Data obfuscation method and device in collaborative maximum pooling calculation, medium and product

Abstract

The invention discloses a data obfuscation method and device in collaborative maximum pooling calculation, a medium and a product, and relates to the field of data obfuscation, the method comprises the following steps: generating a first participant, and generating a private matrix by a second participant; the two parties respectively carry out dimension reduction on the private matrix sum to obtain the private matrix sum; the two parties respectively send the private matrix and the private matrix to each other, the first participant carries out maximum pooling on the addition result of the private matrix sum and then multiplies the addition result with the private matrix A to obtain a first pooling result, and the first pooling result is sent to the calculation requester; and the second participant performs maximum pooling on the addition result of the private matrix sum, multiplies the addition result with the private matrix B to obtain a second pooling result, and sends the second pooling result to the calculation requester.

Description

Technical Field

[0001] This application relates to the field of data obfuscation, and in particular to a data obfuscation method, apparatus, medium, and product in collaborative maximum pooling computation. Background Technology

[0002] Privacy-preserving machine learning, as an important research direction in privacy computing, plays a crucial role in various applications of artificial intelligence. Pooling functions are common operations in neural networks, primarily used to reduce feature dimensionality and enhance model robustness, significantly impacting the training performance of models such as convolutional neural networks. Therefore, achieving secure computation of pooling functions, ensuring that data privacy is protected without affecting model performance, has become one of the most pressing problems to be solved in the field of privacy-preserving machine learning. Current privacy protection schemes for collaborative pooling and their shortcomings are as follows: For nonlinear operations such as secure two-sided pooling functions implemented using aliasing circuit technology, the entire calculation process needs to be converted into Boolean circuits and an aliasing table needs to be generated, resulting in significant consumption of computational and storage resources. At the same time, the framework uses fixed-point numbers to represent floating-point numbers, which will result in a certain loss of precision in numerical expression and affect the accuracy of the final calculation results.

[0003] For computation of secure two-party pooling functions based on secret sharing, private data is split into multiple shares and distributed to the participants, ensuring that data privacy cannot be reconstructed within a security threshold. However, in actual implementation, frequent secret splitting and share exchange lead to significant communication overhead and high communication latency, severely limiting the overall system performance.

[0004] For computations employing homomorphic encryption to implement secure two-party pooling functions in ciphertext, direct computation on encrypted data is permitted, protecting privacy. However, ciphertext computation is highly complex, consuming significant computational and storage resources and incurring substantial communication overhead. Furthermore, relying on third-party cloud platforms for computation carries potential privacy risks; if the platform is attacked, data security is difficult to guarantee.

[0005] Meanwhile, the communication efficiency issues of the above three schemes become more pronounced when extended to multi-party scenarios. When faced with large-scale data, the obfuscated circuit scheme rapidly expands in circuit size, increasing the computational and communication burden; the secret sharing scheme requires frequent information exchange during multi-party computation, resulting in low communication efficiency; and the homomorphic encryption scheme, due to its huge computational and communication overhead, struggles to maintain efficient operation in complex multi-party computation environments.

[0006] In summary, existing data obfuscation methods in collaborative maximum pooling computation suffer from high computational and storage resource consumption and are prone to causing significant communication burdens. Summary of the Invention

[0007] The purpose of this application is to provide a data obfuscation method, device, medium, and product in collaborative maximum pooling computation, which can solve the problems of high consumption of computing and storage resources and easy occurrence of large communication burden in existing data obfuscation methods in collaborative maximum pooling computation.

[0008] To achieve the above objectives, this application provides the following solution: Firstly, this application provides a data obfuscation method in collaborative maximum pooling computation, the data obfuscation method comprising: The first participant possesses its own private matrix. A Obtain the private matrix by increasing the dimensionality. The second participant possesses its own private matrix. B Obtain the private matrix by increasing the dimensionality. ; The first participant will have a private matrix. Multiplying by the first power yields the private matrix. The second participant will use a private matrix. Multiplying by the second power yields the private matrix. In this case, the first and second powers have the same base. The first participant generates a private matrix. The second participant generates a private matrix. Private matrix and Private matrix is ​​calculated using a secure two-way matrix inner product protocol. and The calculations were performed to obtain the result. The first and second participants respectively have access to the private matrix. and Dimensionality reduction is performed to obtain the private matrices respectively. and ; The first participant will have a private matrix. Send it to the second participant, who will then send the private matrix. Send to the first participant; The first participant will have a private matrix. and The sum of the results is then max-pooled and combined with the private matrix. A The first pooling result is obtained by multiplication and then sent to the computation requester. The second participant will have a private matrix. and The sum of the results is then max-pooled and combined with the private matrix. B The result of the second pooling is obtained by multiplication and then sent to the computation requester.

[0009] Secondly, this application provides a data obfuscation device in collaborative maximum pooling computation, the data obfuscation device including a first module applied to a first participant and a second module applied to a second participant; The first module is configured as: a private matrix held by the first participant. A Obtain the private matrix by increasing the dimensionality. , private matrix Multiplying by the first power yields the private matrix. Generate a private matrix For private matrices Dimensionality reduction yields the private matrix , private matrix Send to the second participant, the private matrix and The sum of the results is then max-pooled and combined with the private matrix. A The first pooling result is obtained by multiplication and then sent to the computation requester. The second module is configured as: a private matrix possessed by the second participant. B Obtain the private matrix by increasing the dimensionality. , private matrix Multiplying by the second power yields the private matrix. Generate a private matrix For private matrices Dimensionality reduction yields the private matrix , private matrix Send to the first participant, and send the private matrix and The sum of the results is then max-pooled and combined with the private matrix. B The result of the second pooling is obtained by multiplication and then sent to the computation requester. Where the first and second powers have the same base, private matrix and Private matrix is ​​calculated using a secure two-way matrix inner product protocol. and The calculations yielded the results.

[0010] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps performed by a first participant and / or a second participant in the data obfuscation of the collaborative maximum pooling computation described in any of the above.

[0011] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps performed by the first participant and / or the second participant in the data obfuscation of the cooperative maximum pooling computation described above.

[0012] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps performed by the first participant and / or the second participant in the data obfuscation of the collaborative maximum pooling computation described above.

[0013] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a data obfuscation method, device, medium, and product for collaborative max-pooling computation. It utilizes a secure data obfuscation method to implement a secure two-way matrix inner product protocol, S2PRIP, in a semi-honest scenario, and further implements a max-pooling computation protocol. Compared to existing homomorphic encryption, secret sharing, and obfuscated circuit schemes, this protocol reduces computational complexity to the O(nm) level. Simultaneously, the constant number of rounds of interaction and the use of real numbers in intermediate transmissions ensure that communication overhead costs are kept low, achieving a balance between the impossible triangle of security, lightweight design, and high efficiency. Furthermore, the secure two-way matrix inner product protocol S2PRIP enables parallel computation, resulting in high computational efficiency, fewer communication rounds, and no need for any key encryption operations. Moreover, because the sub-protocol is based on real-number operations, it is not limited to integer computation tasks and effectively avoids the precision problems caused by obfuscated circuits and homomorphic encryption techniques in floating-point processing. Furthermore, since the secure two-party matrix inner product protocol S2PRIP, which is used to implement the secure two-party maximum pooling computation protocol S2PMP, does not rely on third-party cloud platforms, it avoids the risk of privacy data leakage due to attacks on third-party cloud platforms. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0015] Figure 1 This is a schematic diagram of a secure two-party matrix row inner product protocol according to an embodiment of this application; Figure 2 This is a flowchart of a secure two-party matrix row inner product protocol according to an embodiment of this application; Figure 3This is a schematic diagram of a secure two-party average pooling protocol in one embodiment of this application; Figure 4 This is a schematic diagram of a secure two-party maximum pooling protocol in one embodiment of this application; Figure 5 This is a schematic diagram of a data obfuscation method according to an embodiment of this application; Figure 6 This is a flowchart illustrating a data obfuscation method in collaborative maximum pooling computation according to an embodiment of this application; Figure 7 This is a flowchart illustrating a data obfuscation method in collaborative average pooling computation according to an embodiment of this application. Detailed Implementation

[0016] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0017] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0018] First, some concepts involved in this application will be explained.

[0019] Semi-honest model: The semi-honest model (Semi-Honest Adversaries Security) is a specific protocol that assumes all parties involved in the computation will participate in privacy computation honestly and strictly follow the protocol to execute each step of the process. However, there is a risk that some parties may attempt to infer the privacy of other parties through the intermediate or final results of the protocol execution process.

[0020] Secure Two-Party Row Inner Product Protocol (S2PRIP): As the name suggests, the Secure Two-Party Row Inner Product Protocol assumes two mutually distrustful parties, P1 and P2, each holding a secret input matrix x and y, who jointly execute a two-party row inner product protocol. Ultimately, each participant receives the corresponding output. And the output satisfies Throughout the computation process, each participating node only knows the input and output data related to its own computation flow and cannot obtain any intermediate computation results from other participants.

[0021] Specifically, the problems that this protocol needs to solve can be described as follows: It is known that there are two computational methods Alice , Bob They are independent and distrustful of each other. Alice Holding a dimension as n× m Private data matrix stored only on its own computing node A , Bob Holding a dimension as n×m Private data matrix B The two parties hope to achieve this by jointly implementing a secure matrix inner product protocol. Ultimately, each participating node in the computation receives its corresponding dimension. n× 1 Output matrix The results are then sent to the requesting party to aggregate and obtain the desired row inner product of the two matrices. During the calculation process, each participating node can only know its own input and output information and cannot obtain the intermediate settlement results or held data information of other participants. A formal description of the problem can be found in [link to problem description]. Figure 1 .

[0022] Reference Figure 2 The specific process is as follows: Step 1: The auxiliary computing node, also known as the product / service node. Generate two pairs of random matrices, specifically in the form of a matrix with dimension 1. random matrix Dimensions random matrix and dimensions Two random matrices These random matrices must strictly satisfy the following constraints. Then by The auxiliary node will pair the random matrix Send the random matrix to Alice computing node (the first participant) Send to Bob's compute node (the second participant).

[0023] Step 2: Participant Alice receives the corresponding random matrix pair Then, its internal calculation And send it to the participating node Bob; where... This represents the private matrix owned by Alice, which is required to participate in the calculation of the row inner product of the two matrices in security.

[0024] Step 3: Participant Bob receives the corresponding random matrix pair Then, its internal calculation And send it to the participating node Alice; where... This represents the private matrix owned by participant Bob that needs to participate in the calculation of the row inner product of the two matrices in security.

[0025] Step 4: Participant Bob node receives the matrix sent by Alice node. Then, a random matrix is ​​secretly generated internally. And secretly calculate the matrix locally. Send it to the Alice node.

[0026] Step 5: Participant Alice node receives Then, the matrix is ​​obtained through local secret computation. ; Step 6: Participating nodes Alice and Bob each present their corresponding final obfuscation and splitting results. , Send the result to the two parties requesting matrix multiplication, who then sum up the results to obtain the final row product. .

[0027] It is easy to verify that: .

[0028] Secure Two-Party Average Pooling Protocol (S2PAP): As the name suggests, the Secure Two-Party Average Pooling Protocol assumes two mutually distrustful parties, P1 and P2. Each party holds a secret input matrix x and y, and a public parameter s representing the pooling window size. Together, they execute a two-party average pooling protocol. Ultimately, each participant receives the corresponding output. And the output satisfies Throughout the computation process, each participating node only knows the input and output data related to its own computation flow and cannot obtain any intermediate computation results from other participants.

[0029] Specifically, the problems that this protocol needs to solve can be described as follows: It is known that there are two computational methods Alice , Bob They are independent and distrustful of each other. Alice Holding a dimension as n× m Private data matrix stored only on its own computing node A , Bob Holding a dimension as n×m Private data matrix B Both parties have a public parameter.s Representing the pooling window size, they hope to achieve this by jointly implementing a safe average pooling protocol. Ultimately, each participating node in the computation obtains its corresponding output matrix. The pooling window size is then sent to the requesting party to obtain the expected two-way average pooling calculation result. For simplicity, it is assumed that the pooling window size can completely divide the entire original data matrix. During the calculation process, each participating node can only know its own input and output information and cannot obtain the intermediate settlement results and held data information of other participants. A formal description of the problem is provided below. Figure 3 .

[0030] Secure Two-Party Max Pooling Protocol: As the name suggests, Secure Two-Party Max Pooling (S2PMP) assumes two mutually distrustful parties, P1 and P2. Each party holds a secret input matrix x and y, and a public parameter s representing the pooling window size. Together, they execute a two-party max pooling protocol. Ultimately, each participant receives the corresponding output. And the output satisfies Throughout the computation process, each participating node only knows the input and output data related to its own computation flow and cannot obtain any intermediate computation results from other participants.

[0031] Specifically, the problems that this protocol needs to solve can be described as follows: It is known that there are two computational methods Alice , Bob They are independent and distrustful of each other. Alice Holding a dimension as n× m Private data matrix stored only on its own computing node A , Bob Holding a dimension as n×m Private data matrix B Both parties have a public parameter. s Representing the pooling window size, they hope to achieve this by jointly implementing a safe max-pooling protocol. Ultimately, each participating node in the computation obtains its corresponding output matrix. The pooling window size is then sent to the requesting party to obtain its desired two-way max pooling result. For simplicity, it is assumed that the pooling window size can completely divide the entire original data matrix. During the computation, each participating node can only know its own input and output information and cannot obtain the intermediate settlement results and held data information of other participants. A formal description of the problem is provided below. Figure 4 .

[0032] Secure Data Disguising Technology: Secure Data Disguising Technology is a data protection method used to protect the intermediate results of secure multi-party computations. By constructing a reasonable computation protocol, the computation results are randomly split, so that the outputs of multiple parties can be combined in a linear manner to form the real target computation result, ultimately achieving the effect of one-time key data privacy protection.

[0033] Privacy-Preserving Computing Technology: Privacy-preserving computing technology refers to a series of information security technologies that break down data silos, collaborate on multi-party computing, and ultimately achieve complex calculations and modeling analysis of multi-source data without exposing the privacy of the private data of all parties, thus ensuring that data elements are "usable but not visible" during the circulation and integration process.

[0034] Specifically, the problems that privacy-preserving computation needs to solve can be described as follows: For most multi-party computations, achieving secure computation typically involves multiple interactive steps. An unavoidable challenge arises: how to ensure the security of intermediate results? For example, multiplying two matrices... A×B When used as an intermediate result in computation, regardless of the participating nodes... Alice or node Bob The final matrix was obtained A×B The results could potentially lead to the deduction of data from the other party. Therefore, privacy-preserving computation must ensure not only the security of the initial input data but also the security of the intermediate results.

[0035] To achieve privacy-preserving computation, this application proposes a secure data obfuscation encryption technique. This technique involves decomposing any multi-ary operation into a new multi-ary addition to obfuscate the result of intermediate values. To better illustrate its principle, this paper uses a two-way operation as an example. The principle is as follows: Figure 5 As shown. Assume ,in It is the first k The objective function of the step is to compute the following. It belongs to the organization Alice No. k Step's private input data, It belongs to the organization Bob No. k Private input data for the first step. In the execution of the multi-party secure computation protocol... k intermediate results during the step The following constraints will be strictly followed: Alice Only know the results of your own calculations and Bob Only the result is known ,and . formula This illustrates the process of passing intermediate values ​​throughout the entire process. Alice and Bob Exchanging data information with each other is not allowed, including the splitting of intermediate calculation results. and Similarly, for the first k +1 step, its input and By the k Step Alice and Bob Output and Transmission constitutes, and and Its output and satisfy , Alice Only know the results of your own calculations and Bob Only the result is known Therefore, as long as the intermediate value is split into two random data items and stored separately by the two computation participants at each step of the computation, it can be guaranteed that no party can deduce the original data item from this obfuscated and encrypted data, thus making the entire privacy computation process highly secure.

[0036] This application combines privacy computation with collaborative max pooling computation and collaborative average pooling computation, respectively, and provides a data obfuscation method in collaborative max pooling computation and collaborative average pooling computation. Through this data obfuscation method, a secure two-party average pooling protocol and a secure two-party max pooling protocol can be implemented, and data obfuscation is achieved in the process of two-party average pooling and two-party max pooling, so that no participating party can obtain any intermediate calculation results about other participating parties.

[0037] The following sections will introduce the data obfuscation methods in collaborative maximum pooling and collaborative average pooling computations, respectively. It should be noted that the characters used in the conceptual introduction above are not interchangeable with those used in the data obfuscation method introduction below; that is, the character definitions in the two parts must be completely independent. The two parts may contain the same character with different definitions, or they may contain different characters with the same definition. The definition of each character in its respective part shall prevail.

[0038] The data obfuscation method in collaborative maximum pooling computation provided in this application can be applied to distributed computing systems. A distributed computing system can consist of multiple terminals or a terminal and a server. The terminals communicate with the server via a network. For example, the first participant and the second participant each correspond to two terminals, and the computation requester corresponds to the server.

[0039] The terminal can send the first pooling result and the second pooling result to the server, and the server adds the first pooling result and the second pooling result to obtain the final max pooling result of the collaborative max pooling calculation. In addition, in some embodiments, the calculation requester may also correspond to a third terminal.

[0040] The terminal can be, but is not limited to, various desktop computers, laptops, smartphones, tablets, etc. The server can be a standalone server or a server cluster composed of multiple servers, or it can be a cloud server.

[0041] In one exemplary embodiment, such as Figure 2 As shown, a data obfuscation method is provided in collaborative maximum pooling computation. This method is executed by various computer devices in cooperation, specifically by a terminal and a server. In this embodiment, the method is applied to... Figure 1 Taking terminal 101 (two terminals 101 corresponding to the first participant and the second participant respectively) and server 102 (corresponding to the computation requester) as examples, the following explanation will be provided. Figure 6 It includes the following steps 210 to 270.

[0042] Step 210, the first participant's private matrix A Obtain the private matrix by increasing the dimensionality. The second participant possesses its own private matrix. B Obtain the private matrix by increasing the dimensionality. .

[0043] The first participant possesses its own private matrix. A Obtain the private matrix by increasing the dimensionality. Specifically, this includes: the first participant will transfer the private matrix A Each element in is split into Each element is converted into a first vector by generating a set of distinct, random positive numbers and inserting a 1 after each number. The first participant then combines these first vectors to form a private matrix. .

[0044] The second participant possesses its own private matrix. B Obtain the private matrix by increasing the dimensionality. Specifically, this includes: the second participant transferring a private matrix.B Each element in is split into Each element is converted into a second vector by generating a set of distinct, random positive numbers and inserting a 1 before each number. The second participant then combines these second vectors to form a private matrix. .

[0045] For example, private matrix A The Middle i Line number j Column elements The corresponding first vector for: Private Matrix B The Middle i Line number j Column elements The corresponding second vector for: in, Indicates that the element The first obtained by splitting A positive number, Indicates that the element The first obtained by splitting A positive number.

[0046] Private Matrix Multiple first vectors according to the private matrix A The private matrix is ​​obtained by combining rows in order of precedence. Multiple second vectors according to the private matrix B The order of rows in the table is used to obtain the result.

[0047] It should be noted that the two participants can interchange the methods used to increase the dimension of the matrix.

[0048] In other embodiments, the first participant possesses its own private matrix. A Obtain the private matrix by increasing the dimensionality. Specifically, this includes: the first participant will transfer the private matrix A Each element in is split into Each element is converted into a first vector by generating a set of distinct, random positive numbers and inserting a 1 before each number. The first participant then combines these first vectors to form a private matrix. .

[0049] The second participant possesses its own private matrix. B Obtain the private matrix by increasing the dimensionality. Specifically, this includes: the second participant transferring a private matrix. BEach element in is split into Each element is converted into a second vector by generating a set of distinct, random positive numbers and inserting a 1 after each number. The second participant then combines these second vectors to form a private matrix. .

[0050] For example, private matrix A The Middle i Line number j Column elements The corresponding first vector for: Private Matrix B The Middle i Line number j Column elements The corresponding second vector for: in, Indicates that the element The first obtained by splitting A positive number, Indicates that the element The first obtained by splitting A positive number.

[0051] Step 220, the first participant will transfer the private matrix. Multiplying by the first power yields the private matrix. The second participant will use a private matrix. Multiplying by the second power yields the private matrix. The first and second powers have the same base.

[0052] The bases of the first and second powers are: A random prime number between; the exponent of the first power is the first participant's... A randomly selected positive number, the exponent of the second power is the second participant's... A positive number randomly selected from the list.

[0053] It should be noted that steps 210 and 220 above are preprocessing steps.

[0054] Step 230: The first participant generates a private matrix. The second participant generates a private matrix. Private matrix and Private matrix is ​​calculated using a secure two-way matrix inner product protocol. and The calculations yielded the results.

[0055] The calculation process of the secure two-party matrix inner product calculation protocol includes: the auxiliary node generates two random matrices. and and two random matrices and random matrix and With private matrix and Same type, random matrix and The number of rows and random matrix and They are identical and have a column count of 1; among them, the various random matrices satisfy the following: The first participant obtains a random matrix from the auxiliary node. and The second participant obtains a random matrix from the auxiliary node. and The first participant calculates the shared matrix. And share the matrix Send to the second participant; the second participant calculates the shared matrix. And share the matrix Send to the first participant; the second participant randomly generates a private matrix. And calculate the shared matrix. and sending shared matrix To the first participant; the first participant calculates the private matrix. .

[0056] Step 240, the first participant and the second participant respectively set up the private matrix. and Dimensionality reduction is performed to obtain the private matrices respectively. and .

[0057] Step 250, the first participant will transfer the private matrix Send it to the second participant, who will then send the private matrix. Send to the first participant.

[0058] Step 260, the first participant will transfer the private matrix and The sum of the results is then max-pooled and combined with the private matrix. A The result of the first pooling is obtained by multiplying and then sent to the computation requester.

[0059] Step 270, the second participant will transfer the private matrix. and The sum of the results is then max-pooled and combined with the private matrix. BThe result of the second pooling is obtained by multiplication and then sent to the computation requester.

[0060] Ultimately, the requester receives the computational result. , , . This is the result of the first pooling. This is the result of the second pooling. This is the final result of collaborative max pooling.

[0061] In deep neural networks, pooling operations (such as average pooling and max pooling) are key components of convolutional neural networks, used to extract local features while downsampling and enhance the model's translation invariance. In privacy-preserving computing environments, when multiple data holders collaborate on model training or inference, directly performing pooling operations can easily lead to others speculating on the original activation values, posing a risk of privacy breaches.

[0062] Taking medical image processing as an example, different hospitals each possess X-ray or MRI image data for a specific type of disease, and plan to jointly train or infer a convolutional neural network model for lesion detection. In this network, average pooling is used to extract the overall features of image regions, and max pooling is used to capture the most salient local activations. However, these pooling operations involve averaging or maximizing data across regions. In scenarios with multi-party encryption or distributed input, direct computation would expose sensitive intermediate feature information or image details, violating privacy protection requirements.

[0063] To address this, this embodiment proposes a data obfuscation method for collaborative max pooling and collaborative average pooling computations. Through this data obfuscation method, two data holders can collaboratively complete pooling computations without disclosing their respective local input data, effectively protecting the privacy of intermediate activation features.

[0064] This method is applicable to various tasks requiring convolutional neural network inference or training while protecting privacy, including but not limited to intelligent medical image recognition, financial graph analysis, industrial quality inspection image classification, and security surveillance image recognition. This method enables the obfuscation and pooling of intermediate features while maintaining the integrity of the model structure, thereby ensuring the privacy-preserving execution capability of the end-to-end neural network.

[0065] Steps 210 to 270 above constitute a data obfuscation method in collaborative maximum pooling computation. Private matrix A and BThese are intermediate feature matrices obtained by deep learning network models through feature extraction of images. For example, the images can be medical images, financial graphs, industrial quality inspection images, or security monitoring images. Correspondingly, in scenarios such as intelligent medical image recognition, financial graph analysis, industrial quality inspection image classification, and security monitoring image recognition, collaborative training or collaborative inference may be involved. That is, the first and second participants each provide their own images, which serve as input to the deep learning network model. The deep learning network model first extracts features from the images to obtain intermediate feature matrices, and then performs max pooling on the intermediate feature matrices. To prevent one participant from deducing the other participant's original image from the final max pooling result, this embodiment provides a data obfuscation method in collaborative max pooling computation. This data obfuscation method first obfuscates the private matrix of the first participant. A Private matrix of the second participant B Obfuscation is performed, and the obfuscated data is used to generate the final max pooling result, so that no one of the participants can use the final max pooling result to deduce the original image of the other participant.

[0066] In summary, this embodiment provides a data obfuscation method in collaborative max pooling computation, which can be used to implement the S2PMP max pooling computation protocol.

[0067] At the same time, refer to Figure 7 This embodiment also provides a data obfuscation method in collaborative average pooling computation, which includes: the first participant and the second participant respectively averaging their respective private matrices locally and then sending them to the computation requester.

[0068] Specifically, the first participant owns a private matrix. And calculate the private matrix. The second participant possesses a private matrix. And calculate the private matrix. Both parties output their calculation results. The two parties will then and Send to the computation requester, the computation requester receives .

[0069] In summary, this embodiment provides a data obfuscation method in collaborative max pooling and collaborative average pooling calculations, which has the following technical effects: 1. This embodiment utilizes a secure data obfuscation method to implement a secure two-party matrix row product protocol S2PRIP in a semi-honest scenario. It further implements a two-party average pooling and max pooling calculation protocol. Compared with existing homomorphic encryption, secret sharing, and obfuscated circuit schemes, this protocol reduces the computational complexity to the O(nm) level. At the same time, the constant number of rounds of interaction and the intermediate transmitted data are all real numbers, ensuring that the communication overhead cost is controlled within a low range. It balances the requirements of the three impossible triangles of security, lightweight, and efficiency.

[0070] 2. This embodiment utilizes a secure data obfuscation method to implement a secure two-party average pooling computation protocol S2PAP and a max-pooling computation protocol S2PMP. Simultaneously, it employs a secure two-party matrix inner product protocol S2PRIP to achieve parallel computation, resulting in high computational efficiency, fewer communication rounds, and no need for any key encryption operations. Furthermore, because the sub-protocols are based on real-number operations, they are not limited to integer computation tasks, effectively avoiding the precision issues that arise in floating-point number processing caused by obfuscated circuits and homomorphic encryption techniques.

[0071] 3. The secure two-party average pooling computation protocol S2PAP and the maximum pooling computation protocol S2PMP implemented in this embodiment have more security guarantees. Since the secure two-party matrix inner product protocol S2PRIP used to implement the secure two-party maximum pooling computation protocol S2PMP does not rely on a third-party cloud platform, it avoids the risk of privacy data leakage due to attacks on third-party cloud platforms.

[0072] In one exemplary embodiment, a data obfuscation device for collaborative maximum pooling computation is also provided, the data obfuscation device including a first module applied to a first participant and a second module applied to a second participant.

[0073] The first module is configured as: a private matrix held by the first participant. A Obtain the private matrix by increasing the dimensionality. , private matrix Multiplying by the first power yields the private matrix. Generate a private matrix For private matrices Dimensionality reduction yields the private matrix , private matrix Send to the second participant, the private matrix and The sum of the results is then max-pooled and combined with the private matrix. A The result of the first pooling is obtained by multiplying and then sent to the computation requester.

[0074] The second module is configured as: a private matrix possessed by the second participant. B Obtain the private matrix by increasing the dimensionality. , private matrix Multiplying by the second power yields the private matrix. Generate a private matrix For private matrices Dimensionality reduction yields the private matrix , private matrix Send to the first participant, and send the private matrix and The sum of the results is then max-pooled and combined with the private matrix. B The result of the second pooling is obtained by multiplication and then sent to the computation requester.

[0075] Where the first and second powers have the same base, private matrix and Private matrix is ​​calculated using a secure two-way matrix inner product protocol. and The calculations yielded the results.

[0076] It should be noted that the data obfuscation device in the collaborative max pooling computation is used to implement the data obfuscation method in the collaborative max pooling computation in the above embodiments. Therefore, for a detailed description of the device, please refer to the data obfuscation method in the collaborative max pooling computation in the above embodiments.

[0077] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, the memory storing a computer program that, when executed by the processor, implements steps performed by a first participant and / or a second participant in data obfuscation during collaborative maximum pooling computation.

[0078] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps performed by the first and / or second participants in the data obfuscation process described above in the cooperative maximum pooling computation.

[0079] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps performed by the first and / or second participants in the data obfuscation process described above in the cooperative maximum pooling computation.

[0080] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0081] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0082] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0083] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0084] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. In summary, the content of this specification should not be construed as a limitation of this application.

Claims

1. A data obfuscation method in collaborative maximum pooling computation, characterized in that, The data obfuscation method includes: The first participant possesses its own private matrix. A Obtain the private matrix by increasing the dimensionality. The second participant possesses its own private matrix. B Obtain the private matrix by increasing the dimensionality. ; The first participant will have a private matrix. Multiplying by the first power yields the private matrix. The second participant will use a private matrix. Multiplying by the second power yields the private matrix. In this case, the first and second powers have the same base. The first participant generates a private matrix. The second participant generates a private matrix. Private matrix and Private matrix is ​​calculated using a secure two-way matrix inner product protocol. and The calculations were performed to obtain the result. The first and second participants respectively have access to the private matrix. and Dimensionality reduction is performed to obtain the private matrices respectively. and ; The first participant will have a private matrix. Send it to the second participant, who will then send the private matrix. Send to the first participant; The first participant will have a private matrix. and The sum of the results is then max-pooled and combined with the private matrix. A The first pooling result is obtained by multiplication and then sent to the computation requester. The second participant will have a private matrix. and The sum of the results is then max-pooled and combined with the private matrix. B The result of the second pooling is obtained by multiplication and then sent to the computation requester.

2. The data obfuscation method in collaborative maximum pooling computation according to claim 1, characterized in that, The first participant possesses its own private matrix. A Obtain the private matrix by increasing the dimensionality. Specifically, it includes: The first participant will have a private matrix. A Each element in is split into Each element is converted into a first vector by generating a set of distinct, random positive numbers and inserting a 1 after each number. The first participant then combines these first vectors to form a private matrix. ; The second participant possesses its own private matrix. B Obtain the private matrix by increasing the dimensionality. Specifically, it includes: The second participant will have a private matrix. B Each element in is split into Each element is a distinct set of random positive numbers, and a 1 is inserted before each positive number, thus transforming each element into a second vector. The second participant combines multiple second vectors to form a private matrix. .

3. The data obfuscation method in collaborative maximum pooling computation according to claim 2, characterized in that, Private Matrix A The Middle i Line number j Column elements The corresponding first vector for: Private Matrix B The Middle i Line number j Column elements The corresponding second vector for: in, Indicates that the element The first obtained by splitting A positive number, Indicates that the element The first obtained by splitting A positive number; Private Matrix Multiple first vectors according to the private matrix A The private matrix is ​​obtained by combining rows in order of precedence. Multiple second vectors according to the private matrix B The order of rows in the table is used to obtain the result.

4. The data obfuscation method in collaborative maximum pooling computation according to claim 1, characterized in that, The bases of the first and second powers are A random prime number between; The exponent of the first power is the first participant's... A randomly selected positive number, the exponent of the second power is the second participant's... A positive number randomly selected from the list.

5. The data obfuscation method in collaborative maximum pooling computation according to claim 1, characterized in that, The calculation process of the secure two-party matrix inner product calculation protocol includes: The auxiliary node generates two random matrices. and and two random matrices and random matrix and With private matrix and Same type, random matrix and The number of rows and random matrix and They are identical and have a column count of 1; among them, the various random matrices satisfy the following: ; The first participant obtains a random matrix from the auxiliary node. and The second participant obtains a random matrix from the auxiliary node. and ; The first participant calculates the shared matrix. And share the matrix Send to the second participant; The second participant calculates the shared matrix. And share the matrix Send to the first participant; The second participant randomly generates a private matrix. And calculate the shared matrix. and sending shared matrix To the first participant; The first participant calculates the private matrix. .

6. The data obfuscation method in collaborative maximum pooling computation according to claim 1, characterized in that, Private Matrix A and B These are all intermediate feature matrices obtained by deep learning network models extracting features from images.

7. A data obfuscation device in collaborative maximum pooling computation, characterized in that, The data obfuscation device includes a first module applied to a first participant and a second module applied to a second participant; The first module is configured as: a private matrix held by the first participant. A Obtain the private matrix by increasing the dimensionality. , private matrix Multiplying by the first power yields the private matrix. Generate a private matrix For private matrices Dimensionality reduction yields the private matrix , private matrix Send to the second participant, the private matrix and The sum of the results is then max-pooled and combined with the private matrix. A The first pooling result is obtained by multiplication and then sent to the computation requester. The second module is configured as: a private matrix possessed by the second participant. B Obtain the private matrix by increasing the dimensionality. , private matrix Multiplying by the second power yields the private matrix. Generate a private matrix For private matrices Dimensionality reduction yields the private matrix , private matrix Send to the first participant, and send the private matrix and The sum of the results is then max-pooled and combined with the private matrix. B The result of the second pooling is obtained by multiplication and then sent to the computation requester. Where the first and second powers have the same base, private matrix and Private matrix is ​​calculated using a secure two-way matrix inner product protocol. and The calculations yielded the results.

8. A computer device, comprising: A memory, a processor, and a computer program stored on the memory and executable on the processor, characterized in that the processor executes the computer program to implement the steps performed by the first participant and / or the second participant in the data obfuscation of the collaborative maximum pooling computation according to any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the steps performed by the first participant and / or the second participant in the data obfuscation of the collaborative maximum pooling computation as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the steps performed by the first participant and / or the second participant in the data obfuscation of the collaborative maximum pooling computation as described in any one of claims 1-6.

Images