Model data storage method, device, electronic device and computer readable medium

Abstract

The embodiments of the present disclosure disclose a model data storage method, device, electronic device and computer-readable medium. A specific implementation of the method includes: obtaining a model training data set; determining the importance information of the data set; determining the encryption method; performing data encryption processing on each model training data to generate encryption information; determining at least one usage model; performing encryption processing on at least one model source file to generate at least one encrypted model source file information; setting at least one connection identifier; storing the model training data set and the encryption information in a first storage center, and storing at least one usage model and at least one encrypted model source file information in a second storage center. This implementation can realize the encryption processing of the model training data set and the corresponding at least one usage model, effectively perform the confidentiality protection of the model training data and the model source file, and facilitate the use of subsequent training data and model source files.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to the field of computer technology, and in particular to a model data storage method, device, electronic device, and computer-readable medium. Background Art

[0002] At present, with the continuous development of artificial intelligence, model training is also a major development direction in artificial intelligence. For model training, the confidentiality protection of model training data sets and model source files is also an aspect that needs to be paid attention to during the model training process and model application process. For the confidentiality protection of model training data and model source files, the commonly used method is to directly use relevant desensitization algorithms (for example, AES algorithm, differential privacy algorithm) to protect the confidentiality of model training data and model source files.

[0003] However, when using the above method to protect the model training data and model source files, the following technical problems often occur:

[0004] The logic of the relevant desensitizing algorithm is relatively simple, and it is troublesome to jointly retrieve and use the desensitized model training data and the desensitized model source files, resulting in low application efficiency of the model training data and model source files.

[0005] In the process of adopting technical solutions to solve the above technical problem 1, the following technical problem often arises: How to realize data encryption processing based on neural networks has become the current main development direction. For the above technical problems, the conventional solution is generally to realize data encryption processing through BP neural network. However, the above conventional solution still has the following problem 2: the accuracy of the encrypted data obtained is not accurate enough.

[0006] The above information disclosed in this Background section is only for enhancement of understanding of the background of the inventive concept and therefore it may contain information that does not form the prior art that is already known in this country to a person of ordinary skill in the art. Summary of the invention

[0007] The content of this disclosure is used to introduce concepts in a brief form, which will be described in detail in the detailed implementation section below. The content of this disclosure is not intended to identify the key features or essential features of the technical solution claimed for protection, nor is it intended to limit the scope of the technical solution claimed for protection.

[0008] Some embodiments of the present disclosure propose a model data storage method, device, electronic device, and computer-readable medium to solve one or more of the technical problems mentioned in the above background technology section.

[0009] In a first aspect, some embodiments of the present disclosure provide a model data storage method, including: obtaining a model training data set; determining data set importance information corresponding to the above-mentioned model training data set; determining the encryption method corresponding to the above-mentioned model training data set based on the above-mentioned data set importance information; performing data encryption processing on each model training data in the above-mentioned model training data set according to the above-mentioned encryption method to generate encryption information; determining at least one usage model corresponding to the above-mentioned model training data set; encrypting at least one model source file corresponding to the above-mentioned at least one usage model to generate at least one encrypted model source file information; setting at least one connection identifier for the above-mentioned model training data set and the above-mentioned at least one usage model; storing the above-mentioned model training data set and the above-mentioned encryption information in a first storage center, and storing the above-mentioned at least one usage model and the above-mentioned at least one encrypted model source file information in a second storage center, wherein the above-mentioned first storage center and the above-mentioned second storage center communicate through a connection identifier request, and the above-mentioned connection identifier request is generated based on the connection identifier.

[0010] In a second aspect, some embodiments of the present disclosure provide a model data storage device, comprising: an acquisition unit, configured to acquire a model training data set; a first determination unit, configured to determine data set importance information corresponding to the above-mentioned model training data set; a second determination unit, configured to determine the encryption method corresponding to the above-mentioned model training data set according to the above-mentioned data set importance information; a data encryption processing unit, configured to perform data encryption processing on each model training data in the above-mentioned model training data set according to the above-mentioned encryption method to generate encryption information; a third determination unit, configured to determine at least one usage model corresponding to the above-mentioned model training data set; an encryption unit, configured to perform encryption processing on at least one model source file corresponding to the above-mentioned at least one usage model to generate at least one encrypted model source file information; a setting unit, configured to set at least one connection identifier for the above-mentioned model training data set and the above-mentioned at least one usage model; a storage unit, configured to store the above-mentioned model training data set and the above-mentioned encryption information in a first storage center, and store the above-mentioned at least one usage model and the above-mentioned at least one encrypted model source file information in a second storage center, wherein the above-mentioned first storage center and the above-mentioned second storage center communicate through a connection identifier request, and the above-mentioned connection identifier request is generated based on the connection identifier.

[0011] In a third aspect, some embodiments of the present disclosure provide an electronic device comprising: one or more processors; a storage device on which one or more programs are stored, and when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation manner in the first aspect.

[0012] In a fourth aspect, some embodiments of the present disclosure provide a computer-readable medium having a computer program stored thereon, wherein when the program is executed by a processor, the method described in any implementation manner in the first aspect is implemented.

[0013] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the model data storage method of some embodiments of the present disclosure, the encryption processing of the model training data set and the corresponding at least one usage model can be realized, and the confidentiality protection of the model training data and the model source file can be effectively performed, which is convenient for the subsequent use of the training data and the model source file. Specifically, the reason why the confidentiality protection corresponding to the encryption of the relevant training data and the model source file is not effective enough and the subsequent use is not efficient enough is that the logic of the relevant desensitization algorithm is relatively simple, and the common retrieval and use between the desensitized model training data and the desensitized model source file is relatively troublesome, resulting in low application efficiency of the model training data and the model source file. Based on this, the model data storage method of some embodiments of the present disclosure first obtains the model training data set as the data object for subsequent encryption and corresponding storage, which is convenient for the subsequent use of the data set. Then, the data set importance information corresponding to the above-mentioned model training data set is determined. Here, the determined data set importance information can be used to determine the encryption method later. Different data set importance information selects the encryption method corresponding to the confidentiality level to achieve adaptive encryption of resources. Then, according to the above-mentioned data set importance information, the encryption method corresponding to the above-mentioned model training data set can be accurately determined to achieve corresponding confidentiality protection according to the importance of the model training data set. Then, according to the above-mentioned encryption method, each model training data in the above-mentioned model training data set can be accurately encrypted to generate encrypted information, which can achieve targeted confidentiality protection. Secondly, at least one usage model corresponding to the above-mentioned model training data set is determined to facilitate the subsequent use of the model training data set and the retrieval of the corresponding usage model. Further, at least one model source file corresponding to the above-mentioned at least one usage model is encrypted to accurately generate at least one encrypted model source file information to achieve the protection of the model source file. Furthermore, at least one connection identifier is set for the above-mentioned model training data set and the above-mentioned at least one usage model to facilitate the subsequent retrieval of the model and the model training data set. Finally, the above-mentioned model training data set and the above-mentioned encryption information are stored in the first storage center, and the above-mentioned at least one usage model and the above-mentioned at least one encrypted model source file information are stored in the second storage center. Among them, the above-mentioned first storage center and the above-mentioned second storage center communicate through a connection identifier request, and the above-mentioned connection identifier request is generated based on the connection identifier. Here, by differentiating the storage in the first storage center and the second storage center, effective storage and confidentiality protection of the model training data set and the usage model can be achieved. In addition, by connecting the identification request, the model training data set and the usage model can be selectively obtained, which greatly improves the acquisition efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The above and other features, advantages and aspects of the embodiments of the present disclosure will become more apparent with reference to the following detailed description in conjunction with the accompanying drawings. Throughout the accompanying drawings, the same or similar reference numerals represent the same or similar elements. It should be understood that the drawings are schematic and that components and elements are not necessarily drawn to scale.

[0015] Figure 1 is a flow chart of some embodiments of the model data storage method according to the present disclosure;

[0016] Figure 2 is a schematic diagram of the structure of some embodiments of the model data storage device according to the present disclosure;

[0017] Figure 3 It is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. DETAILED DESCRIPTION

[0018] Embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments set forth herein. On the contrary, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0019] It should also be noted that, for ease of description, only the parts related to the invention are shown in the drawings. In the absence of conflict, the embodiments and features in the embodiments of the present disclosure can be combined with each other.

[0020] It should be noted that the concepts such as "first" and "second" mentioned in the present disclosure are only used to distinguish different devices, modules or units, and are not used to limit the order or interdependence of the functions performed by these devices, modules or units.

[0021] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0022] The names of the messages or information exchanged between multiple devices in the embodiments of the present disclosure are only used for illustrative purposes and are not used to limit the scope of these messages or information.

[0023] The present disclosure will be described in detail below with reference to the accompanying drawings and in conjunction with embodiments.

[0024] refer to Figure 1, shows a process 100 of some embodiments of the model data storage method according to the present disclosure. The model data storage method comprises the following steps:

[0025] Step 101, obtaining a model training data set.

[0026] In some embodiments, the execution subject of the above-mentioned model data storage method can obtain the model training data set through a wired connection method or a wireless connection method. Among them, the model training data set can be a training data set for training a neural network model. For example, the model training data set can be a face image training data set.

[0027] Step 102: Determine the data set importance information corresponding to the above model training data set.

[0028] In some embodiments, the execution entity may determine the data set importance information corresponding to the model training data set. The data set importance information may represent the data importance of each model training data in the model training data set. The data set importance information may be a value between 0 and 1. The larger the value, the higher the data importance of each model training data in the corresponding model training data set.

[0029] As an example, first, the execution subject may determine the data quality and data content corresponding to the model training data set. Then, determine the first importance information corresponding to the data quality, and determine the second importance information corresponding to the data content. Finally, the first importance information and the second importance information are averaged to generate the data set importance information.

[0030] Step 103: Determine the encryption method corresponding to the model training data set according to the importance information of the data set.

[0031] In some embodiments, the execution entity may determine the encryption method corresponding to the model training data set according to the importance information of the data set. The encryption method may be a method that characterizes how to encrypt the model training data set. In practice, the encryption methods may include: an encryption method based on a data set label and a data content encryption method. The encryption method based on a data set label may be a method of encrypting a label for a data set label corresponding to a data set. Only by entering the correct encrypted data set label can the model training data set corresponding to the data set label be obtained. The data content encryption method may be a method of encrypting each data in the data set separately.

[0032] As an example, the execution subject may determine the encryption method corresponding to the importance information of the data set by means of an information association table. The information association table may represent the correspondence between the importance of the data set and the encryption method.

[0033] Step 104: perform data encryption processing on each model training data set in the model training data set according to the encryption method to generate encrypted information.

[0034] In some embodiments, the above-mentioned execution entity can execute encryption logic corresponding to the encryption method to perform data encryption processing on each model training data in the above-mentioned model training data set to generate encrypted information.

[0035] In some optional implementations of some embodiments, in response to determining that the encryption method is an encryption method based on a data set label, the execution subject may use a pre-trained encoding and decoding model to encrypt the data set label corresponding to the model training data set to generate the encrypted information. The data set label includes: at least one keyword corresponding to the model training data set. The encoding and decoding model may be a pre-trained model. In practice, the encoding and decoding model includes: an encoding model and a decoding model. The encoding model may be a neural network model that encodes each input data. The input of the encoding model may be specifically set according to different encryption methods. For example, the input of the encoding model may be a data set label, or it may be each model training data. The output of the encoding model may be each encoding matrix corresponding to each model training data. The encoding matrix may represent the data feature semantics of the corresponding model training data. The encoding model may include: multiple cascaded convolutional layers. The decoding model may be a neural network model that decodes each input matrix. The input of the decoding model may be each input matrix. For example, each input matrix may be each encoding matrix, or it may be the output result of the encryption model. The output of the decoding model may be each decoding matrix corresponding to each model training data. The decoding matrix can represent the characteristic semantic information of the output target corresponding to the model training data. For example, for the graph generation model, the encoding and decoding model is a model for generating the target scene graph. The output of the corresponding decoding model can represent the output scene graph corresponding to the model training data. For example, the decoding model can include: multiple cascaded convolutional layers. The model training method of the encoding and decoding model can be a conventional training method based on the gradient descent method.

[0036] As an example, first, the above-mentioned execution entity can input the data set label into the encoding model in the encoding and decoding model to output each first encoding matrix. Then, each first encoding matrix in each first encoding matrix is ​​average pooled to generate a first pooling result set. Finally, each first pooling result in the first pooling result set is spliced ​​to generate encrypted information. That is, in the training process of the encoding and decoding model, it is necessary to perform model training based on the decoding model, and in the process of model application, the output of the encoding model is used as the encryption basic data to realize the generation of encrypted information.

[0037] In some optional implementations of some embodiments, in response to determining that the above-mentioned encryption method is a data content encryption method, a pre-trained encoding and decoding model is used to perform data encryption processing on each model training data set in the above-mentioned model training data set to generate an encrypted data set as the above-mentioned encryption information.

[0038] As an example, first, the execution subject may input each model training data into the encoding model in the encoding and decoding model to output each second encoding matrix. Then, average pooling is performed on each second encoding matrix in each second encoding matrix to generate a second pooling result set. Finally, each second pooling result in the second pooling result set is concatenated to generate encrypted information.

[0039] In some optional implementations of some embodiments, the encoding and decoding models include: an encoding model, an encryption model based on hash encryption, and a decoding model. Among them, the encryption model based on hash encryption can be a neural network model that implements the principle of hash encryption. Specifically, the encryption model based on hash encryption can be a multi-layer cascaded fully connected layer (Fully Connected Layer). The multi-layer cascaded fully connected layer can be pre-trained based on multiple hash training data sets. The input of multiple cascaded fully connected layers can be a matrix or array to be hashed.

[0040] Optionally, the above-mentioned use of the pre-trained encoding and decoding model to encrypt the data set label corresponding to the above-mentioned model training data set to generate the above-mentioned encrypted information may include the following steps:

[0041] The first step is to input the dataset label into the encoding model to generate at least one first label encoding information. The encoding model is at least one first convolutional neural network connected in series. There is a one-to-one correspondence between the first convolutional neural network in the at least one first convolutional neural network and the first label encoding information in the at least one first label encoding information.

[0042] The second step is to determine the first label encoding information at the target position in the at least one first label encoding information as the target label encoding information. The target position may be the position of the last layer of the at least one first convolutional neural network in series. The target label encoding information may be encoding information in the form of a matrix.

[0043] The third step is to perform average pooling on the above target label encoding information to generate pooling information.

[0044] The fourth step is to use the encryption model to encrypt each matrix element included in the pooling information to generate an encrypted matrix as the model encryption information.

[0045] The fifth step is to input the model encryption information into the decoding model to generate at least one first label decoding information. The decoding model is at least one second convolutional neural network connected in series. There is a one-to-one correspondence between the at least one second convolutional neural network and the at least one first label decoding information.

[0046] Step 6: Select the label decoding information of the second position from the at least one first label decoding information as the target label decoding information. The second position may be a pre-selected position in at least one position corresponding to at least one label decoding information. For example, the second position may be a position corresponding to the second first label decoding information in at least one first label decoding information. The target label decoding information may be information in the form of a matrix.

[0047] The seventh step is to combine the second position and the target tag decoding information to generate the encrypted information.

[0048] As an example, first, the execution subject may encode the second position to generate a matrix with the same matrix dimension as the target tag decoding information as the position matrix. Then, the position matrix and the target tag decoding information are combined to generate a combined matrix as the encrypted information.

[0049] Considering the problems of the above conventional solutions, facing the above technical problem 2: the accuracy of the value certificate value obtained is not accurate enough. Combining the advantages / technical status, we can decide to adopt the following solution.

[0050] In some optional implementations of some embodiments, the above-mentioned use of the pre-trained encoding and decoding model to perform data encryption processing on each model training data in the above-mentioned model training data set to generate an encrypted data set as the above-mentioned encryption information may include the following steps:

[0051] In the first step, each model training data in the above model training data set is subjected to the following generation steps:

[0052] Sub-step 1, input the above-mentioned model training data into the above-mentioned encoding and decoding model to generate at least one second label encoding information and at least one second label decoding information. Among them, the second label encoding information in the at least one second label encoding information has a one-to-one correspondence with the first convolutional neural network in the at least one first convolutional neural network. The second label decoding information in the at least one second label decoding information has a one-to-one correspondence with the second convolutional neural network in the at least one second convolutional neural network. The above-mentioned encoding and decoding model is a network model whose output corresponds to the business target under the business scenario corresponding to the model training data set.

[0053] Sub-step 2, determining a random number corresponding to the above-mentioned model training data. The random number is a value in [0, convolution value]. The convolution value can be the minimum value of the information value corresponding to at least one second label encoding information and the information value corresponding to at least one second label decoding information.

[0054] Sub-step 3: determining the random number of the second label encoding information in the at least one second label encoding information as the candidate encoding information. Next, determining the random number of the second label decoding information in the at least one second label decoding information as the candidate decoding information.

[0055] Sub-step 4, concatenating the candidate encoding information, the candidate decoding information and the above random number to generate encrypted data.

[0056] In the second step, the obtained encrypted data set is determined as the above-mentioned encrypted information.

[0057] The content in the above “optionally”, as an inventive point of the present disclosure, solves the technical problem mentioned in the background technology that “the accuracy of the encrypted data obtained is not accurate enough.” Based on this, the present disclosure can determine candidate encoding information and candidate decoding information through at least one second tag encoding information and at least one second tag decoding information, and the randomness of the random number, so as to accurately generate encrypted data subsequently by following the random number.

[0058] Step 105: Determine at least one usage model corresponding to the above model training data set.

[0059] In some embodiments, the execution subject may determine at least one usage model corresponding to the model training data set. The training data set corresponding to the usage model is a model training data set, and the usage model may be a neural network model that can be trained and directly used. Each usage model of at least one usage model has a corresponding model usage scenario. For example, at least one model usage scenario corresponding to at least one usage model includes: a face recognition scenario, a character recognition scenario, and a limb key point determination scenario.

[0060] Step 106: encrypt at least one model source file corresponding to the at least one usage model to generate at least one encrypted model source file information.

[0061] In some embodiments, the execution entity may encrypt at least one model source file corresponding to the at least one usage model to generate at least one encrypted model source file information. There is a one-to-one correspondence between the usage model in the at least one usage model and the model source file in the at least one model source file. The model source file may be a code source file of the usage model. The model source file may characterize the model structure and the corresponding model parameter set of the usage model. There is a one-to-one correspondence between the encrypted model source file information in the at least one encrypted model source file information and the usage model in the at least one usage model.

[0062] As an example, for each usage model, first, the above-mentioned execution entity can determine the model usage description information corresponding to the usage model. Among them, the model usage description information includes: model usage scenario, model input, and model output. Then, the scenario information corresponding to the model usage scenario, the input description information of the model input, and the output description information of the model output are respectively encoded to obtain the scene encoding information, the model input encoding information, and the model output encoding information. Finally, the scene encoding information, the model input encoding information, and the model output encoding information are combined into a matrix form to generate a combined matrix. Finally, for each row element group of the combined matrix, the averaging processing of each element in the element group is performed to generate an average value matrix as the encrypted model source file information.

[0063] Step 107: Set at least one connection identifier for the above-mentioned model training data set and the above-mentioned at least one usage model.

[0064] In some embodiments, the execution entity may set at least one connection identifier for the model training data set and the at least one usage model. The connection identifier may characterize the association between the usage model and the model training data set. The connection identifier may be information in a numerical form. There is a one-to-one correspondence between the connection identifier in at least one connection identifier and the usage model in at least one usage model. The connection identifier is used to subsequently generate a connection identifier request. The connection identifier request may be used for communication between the first storage center and the second storage center. That is, the connection identifier request may obtain the model training data set associated with the corresponding connection identifier in the first storage center. The connection identifier request may obtain the usage model associated with the corresponding connection identifier in the second storage center.

[0065] Step 108: store the model training data set and the encryption information in a first storage center, and store the at least one usage model and the at least one encrypted model source file information in a second storage center.

[0066] In some embodiments, the execution entity may store the model training data set and the encrypted information in a first storage center, and store the at least one usage model and the at least one encrypted model source file information in a second storage center. The first storage center and the second storage center communicate via a connection identifier request. The connection identifier request is based on the practice of generating a connection identifier, and the connection identifier request may be a request generated by filling a connection identifier into a preset request template.

[0067] In some optional implementations of some embodiments, after step 108, the steps further include:

[0068] The first step is to, in response to receiving a model usage request, send the model usage request to the second storage center to parse the model usage request and obtain the first model encryption information and data set requirement information.

[0069] The model use request may be a request to retrieve a model for use. The first model encryption information may be encryption information of a model source file corresponding to the model to be retrieved. The data set requirement information may indicate whether a model training data set corresponding to the model to be retrieved is required. For example, the data set requirement information may include: requirement information indicating that it is not necessary to obtain a model training data set corresponding to the model to be retrieved, and requirement information indicating that it is necessary to obtain a model training data set corresponding to the model to be retrieved.

[0070] As an example, the execution entity may utilize the preset first regular term to parse the model usage request to obtain the first model encryption information and data set requirement information.

[0071] The second step is to match the corresponding model source file in the second storage center according to the first model encryption information as the first model source file.

[0072] As an example, first, the execution subject may obtain the encrypted model source file information matching the first model encryption information from the second storage center as the first target encrypted model source file information. Then, the model source file corresponding to the first target encrypted model source file information is determined in the second storage center as the first model source file.

[0073] In a third step, in response to determining that the dataset requirement information represents the acquisition of the corresponding model training dataset, the dataset requirement information is parsed to generate first dataset encryption information.

[0074] As an example, the execution subject may use the preset second regularization item to parse the data set requirement information to generate the first data set encryption information.

[0075] The fourth step is to generate a first connection identification request for the first model encryption information and the first data set encryption information, wherein the first connection identification request may be request information for obtaining a model training data set corresponding to the first model encryption information and the first data set encryption information.

[0076] As an example, first, a request generation template corresponding to a connection identification request is obtained, and then the first model encryption information and the first data set encryption information are added to the request generation template to generate a first connection identification request.

[0077] In the fifth step, the first connection identifier request is sent to the first storage center to parse the first connection identifier request and obtain the first data set encryption information, wherein the first data set encryption information may be encryption information after the corresponding data set is encrypted.

[0078] Step 6: Match the first model training data set corresponding to the encrypted information of the first data set in the first storage center.

[0079] As an example, first, the first target encryption information matching the first data set encryption information is determined in the first storage center, and then the model training data set corresponding to the first target encryption information is determined as the first model training data set.

[0080] In some optional implementations of some embodiments, after step 108, the steps further include:

[0081] The first step is to, in response to receiving a data set use request, send the data set use request to the first storage center to parse the data set use request and obtain the second data set encryption information and model requirement information.

[0082] The data set use request may be a request for calling a data set. The second data set encryption information may be encryption information corresponding to the data set to be retrieved. The model requirement information may be requirement information indicating whether to obtain a corresponding use model file. The model requirement information may include: requirement information indicating whether to obtain a corresponding use model file, and requirement information indicating whether to not obtain a corresponding use model file.

[0083] The second step is to match the corresponding second model training data set in the first storage center according to the encrypted information of the second data set.

[0084] As an example, first, the second target encryption information corresponding to the second data set encryption information is determined in the first storage center, and then the second model training data set corresponding to the second target encryption information is determined.

[0085] In a third step, in response to determining that the above model requirement information represents obtaining a corresponding usage model file, the above model requirement information is parsed to generate second model encryption information.

[0086] As an example, the execution subject may use the preset third regularization term to parse the model requirement information to generate the second model encryption information. The second model encryption information may be encryption information of a model source file corresponding to the model to be retrieved.

[0087] Step 4: Generate a second connection identification request for the second data set encryption information and the second model encryption information. The second connection identification request may be request information for obtaining a model source file corresponding to the second model encryption information and the second data set encryption information.

[0088] As an example, first, a request generation template corresponding to a connection identification request is obtained, and then the second model encryption information and the second data set encryption information are added to the request generation template to generate a first connection identification request.

[0089] The fifth step is to send the second connection identification request to the second storage center to parse the second connection identification request and obtain the second model encryption information.

[0090] Step 6: Match the model source file corresponding to the second model encryption information in the second storage center as the second model source file.

[0091] As an example, first, the first target encrypted model source file information matching the second model encryption information is determined in the second storage center, and then the model source file corresponding to the first target encrypted model source file information is determined as the second model source file.

[0092] The above-mentioned embodiments of the present disclosure have the following beneficial effects: through the model data storage method of some embodiments of the present disclosure, the encryption processing of the model training data set and the corresponding at least one usage model can be realized, and the confidentiality protection of the model training data and the model source file can be effectively performed, which is convenient for the subsequent use of the training data and the model source file. Specifically, the reason why the confidentiality protection corresponding to the encryption of the relevant training data and the model source file is not effective enough and the subsequent use is not efficient enough is that the logic of the relevant desensitization algorithm is relatively simple, and the common retrieval and use between the desensitized model training data and the desensitized model source file is relatively troublesome, resulting in low application efficiency of the model training data and the model source file. Based on this, the model data storage method of some embodiments of the present disclosure first obtains the model training data set as the data object for subsequent encryption and corresponding storage, which is convenient for the subsequent use of the data set. Then, the data set importance information corresponding to the above-mentioned model training data set is determined. Here, the determined data set importance information can be used to determine the encryption method later. Different data set importance information selects the encryption method corresponding to the confidentiality level to achieve adaptive encryption of resources. Then, according to the above-mentioned data set importance information, the encryption method corresponding to the above-mentioned model training data set can be accurately determined to achieve corresponding confidentiality protection according to the importance of the model training data set. Then, according to the above-mentioned encryption method, each model training data in the above-mentioned model training data set can be accurately encrypted to generate encrypted information, which can achieve targeted confidentiality protection. Secondly, at least one usage model corresponding to the above-mentioned model training data set is determined to facilitate the subsequent use of the model training data set and the retrieval of the corresponding usage model. Further, at least one model source file corresponding to the above-mentioned at least one usage model is encrypted to accurately generate at least one encrypted model source file information to achieve the protection of the model source file. Furthermore, at least one connection identifier is set for the above-mentioned model training data set and the above-mentioned at least one usage model to facilitate the subsequent retrieval of the model and the model training data set. Finally, the above-mentioned model training data set and the above-mentioned encryption information are stored in the first storage center, and the above-mentioned at least one usage model and the above-mentioned at least one encrypted model source file information are stored in the second storage center. Among them, the above-mentioned first storage center and the above-mentioned second storage center communicate through a connection identifier request, and the above-mentioned connection identifier request is generated based on the connection identifier. Here, by differentiating the storage in the first storage center and the second storage center, effective storage and confidentiality protection of the model training data set and the usage model can be achieved. In addition, by connecting the identification request, the model training data set and the usage model can be selectively obtained, which greatly improves the acquisition efficiency.

[0093] Further references Figure 2As an implementation of the methods shown in the above figures, the present disclosure provides some embodiments of a model data storage device, and these device embodiments are Figure 1 Corresponding to the method embodiments shown, the model data storage device can be specifically applied to various electronic devices.

[0094] like Figure 2 As shown, a model data storage device 200 includes: an acquisition unit 201, a first determination unit 202, a second determination unit 203, a data encryption processing unit 204, a third determination unit 205, an encryption unit 206, a setting unit 207 and a storage unit 208. Among them, the acquisition unit 201 is configured to acquire a model training data set; the first determination unit 202 is configured to determine the data set importance information corresponding to the above-mentioned model training data set; the second determination unit 203 is configured to determine the encryption method corresponding to the above-mentioned model training data set according to the above-mentioned data set importance information; the data encryption processing unit 204 is configured to perform data encryption processing on each model training data in the above-mentioned model training data set according to the above-mentioned encryption method to generate encryption information; the third determination unit 205 is configured to determine at least one usage model corresponding to the above-mentioned model training data set; the encryption unit 206 is configured to encrypt at least one model source file corresponding to the above-mentioned at least one usage model to generate at least one encrypted model source file information; the setting unit 207 is configured to set a connection identifier for the above-mentioned model training data set and the above-mentioned at least one usage model; the storage unit 208 is configured to store the above-mentioned model training data set and the above-mentioned encryption information in a first storage center, and store the above-mentioned at least one usage model and the above-mentioned at least one encrypted model source file information in a second storage center, wherein the above-mentioned first storage center and the above-mentioned second storage center communicate through a connection identifier request, and the above-mentioned connection identifier request is generated based on the connection identifier.

[0095] It is understood that the units recorded in the model data storage device 200 and the reference Figure 1 Therefore, the operations, features and beneficial effects described above for the method are also applicable to the model data storage device 200 and the units contained therein, and will not be described in detail here.

[0096] Reference below Figure 3 , which shows a structural schematic diagram of an electronic device (eg, an electronic device) 300 suitable for implementing some embodiments of the present disclosure. Figure 3 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0097] like Figure 3As shown, the electronic device 300 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. In the RAM 303, various programs and data required for the operation of the electronic device 300 are also stored. The processing device 301, the ROM 302, and the RAM 303 are connected to each other via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0098] Typically, the following devices may be connected to the I / O interface 305: input devices 306 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; output devices 307 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; storage devices 308 including, for example, a magnetic tape, a hard disk, etc.; and communication devices 309. The communication devices 309 may allow the electronic device 300 to communicate with other devices wirelessly or by wire to exchange data. Although Figure 3 The electronic device 300 is shown with various devices, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead. Figure 3 Each block shown in the figure may represent one device, or may represent multiple devices as required.

[0099] In particular, according to some embodiments of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, some embodiments of the present disclosure include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In some such embodiments, the computer program can be downloaded and installed from the network through the communication device 309, or installed from the storage device 308, or installed from the ROM 302. When the computer program is executed by the processing device 301, the above-mentioned functions defined in the method of some embodiments of the present disclosure are executed.

[0100] It should be noted that the computer-readable medium in some embodiments of the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In some embodiments of the present disclosure, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In some embodiments of the present disclosure, the computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer readable signal medium may also be any computer readable medium other than a computer readable storage medium, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer readable medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0101] In some embodiments, the client and the server may communicate using any currently known or future developed network protocol such as HTTP (HyperText Transfer Protocol), and may be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or future developed network.

[0102] The computer-readable medium may be included in the electronic device; or it may exist independently without being installed in the electronic device. The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device: obtains a model training data set; determines the data set importance information corresponding to the model training data set; determines the encryption method corresponding to the model training data set according to the data set importance information; performs data encryption processing on each model training data in the model training data set according to the encryption method to generate encryption information; determines at least one usage model corresponding to the model training data set; performs encryption processing on at least one model source file corresponding to the at least one usage model to generate at least one encrypted model source file information; sets at least one connection identifier for the model training data set and the at least one usage model; stores the model training data set and the encryption information in a first storage center, and stores the at least one usage model and the at least one encrypted model source file information in a second storage center, wherein the first storage center and the second storage center communicate through a connection identifier request, and the connection identifier request is generated based on the connection identifier.

[0103] Computer program code for performing the operations of some embodiments of the present disclosure may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0104] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present disclosure. In this regard, each square box in the flow chart or block diagram can represent a module, a program segment or a part of a code, and the module, the program segment or a part of the code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some implementations as replacements, the functions marked in the square box can also occur in a sequence different from that marked in the accompanying drawings. For example, two square boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each square box in the block diagram and / or flow chart, and the combination of the square boxes in the block diagram and / or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0105] The units described in some embodiments of the present disclosure may be implemented by software or by hardware. The described units may also be provided in a processor, for example, may be described as: a processor including an acquisition unit, a first determination unit, a second determination unit, a data encryption processing unit, a third determination unit, an encryption unit, a setting unit, and a storage unit. The names of these units do not, in some cases, constitute limitations on the units themselves, for example, the acquisition unit may also be described as a "unit for acquiring a model training data set".

[0106] The functions described above herein may be performed at least in part by one or more hardware logic components. For example, without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), and the like.

[0107] The above descriptions are only some preferred embodiments of the present disclosure and an explanation of the technical principles used. Those skilled in the art should understand that the scope of the invention involved in the embodiments of the present disclosure is not limited to the technical solutions formed by a specific combination of the above-mentioned technical features, but should also cover other technical solutions formed by any combination of the above-mentioned technical features or their equivalent features without departing from the above-mentioned inventive concept. For example, the above-mentioned features are replaced with the technical features with similar functions disclosed in the embodiments of the present disclosure (but not limited to) and the technical solutions formed.

Claims

1. A model data storage method, comprising: Get the model training dataset; Determining data set importance information corresponding to the model training data set; Determining an encryption method corresponding to the model training data set according to the importance information of the data set; According to the encryption method, data encryption processing is performed on each model training data set in the model training data set to generate encrypted information, wherein: The step of performing data encryption processing on each model training data in the model training data set according to the encryption method to generate encrypted information includes: In response to determining that the encryption method is a data content encryption method, using a pre-trained encoding and decoding model, each model training data in the model training data set is subjected to data encryption processing to generate an encrypted data set as the encryption information, wherein: The method of using the pre-trained encoding and decoding model to perform data encryption processing on each model training data in the model training data set to generate an encrypted data set as the encryption information includes: For each model training data in the model training data set, the following generation steps are performed: Inputting the model training data into the encoding and decoding model to generate at least one second label encoding information and at least one second label decoding information, wherein the second label encoding information in the at least one second label encoding information has a one-to-one correspondence with the first convolutional neural network in the at least one first convolutional neural network, and the second label decoding information in the at least one second label decoding information has a one-to-one correspondence with the second convolutional neural network in the at least one second convolutional neural network, and the encoding and decoding model is a network model whose output corresponds to the business target under the business scenario corresponding to the model training data set; Determine a random number corresponding to the model training data, wherein the random number is a value in [0, convolution value], and the convolution value is a minimum value of an information value corresponding to at least one second label encoding information and an information value corresponding to at least one second label decoding information; Determine a random number of second label encoding information among the at least one second label encoding information as candidate encoding information; Determine a random number of second label decoding information among the at least one second label decoding information as candidate decoding information; Concatenating the candidate encoding information, the candidate decoding information and the random number to generate encrypted data; Determining the obtained encrypted data set as the encrypted information; Determine at least one usage model corresponding to the model training data set; Encrypting at least one model source file corresponding to the at least one usage model to generate at least one encrypted model source file information; Setting at least one connection identifier for the model training data set and the at least one usage model; The model training data set and the encrypted information are stored in a first storage center, and the at least one usage model and the at least one encrypted model source file information are stored in a second storage center, wherein the first storage center and the second storage center communicate via a connection identifier request, and the connection identifier request is generated based on the connection identifier.

2. The method according to claim 1, wherein: The step of performing data encryption processing on each model training data in the model training data set according to the encryption method to generate encrypted information includes: In response to determining that the encryption method is an encryption method based on a data set label, a pre-trained encoding and decoding model is used to encrypt the data set label corresponding to the model training data set to generate the encrypted information, wherein the data set label includes: at least one keyword corresponding to the model training data set.

3. The method according to claim 1, wherein: The method further comprises: In response to receiving the model use request, sending the model use request to the second storage center to parse the model use request and obtain the first model encryption information and data set requirement information; According to the first model encryption information, matching the corresponding model source file in the second storage center as the first model source file; In response to determining that the data set requirement information represents obtaining a corresponding model training data set, the data set requirement information is parsed to generate first data set encryption information; Generate a first connection identification request for the first model encryption information and the first data set encryption information; Sending the first connection identifier request to the first storage center to parse the first connection identifier request and obtain first data set encryption information; A first model training data set corresponding to the first data set encryption information is matched in the first storage center.

4. The method according to claim 1, wherein: The method further comprises: In response to receiving the data set use request, sending the data set use request to the first storage center to parse the data set use request and obtain the second data set encryption information and model requirement information; According to the second data set encryption information, matching the corresponding second model training data set in the first storage center; In response to determining that the model requirement information represents obtaining a corresponding usage model file, the model requirement information is parsed to generate second model encryption information; generating a second connection identification request for the second data set encryption information and the second model encryption information; Sending the second connection identifier request to the second storage center to parse the second connection identifier request and obtain the second model encryption information; The model source file corresponding to the second model encryption information is matched in the second storage center as the second model source file.

5. The method according to claim 1, wherein: The encoding and decoding models include: an encoding model, an encryption model based on hash encryption, and a decoding model; and The using of the pre-trained encoding and decoding model to encrypt the data set label corresponding to the model training data set to generate the encrypted information includes: Inputting the data set label into the encoding model to generate at least one first label encoding information, wherein the encoding model is at least one first convolutional neural network connected in series, and there is a one-to-one correspondence between the at least one first convolutional neural network and the at least one first label encoding information; Determine the first label encoding information at the target position among the at least one first label encoding information as the target label encoding information; Performing average pooling on the target label encoding information to generate pooling information; Using the encryption model, encrypt each matrix element included in the pooling information to generate an encrypted matrix as the model encryption information; Inputting the model encryption information into the decoding model to generate at least one first label decoding information, wherein the decoding model is at least one second convolutional neural network connected in series, and there is a one-to-one correspondence between the at least one second convolutional neural network and the at least one first label decoding information; Selecting the label decoding information at the second position from the at least one first label decoding information as the target label decoding information; The second position and the target tag decoding information are combined to generate the encrypted information.

6. A model data storage device, comprising: An acquisition unit, configured to acquire a model training data set; A first determining unit is configured to determine data set importance information corresponding to the model training data set; A second determining unit is configured to determine an encryption method corresponding to the model training data set according to the data set importance information; A data encryption processing unit is configured to perform data encryption processing on each model training data in the model training data set according to the encryption method to generate encryption information, wherein the data encryption processing on each model training data in the model training data set according to the encryption method to generate encryption information includes: in response to determining that the encryption method is a data content encryption method, using a pre-trained encoding and decoding model to perform data encryption processing on each model training data in the model training data set to generate an encrypted data set as the encryption information, wherein the data encryption processing on each model training data in the model training data set using a pre-trained encoding and decoding model to generate an encrypted data set as the encryption information includes: for each model training data in the model training data set, perform the following generation steps: input the model training data into the encoding and decoding model to generate at least one second label encoding information and at least one second label decoding information, wherein at least one second label encoding The second label encoding information in the code information has a one-to-one correspondence with the first convolutional neural network in at least one first convolutional neural network, and the second label decoding information in at least one second label decoding information has a one-to-one correspondence with the second convolutional neural network in at least one second convolutional neural network. The encoding and decoding model is a network model whose output corresponds to the business target in the business scenario corresponding to the model training data set; determining a random number corresponding to the model training data, wherein the random number is a value in [0, convolution value], and the convolution value is the minimum value of the information value corresponding to the at least one second label encoding information and the information value corresponding to the at least one second label decoding information; determining the random number of the second label encoding information in the at least one second label encoding information as the candidate encoding information; determining the random number of the second label decoding information in the at least one second label decoding information as the candidate decoding information; splicing the candidate encoding information, the candidate decoding information and the random number to generate encrypted data; determining the obtained encrypted data set as the encrypted information; A third determining unit is configured to determine at least one usage model corresponding to the model training data set; an encryption unit, configured to perform encryption processing on at least one model source file corresponding to the at least one usage model to generate at least one encrypted model source file information; A setting unit, configured to set a connection identifier for the model training data set and the at least one usage model; A storage unit is configured to store the model training data set and the encrypted information in a first storage center, and to store the at least one usage model and the at least one encrypted model source file information in a second storage center, wherein the first storage center and the second storage center communicate via a connection identifier request, and the connection identifier request is generated based on the connection identifier.

7. An electronic device comprising: one or more processors; a storage device having one or more programs stored thereon, When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 5.

8. A computer readable medium having a computer program stored thereon, wherein: When the program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

Images