A data processing method, device and equipment based on privacy protection

Abstract

The embodiments of this specification disclose a data processing method, device, and equipment based on privacy protection. The method includes: obtaining target data to be processed, where the target data includes continuous structured data and / or image data. Then, differential privacy processing can be performed on the target data respectively through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget. Based on the differential privacy results corresponding to a variety of different privacy budgets, and through a preset contrastive learning loss algorithm, the target model is trained to obtain a trained target model.

Description

Technical Field

[0001] This document relates to the field of computer technology, and particularly to a data processing method, device, and equipment based on privacy protection. Background Art

[0002] Deep learning can often obtain models with excellent performance through complex large architectures and huge data scales. The training samples used therein often need to be labeled to obtain the label data of each training sample. Since the label data of training samples often requires a high labeling cost in practical applications, unsupervised learning or semi-supervised learning has become an important technical tool in such cases. However, in unsupervised learning or semi-supervised learning, a large number of unlabeled training samples are easy to obtain while also increasing privacy risks. For example, in unsupervised learning, during the process of training a model, it may be allowed for an attacker to extract the original training samples by querying the model, thus causing personal privacy data to be stolen. In the current situation where data compliance is becoming more and more strict, privacy protection in unsupervised learning or semi-supervised learning has received increasing attention. Contrastive learning is a currently commonly used pre-training paradigm and also a typical self-supervised technique. Contrastive learning can achieve self-supervised learning by maximizing the different data augmentations of two identical underlying objects. Therefore, it is necessary to provide a technical solution for privacy protection processing in contrastive learning to achieve pre-training without revealing user privacy. Summary of the Invention

[0003] The purpose of the embodiments of this specification is to provide a technical solution for privacy protection processing in contrastive learning to achieve pre-training without revealing user privacy.

[0004] To achieve the above technical solution, the embodiments of this specification are implemented as follows:

[0005] A data processing method based on privacy protection provided by the embodiments of this specification, the method includes: obtaining target data to be processed, where the target data includes continuous structured data and / or image data. Performing differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget. Training a target model based on the differential privacy results corresponding to a variety of different privacy budgets and through a preset contrastive learning loss algorithm to obtain a trained target model.

[0006] A data processing method based on privacy protection provided by an embodiment of this specification is applied to a blockchain system. The method includes: obtaining rule information for contrastive learning based on privacy protection, generating a corresponding first smart contract using the rule information for contrastive learning based on privacy protection, and deploying the first smart contract to the blockchain system. When obtaining target data to be processed, calling the first smart contract, and respectively performing differential privacy processing on the target data through a plurality of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget, where the target data includes continuous structured data and / or image data. Calling the first smart contract, and training a target model based on the differential privacy results corresponding to a plurality of different privacy budgets through a preset contrastive learning loss algorithm to obtain a trained target model.

[0007] A data processing device based on privacy protection provided by an embodiment of this specification includes: a data acquisition module, which acquires target data to be processed, where the target data includes continuous structured data and / or image data. A differential privacy module, which respectively performs differential privacy processing on the target data through a plurality of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget. A contrastive learning module, which trains a target model based on the differential privacy results corresponding to a plurality of different privacy budgets and through a preset contrastive learning loss algorithm to obtain a trained target model.

[0008] A data processing device based on privacy protection provided by an embodiment of this specification is a device in a blockchain system. The device includes: a contract deployment module, which obtains rule information for contrastive learning based on privacy protection, generates a corresponding first smart contract using the rule information for contrastive learning based on privacy protection, and deploys the first smart contract to the blockchain system. A differential privacy module, which when obtaining target data to be processed, calls the first smart contract, and respectively performs differential privacy processing on the target data through a plurality of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget, where the target data includes continuous structured data and / or image data. A contrastive learning module, which calls the first smart contract, and trains a target model based on the differential privacy results corresponding to a plurality of different privacy budgets through a preset contrastive learning loss algorithm to obtain a trained target model.

[0009] A data processing device based on privacy protection provided by an embodiment of this specification includes: a processor; and a memory arranged to store computer-executable instructions, and when the executable instructions are executed, the processor: obtains target data to be processed, and the target data includes continuous structured data and / or image data. Through a variety of different preset privacy budgets, differential privacy processing is respectively performed on the target data to obtain differential privacy results corresponding to each privacy budget. Based on the differential privacy results corresponding to a variety of different privacy budgets, and through a preset contrastive learning loss algorithm, a target model is trained to obtain a trained target model.

[0010] A data processing device based on privacy protection provided by an embodiment of this specification, the device is a device in a blockchain system, and the data processing device based on privacy protection includes: a processor; and a memory arranged to store computer-executable instructions, and when the executable instructions are executed, the processor: obtains rule information for contrastive learning based on privacy protection, generates a corresponding first smart contract by using the rule information for contrastive learning based on privacy protection, and deploys the first smart contract to the blockchain system. When obtaining target data to be processed, the first smart contract is called, and through a variety of different preset privacy budgets, differential privacy processing is respectively performed on the target data to obtain differential privacy results corresponding to each privacy budget, and the target data includes continuous structured data and / or image data. The first smart contract is called, and based on the differential privacy results corresponding to a variety of different privacy budgets, a target model is trained through a preset contrastive learning loss algorithm to obtain a trained target model.

[0011] An embodiment of this specification also provides a storage medium, wherein the storage medium is used to store computer-executable instructions, and when the executable instructions are executed, the following process is implemented: obtaining target data to be processed, and the target data includes continuous structured data and / or image data. Through a variety of different preset privacy budgets, differential privacy processing is respectively performed on the target data to obtain differential privacy results corresponding to each privacy budget. Based on the differential privacy results corresponding to a variety of different privacy budgets, and through a preset contrastive learning loss algorithm, a target model is trained to obtain a trained target model.

[0012] An embodiment of this specification also provides a storage medium, where the storage medium is used to store computer-executable instructions, and when the executable instructions are executed, the following processes are implemented: obtaining rule information for contrastive learning based on privacy protection, generating a corresponding first smart contract using the rule information for contrastive learning based on privacy protection, and deploying the first smart contract to a blockchain system. When target data to be processed is obtained, the first smart contract is called, and differential privacy processing is respectively performed on the target data through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget, where the target data includes continuous structured data and / or image data. The first smart contract is called, and based on the differential privacy results corresponding to a variety of different privacy budgets, a target model is trained through a preset contrastive learning loss algorithm to obtain a trained target model. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] To more clearly illustrate the technical solutions in the embodiments of this specification or the prior art, the following will briefly introduce the drawings required for use in the description of the embodiments or the prior art. Obviously, the drawings in the following description are only some embodiments recorded in this specification. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 This is an embodiment of a data processing method based on privacy protection in this specification;

[0015] Figure 2 This is a schematic structural diagram of an interface related to data processing based on privacy protection in this specification;

[0016] Figure 3 This is another embodiment of a data processing method based on privacy protection in this specification;

[0017] Figure 4 This is yet another embodiment of a data processing method based on privacy protection in this specification;

[0018] Figure 5A This is yet another embodiment of a data processing method based on privacy protection in this specification;

[0019] Figure 5B This is a schematic diagram of a data processing process based on privacy protection in this specification;

[0020] Figure 6 This is an embodiment of a data processing device based on privacy protection in this specification;

[0021] Figure 7 This is another embodiment of a data processing device based on privacy protection in this specification;

[0022] Figure 8 This is an embodiment of a data processing device based on privacy protection in this specification. Specific implementation manners

[0023] The embodiments of this specification provide a data processing method, device and equipment based on privacy protection.

[0024] In order to enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings in the embodiments of this specification. Obviously, the described embodiments are only a part of the embodiments of this specification, rather than all the embodiments. Based on the embodiments in this specification, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of this specification.

[0025] Embodiment 1

[0026] As Figure 1 shown, the embodiments of this specification provide a data processing method based on privacy protection. The execution subject of this method can be a terminal device or a server. Among them, the terminal device can be a mobile terminal device such as a mobile phone or a tablet computer, or a device such as a personal computer. The server can be an independent server, or a server cluster composed of multiple servers. The server can be a background server such as a financial service or an online shopping service, or a background server of a certain application program. This method can be applied to relevant scenarios such as model training through contrastive learning. In this embodiment, the server is used as the execution subject for detailed description. For the case of the terminal device, reference can be made to the following relevant content and will not be elaborated here. This method can specifically include the following steps:

[0027] In step S102, obtain target data to be processed, where the target data includes continuous structured data and / or image data.

[0028] Among them, the target data can be any data in certain scenarios (such as relevant scenarios involving the financial field within a certain region or cross-regional data interaction, verification, risk prevention and control, etc.). In practical applications, the target data may include sensitive data or privacy data in the above scenarios, and the content of the target data includes continuous structured data and / or image data. The continuous structured data can be structured data with continuous data (that is, its data is not a single or isolated numerical value or number, but structured data containing several decimal places and dense value ranges), and can also be called continuous row data. It can be data logically expressed and implemented by a two-dimensional table structure, and strictly follows the data format and length specifications. The image data can include images and / or videos, etc. It should be noted that the continuous structured data and image data in this embodiment need to satisfy that for each dimension of the data, the value range of the data is bounded. For example, for image data, the value range of each pixel point can be 0-255, etc., and can be specifically set according to the actual situation. This specification embodiment does not make any limitations in this regard.

[0029] In implementation, deep learning can often obtain a model with excellent performance through complex large architectures and huge data scales. Among them, the training samples that need to be used often need to be labeled to obtain the label data of each training sample. Since the label data of the training samples often requires a high labeling cost in practical applications, unsupervised learning or semi-supervised learning has become an important technical tool in such cases. However, in unsupervised learning or semi-supervised learning, a large number of unlabeled training samples are easy to obtain while also increasing privacy risks. For example, in unsupervised learning, during the process of training the model, it may allow an attacker to extract the original training samples by querying the model, thus causing personal privacy data to be stolen. In the current situation where data compliance is becoming more and more strict, privacy protection in unsupervised learning or semi-supervised learning has received more and more attention. Contrastive learning is a currently commonly used pre-training paradigm and also a typical self-supervised technique. Contrastive learning can achieve self-supervised learning by maximizing the different data augmentations of two identical underlying objects. Therefore, it is necessary to provide a technical solution for privacy protection processing in contrastive learning to achieve pre-training without disclosing user privacy. The embodiments of this specification provide an achievable technical solution, which can specifically include the following content:

[0030] The target data to be processed can be obtained in a variety of different ways. For example, an input page with the target data can be preset in advance. The input page may include a data input box for the target data, an OK button, a cancel button, etc. When it is necessary to upload a certain data (i.e., the target data) to the server, the data on the above input page can be obtained and the input page can be displayed. As Figure 2 shown, the user can input the target data in the data input box on the data page. After the input is completed, the OK button on the input page can be clicked. At this time, the server can obtain the target data and use the target data as the target data to be processed. Alternatively, the relevant data of a certain service can be recorded in the server. When it is necessary to obtain the target data, the data that meets the specified requirements can be obtained from the relevant data of the above service and used as the target data. In addition to obtaining the target data through the above methods, the target data can also be obtained in a variety of different ways, which can be specifically set according to the actual situation. The embodiments of this specification do not limit this.

[0031] In step S104, differential privacy processing is performed on the target data respectively through a variety of different preset privacy budgets to obtain the differential privacy results corresponding to each privacy budget.

[0032] Among them, the privacy budget, as the core parameter in the differential privacy protection mechanism, determines both the protection level of differential privacy and the degree of privacy leakage. In the differential privacy protection mechanism, the privacy budget does not have to be uniformly used, but can be allocated multiple times. For example, the privacy budget can be divided into three parts, such as sub-privacy budget 1, sub-privacy budget 2, and sub-privacy budget 3, where sub-privacy budget 1 + sub-privacy budget 2 + sub-privacy budget 3 = privacy budget, and the numerical sizes of the three parts can be set according to the actual situation. A privacy budget can be set for each differential privacy process. In this way, multiple different privacy budgets can be preset. For example, two privacy budgets can be set, or three privacy budgets can be preset in advance, etc., which can be specifically set according to the actual situation. Differential privacy aims to protect the data collected above to a certain extent when the user whose data is collected does not trust the data collector, although the user still uploads the corresponding data to the data collector. Differential privacy can maximize the accuracy of data query while minimizing the probability of identifying its records when querying from a statistical database. Differential privacy achieves the purpose of protecting privacy by perturbing the data, and the perturbation mechanism can include various ones, such as the Laplace mechanism, the exponential mechanism, etc. Differential privacy can include centralized differential privacy and local differential privacy (Local Differential Privacy, LDP). Among them, local differential privacy is that before the data to be uploaded is collected, the user first perturbs the data to be uploaded locally, and then uploads the noisy data to be uploaded to the server (or service center). Local differential privacy can include the following definition: Algorithm A is local differential privacy (∈-LDP) that satisfies ∈, where ∈ ≥ 0, if and only if for any two data v and v', the following formula is satisfied:

[0033]

[0034] Among them, v and v' belong to the domain of A, and y belongs to the range of A. From the user's perspective, local differential privacy can better protect the privacy of user data. Before the user data is collected, it has been perturbed locally, and the privacy content in the user data has been erased.

[0035] In implementation, to avoid the leakage of private data contained in the target data to be processed in contrastive learning, differential privacy processing can be performed on the target data, so that the private data in the target data is disrupted. Even if the above target data is leaked, the private data in the target data cannot be identified, preventing the private data in the target data from being known to others and protecting the privacy of users. Among them, differential privacy can include multiple implementation methods. The following provides an optional implementation method, which specifically can include the following: Through the differential privacy algorithm of the exponential mechanism, based on a variety of different preset privacy budgets, differential privacy processing is respectively performed on the target data to obtain the differential privacy results corresponding to each privacy budget. Specifically, for the differential privacy algorithm of the exponential mechanism and any one of the variety of different preset privacy budgets, let the output domain of the query function be R, and each output value r ∈ R in the output domain. The function q(D, r) → R becomes the availability function of the output value r, used to evaluate the quality of the output value r. If the input of the random algorithm M is set as the dataset D, the output is the object r ∈ R, the function q(D, r) → R is the availability function, and Δq is the sensitivity of the function q(D, r) → R. If the algorithm M selects and outputs r from R with a probability proportional to exp(∈q(D, r) / 2Δq), then the algorithm M provides ∈-differential privacy protection. Based on the above method, for the input of the algorithm M being the target data and the output value r ∈ R corresponding to the target data, the algorithm M selects and outputs r from R with a probability proportional to exp(∈q(target data, r) / 2Δq), thereby obtaining the differential privacy result corresponding to this privacy budget. Through the above method, the differential privacy results corresponding to other privacy budgets can be obtained, and then the differential privacy results corresponding to each privacy budget can be obtained.

[0036] It should be noted that the above processing process is only one implementable way of differential privacy. In practical applications, in addition to being processed in the above way, the target data can also be differentially privately processed in multiple ways, which can be specifically set according to the actual situation, and the embodiments of this specification do not limit this. In addition, the above is only the differential privacy processing of the target data. In practical applications, differential privacy processing can also be performed on the private data in the target data, that is, differential privacy processing can be respectively performed on the target data and the private data in the target data to obtain the processed data, which can be specifically set according to the actual situation, and the embodiments of this specification do not limit this.

[0037] In step S106, based on the differential privacy results corresponding to a variety of different privacy budgets, the target model is trained through a preset contrastive learning loss algorithm to obtain the trained target model.

[0038] Among them, the contrast loss algorithm can be a common loss algorithm used in contrast learning, such as the InfoNCE algorithm or the MoCo algorithm, etc., which can be specifically set according to the actual situation, and the embodiments of this specification do not limit this. The target model can be any model, such as a classification model, a neural network model, etc., which can be specifically set according to the actual situation, and the embodiments of this specification do not limit this.

[0039] In implementation, an initial architecture of the target model can be constructed based on a preset algorithm according to the actual situation. Differential privacy results corresponding to multiple different privacy budgets can be input into the initial architecture of the target model to obtain corresponding results. Then, a preset contrast learning loss algorithm can be used to calculate the loss value corresponding to this result, and the parameter values of the undetermined parameters in the target model can be adjusted based on the obtained loss value. After that, an adjusted target model can be obtained. More differential privacy results corresponding to different privacy budgets can be obtained based on the above method, and this differential privacy result can be input into the adjusted target model to obtain corresponding results. Then, a preset contrast learning loss algorithm can be used to calculate the loss value corresponding to this result, and the parameter values of the undetermined parameters in the adjusted target model can be adjusted based on the obtained loss value. Finally, a trained target model can be obtained. The trained model can be deployed to the corresponding application scenarios.

[0040] The embodiments of this specification provide a data processing method based on privacy protection. The target data to be processed is obtained. The target data includes continuous structured data and / or image data. Then, through a preset variety of different privacy budgets, differential privacy processing is respectively performed on the target data to obtain differential privacy results corresponding to each privacy budget. Based on the differential privacy results corresponding to a variety of different privacy budgets, and through a preset contrast learning loss algorithm, the target model is trained to obtain a trained target model. In this way, self-supervised contrast learning for privacy protection is achieved by constructing privacy-protected data, achieving the purpose of pre-training without disclosing user privacy. In addition, this technical solution is applicable to image data and other continuous structured data, with a wide application range. Moreover, this technical solution provides a controllable privacy protection mechanism, and this privacy protection mechanism can also be used to control the similarity degree of positive samples in contrast learning, etc.

[0041] Embodiment 2

[0042] Such as Figure 3As shown in the figure, an embodiment of this specification provides a data processing method based on privacy protection. The execution subject of this method can be a terminal device or a server. Among them, the terminal device can be a mobile terminal device such as a mobile phone or a tablet computer, or a device such as a personal computer. The server can be an independent server, or a server cluster composed of multiple servers, etc. The server can be a background server such as a financial service or an online shopping service, or a background server of a certain application program, etc. This method can be applied to relevant scenarios such as model training through contrastive learning. In this embodiment, the server is used as the execution subject for detailed description. For the case of the terminal device, reference can be made to the following relevant content and will not be elaborated here. The method specifically may include the following steps:

[0043] In step S302, obtain the target data to be processed, where the target data includes continuous structured data and / or image data.

[0044] In step S304, through a variety of preset privacy budgets, use a preset differential privacy algorithm to perform differential privacy processing on the target data respectively, and obtain the differential privacy results corresponding to each privacy budget. The differential privacy algorithm includes one or more of the differential privacy algorithm based on the Laplace mechanism, the differential privacy algorithm based on the exponential mechanism, and the differential privacy algorithm based on the Gaussian mechanism.

[0045] Among them, the privacy budget is determined based on the dimension information of the vector corresponding to the target data and the goal of privacy protection for the target data. Generally, on the premise of satisfying differential privacy, the value of the privacy budget can not exceed 20. If the privacy protection intensity corresponding to the differential privacy is relatively low (such as the privacy protection protocol corresponding to the differential privacy is the Protection Against Reconstruction protocol, etc.), then the maximum value of the privacy budget can be sqrt(d), where d is the number of dimensions included in the target data.

[0046] In implementation, the quantity of privacy budgets can be preset according to actual situations. For example, if 2 types of privacy budgets are set, for the first type of privacy budget, a certain differential privacy algorithm (such as a differential privacy algorithm based on the Laplace mechanism, a differential privacy algorithm based on the exponential mechanism, or a differential privacy algorithm based on the Gaussian mechanism, etc.) can be used to perform differential privacy processing on the target data to obtain the differential privacy result corresponding to the first type of privacy budget. For the second type of privacy budget, a certain differential privacy algorithm (such as a differential privacy algorithm based on the Laplace mechanism, a differential privacy algorithm based on the exponential mechanism, or a differential privacy algorithm based on the Gaussian mechanism, etc.) can be used to perform differential privacy processing on the target data to obtain the differential privacy result corresponding to the second type of privacy budget. Through the above method, the same target data can be subjected to differential privacy processing with 2 different privacy budgets respectively, obtaining 2 corresponding differential privacy results.

[0047] If 3 types of privacy budgets are preset, the same target data can be subjected to differential privacy processing respectively through the above method to obtain the corresponding differential privacy results, that is, the differential privacy results corresponding to each of the 3 types of privacy budgets. Correspondingly, if more than 3 types of privacy budgets are preset, the differential privacy results corresponding to each type of privacy budget can be obtained through the above method.

[0048] Among them, for the specific processing processes such as using a differential privacy algorithm based on the Laplace mechanism to perform differential privacy processing on the target data, using a differential privacy algorithm based on the exponential mechanism to perform differential privacy processing on the target data, and using a differential privacy algorithm based on the Gaussian mechanism to perform differential privacy processing on the target data, etc., the processing can be carried out according to the algorithm steps determined by different differential privacy algorithms, and the embodiments of this specification will not elaborate here.

[0049] The specific processing method of the above step S304 can be various. The following provides an optional processing method, which can specifically include the processing from step A2 to step A6.

[0050] In step A2, obtain the quantity of the dimensions of the vector corresponding to the target data.

[0051] In implementation, a vectorization processing mechanism can be preset. There can be various types of vectorization processing mechanisms. For example, it can be a vectorization processing mechanism based on data features or a vectorization processing mechanism based on the correspondence between data segments and vectors (specifically, for a certain data, key data segments can be selected from the data, and vectors corresponding to the key data segments can be obtained from a pre-constructed correspondence. Then, the vectors corresponding to the selected key data segments can be fused to obtain the vector corresponding to the data, etc.). Specifically, it can be set according to the actual situation, and the embodiments of this specification do not limit this. The target data can be vectorized through this vectorization processing mechanism to obtain the vector corresponding to the target data. The obtained vector can include multiple dimensions. At this time, the number of dimensions included in the vector corresponding to the target data can be obtained, such as 256 dimensions or 128 dimensions, etc., which can be determined according to the actual situation.

[0052] In step A4, based on a variety of preset different privacy budgets and the number of the above dimensions, determine the privacy budget corresponding to each dimension.

[0053] In implementation, for any one privacy budget, since the target data includes data in multiple dimensions, therefore, this privacy budget can be allocated to each dimension of the target data. Based on this, this privacy budget can be divided into multiple parts according to the number of the above dimensions. For example, if the number of the above dimensions is 128, then this privacy budget can be divided into 128 sub-privacy budgets. The process of dividing a certain privacy budget into multiple sub-privacy budgets can be achieved through various different methods. For example, this privacy budget can be randomly divided into multiple sub-privacy budgets, or this privacy budget can be evenly distributed to obtain multiple identical sub-privacy budgets, etc. Specifically, it can be set according to the actual situation, and the embodiments of this specification do not limit this. Through the above method, the above processing can be performed on other privacy budgets. Finally, for each privacy budget among a variety of different privacy budgets, the privacy budget corresponding to each dimension can be obtained.

[0054] In step A6, use the preset differential privacy algorithm and the privacy budget corresponding to each dimension to perform differential privacy processing on the target data respectively to obtain the differential privacy result corresponding to each privacy budget.

[0055] In implementation, for any one privacy budget, through the above method, the privacy budget corresponding to each dimension can be obtained. Then, the preset differential privacy algorithm can be used to perform differential privacy processing on the target data to obtain the differential privacy result corresponding to this privacy budget. Through the above method, the differential privacy results corresponding to each privacy budget among a variety of different privacy budgets can be obtained.

[0056] In step S306, based on differential privacy results corresponding to multiple different privacy budgets, the target model is trained through a preset contrastive learning loss algorithm to obtain the trained target model.

[0057] In implementation, during the process of training the target model, when calculating the contrastive loss through a preset contrastive learning loss algorithm, it can be calculated in the following manner: There is a randomly generated data batch with a batch size of B. After the above differential privacy processing and / or data augmentation processing, 2B privacy-protected data can be generated. Then, contrastive learning loss algorithms such as InfoNCE or MoCo can be used to calculate the corresponding loss value (i.e., the contrastive loss value) for the data pairs in the above data batch. The corresponding parameters in the target model can be adjusted based on this loss value, and the above process is repeated until the finally trained target model is obtained.

[0058] An embodiment of this specification provides a data processing method based on privacy protection. The target data to be processed is obtained, and the target data includes continuous structured data and / or image data. Then, through a preset variety of different privacy budgets, differential privacy processing is respectively performed on the target data to obtain differential privacy results corresponding to each privacy budget. Based on the differential privacy results corresponding to multiple different privacy budgets, the target model is trained through a preset contrastive learning loss algorithm to obtain the trained target model. In this way, self-supervised contrastive learning for privacy protection is achieved by constructing privacy-protected data, achieving the purpose of pre-training without disclosing user privacy. In addition, this technical solution is applicable to image data and other continuous structured data, with a wide application range. Moreover, this technical solution provides a controllable privacy protection mechanism, and this privacy protection mechanism can also be used to control the similarity degree of positive samples in contrastive learning, etc.

[0059] Embodiment III

[0060] As Figure 4 shown, an embodiment of this specification provides a data processing method based on privacy protection. The execution subject of this method can be a terminal device or a server. Among them, the terminal device can be a mobile terminal device such as a mobile phone or a tablet computer, or a device such as a personal computer. The server can be an independent server or a server cluster composed of multiple servers. The server can be a background server such as a financial service or an online shopping service, or a background server of a certain application program. This method can be applied to relevant scenarios where model training is performed through contrastive learning. In this embodiment, the server is used as the execution subject for detailed description. For the case of the terminal device, the relevant content below can be referred to and will not be elaborated here. This method can specifically include the following steps:

[0061] In step S402, target data to be processed is obtained, and the target data includes image data.

[0062] In step S404, according to the number of preset privacy budgets, data augmentation processing is respectively performed on the target data to obtain multiple augmented data.

[0063] Among them, the privacy budget is determined based on the dimension information of the vector corresponding to the target data and the goal of privacy protection for the target data. Data augmentation processing can be to increase existing data to obtain more data, and data augmentation processing can be implemented in various ways. For example, it can be by flipping the image horizontally and / or vertically, by rotating the image, by scaling the image inwards or outwards, by cropping the image, by shifting the image, by adding Gaussian noise to the image, etc. Specifically, it can be set according to the actual situation, and the embodiments of this specification do not limit this.

[0064] In implementation, the number of preset privacy budgets can be 2, then data augmentation processing can be performed on the target data twice, that is, data augmentation processing is performed once for each privacy budget. Specifically, for the first type of privacy budget, data augmentation processing can be performed on the target data through any of the above methods (such as by flipping the image horizontally and / or vertically, by rotating the image, by scaling the image inwards or outwards, by cropping the image, by shifting the image, or by adding Gaussian noise to the image) to obtain the corresponding augmented data. Similarly, for the second type of privacy budget, data augmentation processing can be performed on the target data through any of the above methods (such as by flipping the image horizontally and / or vertically, by rotating the image, by scaling the image inwards or outwards, by cropping the image, by shifting the image, or by adding Gaussian noise to the image) to obtain the corresponding augmented data, so that 2 augmented data corresponding to 2 different privacy budgets can be obtained.

[0065] The specific processing method of the above step S404 can be various. The following provides an optional processing method, which can specifically include the processing of step B2 and step B4.

[0066] In step B2, according to the number of the above privacy budgets, the processing method required for data augmentation processing of the target data is determined.

[0067] Among them, the determined processing method is a preset processing method, or the determined processing method includes multiple different processing methods. The determined processing method can be one or more of the above multiple methods. In this embodiment, the determined processing method can include one or more of random rotation (i.e., by rotating the image) and image cropping (i.e., by cropping the image).

[0068] In step B4, according to the quantity of the above privacy budget, the target data is respectively subjected to data augmentation processing by the determined processing method to obtain multiple augmented data.

[0069] In implementation, if the quantity of the above privacy budget is 2, the determined processing method corresponding to the first privacy budget can be random rotation, and the determined processing method corresponding to the second privacy budget can be random rotation. Then, the target data can be respectively subjected to data augmentation processing by random rotation for the above two privacy budgets to obtain 2 augmented data. Or, the determined processing method corresponding to the first privacy budget can be random rotation, and the determined processing method corresponding to the second privacy budget can be image cropping. Then, the target data can be subjected to data augmentation processing by random rotation for the above first privacy budget to obtain the corresponding augmented data, and the target data can be subjected to data augmentation processing by image cropping for the above second privacy budget to obtain the corresponding augmented data.

[0070] In step S406, differential privacy processing is performed on the above augmented data through a preset variety of different privacy budgets to obtain a differential privacy result corresponding to each privacy budget.

[0071] The specific processing method of the above step S406 can be various. The following provides an optional processing method, which can specifically include the following: Through a preset variety of different privacy budgets, the differential privacy algorithm is used to respectively perform differential privacy processing on the above augmented data to obtain a differential privacy result corresponding to each privacy budget. The differential privacy algorithm includes one or more of a differential privacy algorithm based on the Laplace mechanism, a differential privacy algorithm based on the exponential mechanism, and a differential privacy algorithm based on the Gaussian mechanism.

[0072] The specific processing method of the above content can be various. The following provides an optional processing method, which can specifically include the processing from step C2 to step C6.

[0073] In step C2, obtain the number of dimensions of the vector corresponding to the above augmented data.

[0074] In step C4, based on a preset variety of different privacy budgets and the above number of dimensions, determine the privacy budget corresponding to each dimension.

[0075] In step C6, the above enhanced data is differentially privately processed respectively by using a preset differential privacy algorithm and the privacy budget corresponding to each dimension, and differential privacy results corresponding to each privacy budget are obtained.

[0076] For the specific processing procedures of the above steps, reference can be made to the relevant content above, which will not be elaborated here.

[0077] In step S408, based on the differential privacy results corresponding to multiple different privacy budgets, the target model is trained by using a preset contrastive learning loss algorithm, and the trained target model is obtained.

[0078] In implementation, during the process of training the target model, when calculating the contrastive loss by using a preset contrastive learning loss algorithm, the following method can be adopted for calculation: there is a randomly generated data batch with a batch size of B. After the above data augmentation processing, 2B privacy-protected data can be generated. Then, a contrastive learning loss algorithm such as InfoNCE or MoCo can be used to calculate the corresponding loss value (i.e., the contrastive loss value) for the data pairs in the above data batch. The corresponding parameters in the target model can be adjusted based on this loss value, and the above processing is repeated until the finally trained target model is obtained.

[0079] The embodiment of this specification provides a data processing method based on privacy protection. The target data to be processed is obtained, and the target data includes continuous structured data and / or image data. Then, the target data can be differentially privately processed respectively by using multiple preset different privacy budgets, and differential privacy results corresponding to each privacy budget are obtained. Based on the differential privacy results corresponding to multiple different privacy budgets, the target model is trained by using a preset contrastive learning loss algorithm, and the trained target model is obtained. In this way, self-supervised contrastive learning for privacy protection is realized by constructing privacy-protected data, achieving the purpose of pre-training without disclosing user privacy. In addition, this technical solution is applicable to image data and other continuous structured data, with a wide application range. Moreover, this technical solution provides a controllable privacy protection mechanism, and this privacy protection mechanism can also be used to control the similarity degree of positive samples in contrastive learning, etc.

[0080] Embodiment Four

[0081] As Figure 5A and Figure 5BAs shown in the figure, an embodiment of this specification provides a data processing method based on privacy protection. The execution subject of this method can be a blockchain system, which can be composed of terminal devices and / or servers, etc. Among them, the terminal device can be a mobile terminal device such as a mobile phone or a tablet computer, or a device such as a personal computer. The server can be an independent server or a server cluster composed of multiple servers. The server can be a background server such as a financial business or an online shopping business, or a background server of a certain application program. This method can be applied to relevant scenarios such as model training through contrastive learning. Specifically, this method can include the following steps:

[0082] In step S502, obtain the rule information of contrastive learning based on privacy protection, generate a corresponding first smart contract using the rule information of contrastive learning based on privacy protection, and deploy the first smart contract to the blockchain system.

[0083] Among them, a smart contract can be a computer protocol designed to spread, verify, or execute a contract in an information-based manner. A smart contract allows for trusted interactions without a third party, and the above interaction process is traceable and irreversible. The smart contract includes an agreement on which the contract parties can execute the rights and obligations agreed upon by the contract parties.

[0084] In implementation, in order to make the traceability of the contrastive learning process based on privacy protection better, a specified blockchain system can be created or joined. In this way, contrastive learning based on privacy protection can be executed based on the blockchain system. Specifically, a corresponding application program can be installed in the blockchain node, and an input box and / or a selection box, etc., for the rule information of contrastive learning based on privacy protection can be set in the application program. Then, corresponding information can be set in the above input box and / or selection box. Then, the blockchain system can receive the rule information of contrastive learning based on privacy protection. The blockchain system can generate a corresponding first smart contract through the rule information of contrastive learning based on privacy protection, and can deploy the first smart contract to the blockchain system. In this way, the rule information of contrastive learning based on privacy protection and the corresponding first smart contract are stored in the blockchain system, and other users cannot tamper with the rule information of contrastive learning based on privacy protection and the corresponding first smart contract. Moreover, the blockchain system executes contrastive learning based on privacy protection through the first smart contract.

[0085] In step S504, when the target data to be processed is obtained, call the first smart contract, and perform differential privacy processing on the target data respectively through a variety of different privacy budgets preset to obtain the differential privacy results corresponding to each privacy budget. The target data includes continuous structured data and / or image data.

[0086] Among them, the privacy budget is determined based on the dimensional information of the vector corresponding to the target data and the goal of privacy protection for the target data.

[0087] In implementation, the first smart contract can be set with relevant rule information for differentially privately processing the target data through a variety of preset different privacy budgets. In this way, the corresponding processing can be achieved based on the above rule information in the first smart contract. For specific details, please refer to the above relevant content and will not be elaborated here.

[0088] If the target data includes image data, before calling the first smart contract in step S504 above and differentially privately processing the target data through a variety of preset different privacy budgets to obtain the differential privacy results corresponding to each privacy budget, the following processing can also be included: Based on the second smart contract pre-deployed in the blockchain system, according to the quantity of the above privacy budget, the target data is respectively subjected to data augmentation processing to obtain multiple augmented data.

[0089] In implementation, the second smart contract can be set with relevant rule information for respectively performing data augmentation processing on the target data according to the quantity of the above privacy budget. In this way, the corresponding processing can be achieved based on the above rule information in the second smart contract. For specific details, please refer to the above relevant content and will not be elaborated here.

[0090] The processing methods for the above data augmentation processing of the target data can be diverse. The following provides an optional processing method, which can specifically include the following: Based on the second smart contract, according to the quantity of the above privacy budget, determine the processing method required for data augmentation processing of the target data; Based on the second smart contract, according to the quantity of the above privacy budget, respectively perform data augmentation processing on the target data through the determined processing method to obtain multiple augmented data.

[0091] Among them, the determined processing method is a preset processing method, or the determined processing method includes a variety of different processing methods, and the determined processing method includes one or more of random rotation and image cropping.

[0092] In implementation, the second smart contract can be set with relevant rule information for determining the processing method required for data augmentation processing of the target data according to the quantity of the above privacy budget, and for respectively performing data augmentation processing on the target data through the determined processing method according to the quantity of the above privacy budget. In this way, the corresponding processing can be achieved based on the above rule information in the second smart contract. For specific details, please refer to the above relevant content and will not be elaborated here.

[0093] Based on the above processing, the process of calling the first smart contract in step S504 to perform differential privacy processing on the target data through a variety of preset different privacy budgets to obtain the differential privacy results corresponding to each privacy budget may include: calling the first smart contract to perform differential privacy processing on the above enhanced data through a variety of preset different privacy budgets to obtain the differential privacy results corresponding to each privacy budget.

[0094] In implementation, relevant rule information for performing differential privacy processing on the above enhanced data through a variety of preset different privacy budgets may be set in the first smart contract. In this way, the corresponding processing can be implemented based on the above rule information in the first smart contract. For specific details, refer to the above relevant content and will not be elaborated here.

[0095] The process of calling the first smart contract in step S504 to perform differential privacy processing on the target data through a variety of preset different privacy budgets to obtain the differential privacy results corresponding to each privacy budget can be various. The following provides an optional processing method, which may specifically include the following: calling the first smart contract, and through a variety of preset different privacy budgets, using a preset differential privacy algorithm to perform differential privacy processing on the target data respectively to obtain the differential privacy results corresponding to each privacy budget. The differential privacy algorithm includes one or more of the differential privacy algorithm based on the Laplace mechanism, the differential privacy algorithm based on the exponential mechanism, and the differential privacy algorithm based on the Gaussian mechanism.

[0096] The processing method of the above content can be various. The following provides an optional processing method, which may specifically include the processing from step D2 to step D6.

[0097] In step D2, based on the first smart contract, obtain the number of dimensions of the vector corresponding to the target data.

[0098] In step D4, based on the first smart contract, through a variety of preset different privacy budgets and the number of the above dimensions, determine the privacy budget corresponding to each dimension.

[0099] In step D6, based on the first smart contract, use the preset differential privacy algorithm and the privacy budget corresponding to each dimension to perform differential privacy processing on the target data respectively to obtain the differential privacy results corresponding to each privacy budget.

[0100] In step S506, call the first smart contract, and based on the differential privacy results corresponding to a variety of different privacy budgets, train the target model through a preset contrastive learning loss algorithm to obtain the trained target model.

[0101] In implementation, the first smart contract can be set with differential privacy results corresponding to multiple different privacy budgets and relevant rule information for training the target model through a preset contrast learning loss algorithm. In this way, the corresponding processing can be implemented based on the above rule information in the first smart contract. For specific details, reference can be made to the above relevant content, which will not be elaborated here.

[0102] For the specific processing of the above steps S504 to S506, reference can be made to the relevant content in the above-mentioned first to third embodiments, that is, various processes involved in the above-mentioned first to third embodiments can be implemented through the corresponding smart contract.

[0103] An embodiment of this specification provides a data processing method based on privacy protection, which is applied to a blockchain system. It obtains rule information for contrast learning based on privacy protection, generates a corresponding first smart contract using the rule information for contrast learning based on privacy protection, and deploys the first smart contract to the blockchain system. When the target data to be processed is obtained, the first smart contract is called, and differential privacy processing is performed on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget. The target data includes continuous structured data and / or image data. Then, the first smart contract is called, and based on the differential privacy results corresponding to a variety of different privacy budgets, the target model is trained through a preset contrast learning loss algorithm to obtain the trained target model. In this way, self-supervised contrast learning with privacy protection is achieved by constructing privacy-protected data, achieving the purpose of pre-training without disclosing user privacy. In addition, this technical solution is applicable to image data and other continuous structured data, with a wide range of applications. Moreover, this technical solution provides a controllable privacy protection mechanism, which can also be used to control the similarity degree of positive samples in contrast learning, etc.

[0104] Embodiment Five

[0105] The above is the data processing method based on privacy protection provided by the embodiments of this specification. Based on the same idea, the embodiments of this specification also provide a data processing device based on privacy protection, as Figure 6 shown.

[0106] The data processing device based on privacy protection includes: a data acquisition module 601, a differential privacy module 602, and a contrast learning module 603, where:

[0107] The data acquisition module 601 acquires the target data to be processed, and the target data includes continuous structured data and / or image data;

[0108] The differential privacy module 602 performs differential privacy processing on the target data respectively through a variety of preset different privacy budgets, and obtains differential privacy results corresponding to each privacy budget;

[0109] The contrastive learning module 603 trains the target model based on the differential privacy results corresponding to a variety of different privacy budgets and trains the target model through a preset contrastive learning loss algorithm to obtain the trained target model.

[0110] In the embodiments of this specification, the differential privacy module 602 performs differential privacy processing on the target data respectively through a variety of preset different privacy budgets by using a preset differential privacy algorithm, and obtains differential privacy results corresponding to each privacy budget. The differential privacy algorithm includes one or more of a differential privacy algorithm based on the Laplace mechanism, a differential privacy algorithm based on the exponential mechanism, and a differential privacy algorithm based on the Gaussian mechanism.

[0111] In the embodiments of this specification, the differential privacy module 602 includes:

[0112] The quantity determination unit obtains the quantity of the dimensions of the vector corresponding to the target data;

[0113] The privacy budget unit determines the privacy budget corresponding to each dimension based on a variety of preset different privacy budgets and the quantity of the dimensions;

[0114] The differential privacy unit performs differential privacy processing on the target data respectively by using a preset differential privacy algorithm and the privacy budget corresponding to each dimension, and obtains differential privacy results corresponding to each privacy budget.

[0115] In the embodiments of this specification, the target data includes image data, and the device further includes:

[0116] The data augmentation module performs data augmentation processing on the target data respectively according to the quantity of the privacy budgets to obtain a plurality of augmented data;

[0117] The differential privacy module 602 performs differential privacy processing on the augmented data through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget.

[0118] In the embodiments of this specification, the data augmentation module includes:

[0119] The method determination unit determines the processing method required for performing data augmentation processing on the target data according to the quantity of the privacy budgets;

[0120] A data augmentation unit performs data augmentation processing on the target data respectively through a determined processing method according to the quantity of the privacy budget, and obtains a plurality of augmented data.

[0121] In the embodiments of the present specification, the determined processing method is a preset processing method, or the determined processing method includes multiple different processing methods, and the determined processing method includes one or more of random rotation and image cropping.

[0122] In the embodiments of the present specification, the privacy budget is determined based on the dimension information of the vector corresponding to the target data and the goal of privacy protection for the target data.

[0123] The embodiments of the present specification provide a data processing device based on privacy protection. The device obtains target data to be processed, and the target data includes continuous structured data and / or image data. Then, through a variety of preset different privacy budgets, differential privacy processing is respectively performed on the target data to obtain differential privacy results corresponding to each privacy budget. Based on the differential privacy results corresponding to a variety of different privacy budgets, the target model is trained through a preset contrast learning loss algorithm to obtain a trained target model. In this way, self-supervised contrast learning for privacy protection is realized by constructing privacy-protected data, achieving the purpose of pre-training without disclosing user privacy. In addition, this technical solution is applicable to image data and other continuous structured data, with a wide application range. Moreover, this technical solution provides a controllable privacy protection mechanism, and this privacy protection mechanism can also be used to control the similarity degree of positive samples in contrast learning, etc.

[0124] Embodiment Six

[0125] Based on the same idea, the embodiments of the present specification also provide a data processing device based on privacy protection, and this device is a device in a blockchain system, as Figure 7 shown.

[0126] The data processing device based on privacy protection includes: a contract deployment module 701, a differential privacy module 702, and a contrast learning module 703, where:

[0127] The contract deployment module 701 obtains rule information for contrast learning based on privacy protection, generates a corresponding first smart contract by using the rule information for contrast learning based on privacy protection, and deploys the first smart contract into the blockchain system;

[0128] The differential privacy module 702, when obtaining the target data to be processed, calls the first smart contract, and performs differential privacy processing on the target data respectively through a variety of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget. The target data includes continuous structured data and / or image data;

[0129] The contrastive learning module 703 calls the first smart contract, and based on the differential privacy results corresponding to a variety of different privacy budgets, trains the target model through a preset contrastive learning loss algorithm to obtain the trained target model.

[0130] In the implementation of this specification, the differential privacy module 702 calls the first smart contract, and performs differential privacy processing on the target data respectively through a variety of different preset privacy budgets by using a preset differential privacy algorithm to obtain differential privacy results corresponding to each privacy budget. The differential privacy algorithm includes one or more of a differential privacy algorithm based on the Laplace mechanism, a differential privacy algorithm based on the exponential mechanism, and a differential privacy algorithm based on the Gaussian mechanism.

[0131] In the implementation of this specification, the differential privacy module 702 includes:

[0132] The quantity determination unit obtains the quantity of the dimensions of the vector corresponding to the target data based on the first smart contract;

[0133] The privacy budget unit determines the privacy budget corresponding to each dimension based on the first smart contract, through a variety of different preset privacy budgets and the quantity of the dimensions;

[0134] The differential privacy unit performs differential privacy processing on the target data respectively based on the first smart contract, by using a preset differential privacy algorithm and the privacy budget corresponding to each dimension, to obtain differential privacy results corresponding to each privacy budget.

[0135] In the implementation of this specification, when the target data includes image data, the device further includes:

[0136] The data augmentation module, based on the second smart contract pre-deployed in the blockchain system, performs data augmentation processing on the target data respectively according to the quantity of the privacy budgets to obtain a plurality of augmented data;

[0137] The differential privacy module 702 calls the first smart contract, and performs differential privacy processing on the augmented data through a variety of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget.

[0138] In the implementation of this specification, the data augmentation module includes:

[0139] A method determination unit determines, based on the second smart contract and according to the quantity of the privacy budget, a processing method to be adopted for performing data enhancement processing on the target data.

[0140] A data enhancement unit performs data enhancement processing on the target data respectively by the determined processing method based on the second smart contract and according to the quantity of the privacy budget, to obtain a plurality of enhanced data.

[0141] In the implementation of this specification, the determined processing method is a preset processing method, or the determined processing method includes multiple different processing methods, and the determined processing method includes one or more of random rotation and image cropping.

[0142] In the implementation of this specification, the privacy budget is determined based on the dimension information of the vector corresponding to the target data and the goal of privacy protection for the target data.

[0143] An embodiment of this specification provides a data processing device based on privacy protection, which obtains rule information of contrastive learning based on privacy protection, generates a corresponding first smart contract by using the rule information of contrastive learning based on privacy protection, and deploys the first smart contract into the blockchain system. When the target data to be processed is obtained, the first smart contract is called, and differential privacy processing is respectively performed on the target data through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget. The target data includes continuous structured data and / or image data. Then, the first smart contract is called, and based on the differential privacy results corresponding to the plurality of different privacy budgets, the target model is trained through a preset contrastive learning loss algorithm to obtain a trained target model. In this way, self-supervised contrastive learning with privacy protection is realized by constructing privacy-protected data, achieving the purpose of pre-training without disclosing user privacy. In addition, this technical solution is applicable to image data and other continuous structured data, with a wide application range, and this technical solution provides a controllable privacy protection mechanism, which can also be used to control the similarity degree of positive samples in contrastive learning, etc.

[0144] Embodiment Seven

[0145] The above is the data processing device based on privacy protection provided by the embodiments of this specification. Based on the same idea, the embodiments of this specification also provide a data processing device based on privacy protection, as Figure 8 shown.

[0146] The data processing device based on privacy protection may be a terminal device, a server, or a device in the blockchain system provided in the above embodiments, etc.

[0147] Data processing devices based on privacy protection can vary significantly due to configuration or performance differences. They can include one or more processors 801 and a memory 802. One or more stored application programs or data can be stored in the memory 802. Among them, the memory 802 can be transient storage or persistent storage. The application programs stored in the memory 802 can include one or more modules (not shown in the figure), and each module can include a series of computer-executable instructions for the data processing device based on privacy protection. Further, the processor 801 can be set to communicate with the memory 802 and execute a series of computer-executable instructions in the memory 802 on the data processing device based on privacy protection. The data processing device based on privacy protection can also include one or more power supplies 803, one or more wired or wireless network interfaces 804, one or more input / output interfaces 805, and one or more keyboards 806.

[0148] Specifically, in this embodiment, the data processing device based on privacy protection includes a memory and one or more programs. One or more of the programs are stored in the memory, and one or more of the programs can include one or more modules. Each module can include a series of computer-executable instructions for the data processing device based on privacy protection and is configured to be executed by one or more processors. The one or more programs include the following computer-executable instructions for:

[0149] Obtain target data to be processed, where the target data includes continuous structured data and / or image data;

[0150] Perform differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget;

[0151] Based on the differential privacy results corresponding to a variety of different privacy budgets, and through a preset contrastive learning loss algorithm, train the target model to obtain the trained target model.

[0152] In the implementation of this specification, the performing differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget includes:

[0153] Perform differential privacy processing on the target data respectively through a preset variety of different privacy budgets by using a preset differential privacy algorithm to obtain differential privacy results corresponding to each privacy budget. The differential privacy algorithm includes one or more of a differential privacy algorithm based on the Laplace mechanism, a differential privacy algorithm based on the exponential mechanism, and a differential privacy algorithm based on the Gaussian mechanism.

[0154] In the implementation of this specification, when performing differential privacy processing on the target data respectively using a plurality of preset different privacy budgets and a preset differential privacy algorithm to obtain differential privacy results corresponding to each privacy budget, it includes:

[0155] Obtain the number of dimensions of the vector corresponding to the target data;

[0156] Based on a plurality of preset different privacy budgets and the number of dimensions, determine the privacy budget corresponding to each dimension;

[0157] Use the preset differential privacy algorithm and the privacy budget corresponding to each dimension to perform differential privacy processing on the target data respectively to obtain differential privacy results corresponding to each privacy budget.

[0158] In the implementation of this specification, the target data includes image data. Before performing differential privacy processing on the target data respectively using a plurality of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget, it further includes:

[0159] According to the number of privacy budgets, perform data augmentation processing on the target data respectively to obtain a plurality of augmented data;

[0160] When performing differential privacy processing on the target data respectively using a plurality of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget, it includes:

[0161] Perform differential privacy processing on the augmented data using a plurality of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget.

[0162] In the implementation of this specification, when performing data augmentation processing on the target data respectively according to the number of privacy budgets to obtain a plurality of augmented data, it includes:

[0163] According to the number of privacy budgets, determine the processing method to be used for performing data augmentation processing on the target data;

[0164] According to the number of privacy budgets, perform data augmentation processing on the target data respectively through the determined processing method to obtain a plurality of augmented data.

[0165] In the implementation of this specification, the determined processing method is a preset processing method, or the determined processing method includes a plurality of different processing methods, and the determined processing method includes one or more of random rotation and image cropping.

[0166] In the implementation of this specification, the privacy budget is determined based on the dimensionality information of the vector corresponding to the target data and the goal of privacy protection for the target data.

[0167] In addition, specifically in this embodiment, the data processing device based on privacy protection includes a memory and one or more programs. One or more of the programs are stored in the memory, and one or more of the programs may include one or more modules. Each module may include a series of computer-executable instructions in the data processing device based on privacy protection and is configured to be executed by one or more processors. The one or more programs include computer-executable instructions for performing the following:

[0168] Obtain the rule information of contrastive learning based on privacy protection, generate a corresponding first smart contract using the rule information of contrastive learning based on privacy protection, and deploy the first smart contract to the blockchain system;

[0169] When the target data to be processed is obtained, call the first smart contract, and perform differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget. The target data includes continuous structured data and / or image data;

[0170] Call the first smart contract, and based on the differential privacy results corresponding to a variety of different privacy budgets, train the target model through a preset contrastive learning loss algorithm to obtain the trained target model.

[0171] In the implementation of this specification, the target data includes image data. Before calling the first smart contract and performing differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget, it further includes:

[0172] Based on the second smart contract pre-deployed in the blockchain system, perform data augmentation processing on the target data respectively according to the number of privacy budgets to obtain multiple augmented data;

[0173] The step of calling the first smart contract and performing differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget includes:

[0174] Call the first smart contract and perform differential privacy processing on the augmented data through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget.

[0175] An embodiment of this specification provides a data processing device based on privacy protection, which obtains target data to be processed. The target data includes continuous structured data and / or image data. Then, through a variety of different preset privacy budgets, differential privacy processing can be performed on the target data respectively to obtain differential privacy results corresponding to each privacy budget. Based on the differential privacy results corresponding to a variety of different privacy budgets, and through a preset contrast learning loss algorithm, the target model is trained to obtain a trained target model. In this way, self-supervised contrast learning for privacy protection is achieved by constructing privacy-protected data, achieving the purpose of pre-training without disclosing user privacy. In addition, this technical solution is applicable to image data and other continuous structured data, with a wide range of applications. Moreover, this technical solution provides a controllable privacy protection mechanism, which can also be used to control the similarity degree of positive samples in contrast learning, etc.

[0176] Embodiment VIII

[0177] Further, based on the above Figures 1 to 5B shown method, one or more embodiments of this specification also provide a storage medium for storing computer-executable instruction information. In a specific embodiment, the storage medium can be a USB flash drive, an optical disc, a hard disk, etc. When the computer-executable instruction information stored in the storage medium is executed by a processor, the following processes can be realized:

[0178] Obtain target data to be processed, where the target data includes continuous structured data and / or image data;

[0179] Through a variety of different preset privacy budgets, perform differential privacy processing on the target data respectively to obtain differential privacy results corresponding to each privacy budget;

[0180] Based on the differential privacy results corresponding to a variety of different privacy budgets, and through a preset contrast learning loss algorithm, train the target model to obtain a trained target model.

[0181] In the implementation of this specification, the process of performing differential privacy processing on the target data respectively through a variety of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget includes:

[0182] Through a variety of different preset privacy budgets, use a preset differential privacy algorithm to perform differential privacy processing on the target data respectively to obtain differential privacy results corresponding to each privacy budget. The differential privacy algorithm includes one or more of a differential privacy algorithm based on the Laplace mechanism, a differential privacy algorithm based on the exponential mechanism, and a differential privacy algorithm based on the Gaussian mechanism.

[0183] In the implementation of this specification, the differential privacy processing of the target data is performed respectively using a preset differential privacy algorithm with a variety of different privacy budgets, and the differential privacy results corresponding to each privacy budget are obtained, including:

[0184] Obtain the number of dimensions of the vector corresponding to the target data;

[0185] Based on a variety of different privacy budgets preset and the number of dimensions, determine the privacy budget corresponding to each dimension;

[0186] Use the preset differential privacy algorithm and the privacy budget corresponding to each dimension to perform differential privacy processing on the target data respectively, and obtain the differential privacy results corresponding to each privacy budget.

[0187] In the implementation of this specification, the target data includes image data. Before performing differential privacy processing on the target data respectively using a variety of different privacy budgets preset to obtain the differential privacy results corresponding to each privacy budget, it further includes:

[0188] According to the number of privacy budgets, perform data augmentation processing on the target data respectively to obtain multiple augmented data;

[0189] The differential privacy processing of the target data is performed respectively using a variety of different privacy budgets preset to obtain the differential privacy results corresponding to each privacy budget, including:

[0190] Perform differential privacy processing on the augmented data using a variety of different privacy budgets preset to obtain the differential privacy results corresponding to each privacy budget.

[0191] In the implementation of this specification, the performing data augmentation processing on the target data respectively according to the number of privacy budgets to obtain multiple augmented data includes:

[0192] According to the number of privacy budgets, determine the processing method to be used for performing data augmentation processing on the target data;

[0193] According to the number of privacy budgets, perform data augmentation processing on the target data respectively through the determined processing method to obtain multiple augmented data.

[0194] In the implementation of this specification, the determined processing method is a preset processing method, or the determined processing method includes a variety of different processing methods, and the determined processing method includes one or more of random rotation and image cropping.

[0195] In the implementation of this specification, the privacy budget is determined based on the dimension information of the vector corresponding to the target data and the goal of privacy protection for the target data.

[0196] In addition, in another specific embodiment, the storage medium may be a USB flash drive, an optical disc, a hard disk, etc. When the computer-executable instruction information stored in the storage medium is executed by a processor, the following processes can be achieved:

[0197] Obtain the rule information of contrastive learning based on privacy protection, generate a corresponding first smart contract by using the rule information of contrastive learning based on privacy protection, and deploy the first smart contract to the blockchain system;

[0198] When the target data to be processed is obtained, call the first smart contract, and perform differential privacy processing on the target data respectively through a variety of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget. The target data includes continuous structured data and / or image data;

[0199] Call the first smart contract, and train the target model through a preset contrastive learning loss algorithm based on the differential privacy results corresponding to a variety of different privacy budgets to obtain the trained target model.

[0200] In the implementation of this specification, when the target data includes image data, before calling the first smart contract and performing differential privacy processing on the target data respectively through a variety of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget, it further includes:

[0201] Based on a second smart contract pre-deployed in the blockchain system, perform data augmentation processing on the target data respectively according to the number of the privacy budgets to obtain a plurality of augmented data;

[0202] The step of calling the first smart contract and performing differential privacy processing on the target data respectively through a variety of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget includes:

[0203] Call the first smart contract and perform differential privacy processing on the augmented data through a variety of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget.

[0204] An embodiment of this specification provides a storage medium. The target data to be processed is obtained, and the target data includes continuous structured data and / or image data. Then, differential privacy processing can be performed on the target data respectively through a variety of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget. Based on the differential privacy results corresponding to a variety of different privacy budgets, the target model is trained through a preset contrastive learning loss algorithm to obtain the trained target model. In this way, self-supervised contrastive learning for privacy protection is achieved by constructing privacy-protected data, achieving the purpose of pre-training without disclosing user privacy. In addition, this technical solution is applicable to image data and other continuous structured data, with a wide range of applications. Moreover, this technical solution provides a controllable privacy protection mechanism, and this privacy protection mechanism can also be used to control the similarity degree of positive samples in contrastive learning, etc.

[0205] The specific embodiments of this specification are described above. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims can be performed in a different order than in the embodiments and still achieve the desired result. Additionally, the processes depicted in the figures do not necessarily require the particular order or sequential order shown to achieve the desired result. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0206] In the 1990s, it was obvious to distinguish whether an improvement to a technology was a hardware improvement (e.g., improvement to circuit structures such as diodes, transistors, switches, etc.) or a software improvement (improvement to method flows). However, with the development of technology, many improvements to method flows today can be regarded as direct improvements to hardware circuit structures. Almost all designers obtain the corresponding hardware circuit structure by programming the improved method flow into the hardware circuit. Therefore, it cannot be said that an improvement to a method flow cannot be implemented with a hardware entity module. For example, a Programmable Logic Device (PLD) (e.g., a Field Programmable Gate Array (FPGA)) is such an integrated circuit whose logical function is determined by the user's programming of the device. The designer can program by himself to "integrate" a digital system on a PLD, without having to ask a chip manufacturer to design and fabricate a dedicated integrated circuit chip. Moreover, nowadays, instead of manually fabricating integrated circuit chips, this programming is mostly implemented using "logic compiler" software, which is similar to the software compiler used in program development and writing. The original code before compilation also has to be written in a specific programming language, which is called a Hardware Description Language (HDL). And there is not only one kind of HDL, but many kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), Confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), Lava, Lola, MyHDL, PALASM, RHDL (Ruby Hardware Description Language), etc. The most commonly used ones currently are VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog. Those skilled in the art should also be aware that as long as the method flow is slightly logically programmed with the above-mentioned several hardware description languages and programmed into the integrated circuit, it is easy to obtain the hardware circuit that implements the logical method flow.

[0207] The controller can be implemented in any suitable manner. For example, the controller can take the form of, for example, a microprocessor or a processor and a computer-readable medium storing computer-readable program code (such as software or firmware) executable by the (micro)processor, logic gates, switches, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller. Examples of the controller include, but are not limited to, the following microcontrollers: ARC625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320. The memory controller can also be implemented as part of the control logic of the memory. Those skilled in the art also know that in addition to implementing the controller in the form of pure computer-readable program code, it is entirely possible to logically program the method steps to enable the controller to be implemented in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. to achieve the same function. Therefore, such a controller can be considered a hardware component, and the devices included therein for implementing various functions can also be regarded as the structures within the hardware component. Or even, the devices for implementing various functions can be regarded as either software modules for implementing the method or structures within the hardware component.

[0208] The systems, devices, modules, or units illustrated in the above embodiments can be specifically implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, the computer can be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or any combination of these devices.

[0209] For the convenience of description, when describing the above devices, they are described separately as various units according to their functions. Of course, when implementing one or more embodiments of this specification, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0210] Those skilled in the art should understand that the embodiments of this specification can be provided as a method, a system, or a computer program product. Therefore, one or more embodiments of this specification can take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware aspects. Moreover, one or more embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk memories, CD-ROMs, optical memories, etc.) containing computer-usable program code.

[0211] Embodiments of this specification are described with reference to the flowcharts and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It should be understood that each flow and / or block in the flowchart and / or block diagram, and the combination of flows and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to the processors of general-purpose computers, special-purpose computers, embedded processors, or other programmable parallel and serial devices for fraud cases to generate a machine, such that the instructions executed by the processors of the computer or other programmable parallel and serial devices for fraud cases generate means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0212] These computer program instructions can also be stored in a computer-readable memory capable of guiding a computer or other programmable parallel and serial devices for fraud cases to work in a specific manner, such that the instructions stored in the computer-readable memory generate a manufactured article including instruction means, and the instruction means implement the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0213] These computer program instructions can also be loaded onto a computer or other programmable parallel and serial devices for fraud cases, such that a series of operation steps are executed on the computer or other programmable devices to generate a computer-implemented process, and thus the instructions executed on the computer or other programmable devices provide steps for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0214] In a typical configuration, a computing device includes one or more processors (CPUs), an input / output interface, a network interface, and a memory.

[0215] The memory may include non-permanent memory in the form of computer-readable media, random access memory (RAM), and / or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). The memory is an example of computer-readable media.

[0216] Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information. Information can be computer readable instructions, data structures, program modules or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device. As defined herein, computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.

[0217] It should also be noted that the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, commodity or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, commodity or device. In the absence of more restrictions, the elements defined by the sentence "comprises a ..." do not exclude the existence of other identical elements in the process, method, commodity or device including the elements.

[0218] It should be understood by those skilled in the art that the embodiments of this specification may be provided as methods, systems or computer program products. Therefore, one or more embodiments of this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, one or more embodiments of this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program codes.

[0219] One or more embodiments of the present specification may be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types. One or more embodiments of the present specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network. In a distributed computing environment, program modules may be located in local and remote computer storage media, including storage devices.

[0220] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and for the relevant parts, reference can be made to the partial description of the method embodiment.

[0221] The above description is only for the embodiments of this specification and is not intended to limit this specification. For those skilled in the art, various changes and modifications can be made to this specification. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of this specification shall be included within the scope of the claims of this specification.

Claims

1. A data processing method based on privacy protection, the method comprising: Obtaining target data to be processed, the target data including continuous structured data and / or image data; Performing differential privacy processing on the target data respectively through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget; Training a target model based on the differential privacy results corresponding to a plurality of different privacy budgets and through a preset contrastive learning loss algorithm to obtain a trained target model; When the target data includes image data, before performing differential privacy processing on the target data respectively through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget, the method further comprises: Performing data augmentation processing on the target data respectively according to the number of the privacy budgets to obtain a plurality of augmented data; The performing differential privacy processing on the target data respectively through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget includes: Performing differential privacy processing on the augmented data through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget.

2. The method according to claim 1, wherein the performing differential privacy processing on the target data respectively through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget includes: Performing differential privacy processing on the target data respectively through a plurality of different preset privacy budgets by using a preset differential privacy algorithm to obtain differential privacy results corresponding to each privacy budget, the differential privacy algorithm including one or more of a differential privacy algorithm based on the Laplace mechanism, a differential privacy algorithm based on the exponential mechanism, and a differential privacy algorithm based on the Gaussian mechanism.

3. The method according to claim 2, wherein the performing differential privacy processing on the target data respectively through a plurality of different preset privacy budgets by using a preset differential privacy algorithm to obtain differential privacy results corresponding to each privacy budget includes: Obtaining the number of dimensions of the vector corresponding to the target data; Determining the privacy budget corresponding to each dimension based on a plurality of different preset privacy budgets and the number of dimensions; Performing differential privacy processing on the target data respectively by using the preset differential privacy algorithm and the privacy budget corresponding to each dimension to obtain differential privacy results corresponding to each privacy budget.

4. The method according to claim 1, wherein the performing data augmentation processing on the target data respectively according to the number of the privacy budgets to obtain a plurality of augmented data includes: Determining the processing method to be used for performing data augmentation processing on the target data according to the number of the privacy budgets; Performing data augmentation processing on the target data respectively through the determined processing method according to the number of the privacy budgets to obtain a plurality of augmented data.

5. The method according to claim 4, wherein the determined processing manner is a preset processing manner, or the determined processing manner includes multiple different processing manners, and the determined processing manner includes one or more of random rotation and image cropping.

6. The method according to claim 1, wherein the privacy budget is determined based on the dimension information of the vector corresponding to the target data and the goal of privacy protection for the target data.

7. A data processing method based on privacy protection, which is applied to a blockchain system. The method includes: Obtaining rule information of contrastive learning based on privacy protection, generating a corresponding first smart contract by using the rule information of contrastive learning based on privacy protection, and deploying the first smart contract to the blockchain system; When obtaining target data to be processed, calling the first smart contract, and respectively performing differential privacy processing on the target data through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget, where the target data includes continuous structured data and / or image data; Calling the first smart contract, and training a target model through a preset contrastive learning loss algorithm based on the differential privacy results corresponding to a plurality of different privacy budgets to obtain a trained target model; When the target data includes image data, before calling the first smart contract and respectively performing differential privacy processing on the target data through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget, the method further includes: Based on a second smart contract pre-deployed in the blockchain system, respectively performing data augmentation processing on the target data according to the number of the privacy budgets to obtain a plurality of augmented data; The calling the first smart contract and respectively performing differential privacy processing on the target data through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget includes: Calling the first smart contract and performing differential privacy processing on the augmented data through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget.

8. A data processing device based on privacy protection, the device includes: A data acquisition module, which acquires target data to be processed, where the target data includes continuous structured data and / or image data; A differential privacy module, which respectively performs differential privacy processing on the target data through a plurality of different preset privacy budgets to obtain differential privacy results corresponding to each privacy budget; A contrastive learning module, which trains a target model based on the differential privacy results corresponding to a plurality of different privacy budgets and through a preset contrastive learning loss algorithm to obtain a trained target model; When the target data includes image data, the device further includes: A data augmentation module, which respectively performs data augmentation processing on the target data according to the number of the privacy budgets to obtain a plurality of augmented data; The differential privacy module performs differential privacy processing on the enhanced data through a variety of preset different privacy budgets, and obtains differential privacy results corresponding to each privacy budget.

9. A data processing device based on privacy protection, where the device is a device in a blockchain system, and the device includes: A contract deployment module, which obtains rule information for contrastive learning based on privacy protection, generates a corresponding first smart contract using the rule information for contrastive learning based on privacy protection, and deploys the first smart contract to the blockchain system; A differential privacy module, when obtaining target data to be processed, calls the first smart contract, and performs differential privacy processing on the target data respectively through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget, where the target data includes continuous structured data and / or image data; A contrastive learning module, which calls the first smart contract, and trains a target model based on the differential privacy results corresponding to a variety of different privacy budgets through a preset contrastive learning loss algorithm to obtain a trained target model; When the target data includes image data, the device further includes: A data enhancement module, based on a second smart contract pre-deployed in the blockchain system, performs data enhancement processing on the target data respectively according to the number of privacy budgets to obtain multiple enhanced data; The differential privacy module, calls the first smart contract, and performs differential privacy processing on the enhanced data through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget.

10. A data processing device based on privacy protection, where the data processing device based on privacy protection includes: A processor; And A memory arranged to store computer-executable instructions, and when the executable instructions are executed, the processor: Obtains target data to be processed, where the target data includes continuous structured data and / or image data; Performs differential privacy processing on the target data respectively through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget; Based on the differential privacy results corresponding to a variety of different privacy budgets, and trains a target model through a preset contrastive learning loss algorithm to obtain a trained target model; When the target data includes image data, before performing differential privacy processing on the target data respectively through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget, it further includes: Performs data enhancement processing on the target data respectively according to the number of privacy budgets to obtain multiple enhanced data; The performing differential privacy processing on the target data respectively through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget includes: Performing differential privacy processing on the enhanced data through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget.

11. A data processing device based on privacy protection, the device being a device in a blockchain system, the data processing device based on privacy protection comprising: A processor; And A memory arranged to store computer-executable instructions, the executable instructions, when executed, causing the processor to: Obtain rule information for privacy-preserving contrast learning, generate a corresponding first smart contract using the rule information for privacy-preserving contrast learning, and deploy the first smart contract to the blockchain system; When obtaining target data to be processed, call the first smart contract, and perform differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget, the target data including continuous structured data and / or image data; Call the first smart contract, and based on the differential privacy results corresponding to a variety of different privacy budgets, train a target model through a preset contrast learning loss algorithm to obtain a trained target model; The target data includes image data. Before calling the first smart contract and performing differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget, it further includes: Based on a second smart contract pre-deployed in the blockchain system, perform data augmentation processing on the target data respectively according to the number of the privacy budgets to obtain a plurality of augmented data; The calling the first smart contract and performing differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget includes: Call the first smart contract and perform differential privacy processing on the augmented data through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget.

12. A storage medium for storing computer-executable instructions, the executable instructions, when executed by a processor, implementing the following process: Obtain target data to be processed, the target data including continuous structured data and / or image data; Perform differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget; Based on the differential privacy results corresponding to a variety of different privacy budgets and through a preset contrast learning loss algorithm, train a target model to obtain a trained target model; The target data includes image data. Before performing differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget, it further includes: Perform data augmentation processing on the target data respectively according to the number of the privacy budgets to obtain a plurality of augmented data; The performing differential privacy processing on the target data respectively through a preset variety of different privacy budgets to obtain differential privacy results corresponding to each privacy budget includes: Perform differential privacy processing on the enhanced data through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget.

13. A storage medium for storing computer-executable instructions that, when executed by a processor, implement the following process: Obtain rule information for privacy-preserving contrast learning, generate a corresponding first smart contract using the rule information for privacy-preserving contrast learning, and deploy the first smart contract to a blockchain system; When target data to be processed is obtained, call the first smart contract and perform differential privacy processing on the target data through a variety of preset different privacy budgets respectively to obtain differential privacy results corresponding to each privacy budget, where the target data includes continuous structured data and / or image data; Call the first smart contract and train a target model through a preset contrast learning loss algorithm based on the differential privacy results corresponding to a variety of different privacy budgets to obtain a trained target model; When the target data includes image data, before calling the first smart contract and performing differential privacy processing on the target data through a variety of preset different privacy budgets respectively to obtain differential privacy results corresponding to each privacy budget, further include: Based on a second smart contract pre-deployed in the blockchain system, perform data enhancement processing on the target data respectively according to the number of the privacy budgets to obtain multiple enhanced data; The step of calling the first smart contract and performing differential privacy processing on the target data through a variety of preset different privacy budgets respectively to obtain differential privacy results corresponding to each privacy budget includes: Call the first smart contract and perform differential privacy processing on the enhanced data through a variety of preset different privacy budgets to obtain differential privacy results corresponding to each privacy budget.

Images