 .

[0047] The memory 601 , processor 602 , input device 603 and output device 604 are connected via a bus 605 .

[0048] The input device 603 may be a camera, a touch panel, a physical button or a mouse, etc. The output device 604 may be a display screen.

[0049] The memory 601 may be a high-speed random access memory (RAM) memory, or a non-volatile memory, such as a disk memory. The memory 601 is used to store a set of executable program codes, and the processor 602 is coupled to the memory 601 .

[0050] Furthermore, the embodiment of the present application also provides a computer-readable storage medium, which may be disposed in the electronic device in each of the above embodiments, and the computer-readable storage medium may be the memory 601 in the above embodiments. The computer-readable storage medium stores a computer program, which, when executed by the processor 602, implements the privacy protection network generation method based on Bayesian optimization described in the above method embodiment.

[0051] Furthermore, the computer storable medium may also be a U disk, a mobile hard disk, a read-only memory (ROM), a RAM, a magnetic disk or an optical disk, or other medium that can store program codes.

[0052] 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 modules is only a logical function division. There may be other division methods in actual implementation, such as multiple modules 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 modules, which can be electrical, mechanical or other forms.

[0053] The modules described as separate components may or may not be physically separated, and the components shown as modules may or may not be physical modules, that is, they may be located in one place or distributed on multiple network modules. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0054] In addition, each functional module in each embodiment of the present invention may be integrated into one processing module, or each module may exist physically separately, or two or more modules may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of software functional modules.

[0055] If the integrated module is implemented in the form of a software function module 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 invention, 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 several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention.

[0056] It should be noted that, for the convenience of description, the aforementioned method embodiments are all described as a series of action combinations, but those skilled in the art should be aware that the present invention is not limited by the described action sequence, because according to the present invention, certain steps can be performed in other sequences or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required by the present invention.

[0057] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0058] In summary, although the present invention has been disclosed as above in terms of preferred embodiments, the above preferred embodiments are not intended to limit the present invention. A person skilled in the art may make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention shall be based on the scope defined in the claims.

Claims

1. A privacy-preserving network generation method based on Bayesian optimization, characterized in that: include: The HEMET model is constructed, and the parameter quantity constraint is embedded into the Bayesian optimization acquisition function through exponential penalty to obtain an improved acquisition function, and the surrogate model is constructed using the sample set; Calculating the improved acquisition function using the proxy model to obtain the next possible optimal hyperparameter combination; Using the optimal hyperparameter combination to construct and verify a new HEMET model, and calculate its parameter quantity; If the parameter amount is greater than a preset parameter amount threshold of the HEMET model, then continue to iterate the new HEMET model; if the parameter amount is less than or equal to the parameter amount threshold, then train and evaluate the new HEMET model on the test set to obtain the accuracy of the test set; Adding new hyperparameter combinations from the HEMET and their corresponding accuracy rates to the sample set; If the accuracy of the test set is greater than the accuracy of the new hyperparameter combination, the new hyperparameter combination is updated using the optimal hyperparameter combination, and the HEMET model is updated using the new HEMET model.

2. The privacy-preserving network generation method based on Bayesian optimization according to claim 1, characterized in that: The improved acquisition function is: ； Among them, EI is the expected improvement function, is the weight coefficient, is the currently known optimal objective function value, is the objective function value obtained in the previous round of calculation, is the hyperparameter combination of the HEMET model, is the total number of parameters of the HEMET model, The parameter threshold of the HEMET model is set.

3. The privacy-preserving network generation method based on Bayesian optimization according to claim 2, characterized in that: In the improved acquisition function, if ,but , the candidate solutions are filtered; if , the penalty term degenerates into a constant, maintaining the optimization logic of traditional EI, where the adjustment To achieve Pareto optimization of accuracy-parameter quantity.

4. The privacy-preserving network generation method based on Bayesian optimization according to claim 1, characterized in that: The improved acquisition function is calculated by using the proxy model to obtain the calculation formula for the next possible optimal hyperparameter combination. include: ； in, is the next possible optimal hyperparameter combination.

5. The privacy-preserving network generation method based on Bayesian optimization according to claim 1, characterized in that: The new HEMET model is trained and evaluated on the test set, and the accuracy of the test set is obtained as follows: Compile a new HEMET model using the Adam optimizer , using the validation set To verify, in the test set Upper Assessment , get the accuracy of the test set .

6. The privacy-preserving network generation method based on Bayesian optimization according to claim 1, characterized in that: Also includes: The optimal hyperparameter combination and the updated new HEMET model are output.

7. A privacy-preserving network generation system based on Bayesian optimization, characterized in that: include: The model preparation module is used to build the HEMET model and embed the parameter quantity constraint into the Bayesian optimization acquisition function through exponential penalty to obtain the improved acquisition function, and use the sample set to build the proxy model; An acquisition function calculation module, used to calculate the improved acquisition function using the proxy model to obtain the next possible optimal hyperparameter combination; A model building and verification module is used to build and verify a new HEMET model using the optimal hyperparameter combination and calculate its parameter quantity; A model training and evaluation module, configured to continue iterating the new HEMET model if the parameter amount is greater than a preset parameter amount threshold of the HEMET model, and to train and evaluate the new HEMET model on a test set if the parameter amount is less than or equal to the parameter amount threshold to obtain the accuracy of the test set; A sample updating module, used for adding new hyperparameter combinations and their corresponding accuracy rates from the HEMET to the sample set; The optimal solution iteration module is used to update the new hyperparameter combination using the optimal hyperparameter combination if the accuracy of the test set is greater than the accuracy of the new hyperparameter combination, and to update the HEMET model using the new HEMET model.

8. An electronic device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the method described in any one of claims 1 to 6 is implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method described in any one of claims 1 to 6 is implemented.

10. A computer program, characterized in that Used to perform the method according to any one of claims 1 to 6.

Images