Privacy computing method, system and medium for secure two-way matrix hybrid multiplication

Through the decomposition technology of the basic protocol of secure two-party matrix multiplication 2PMP and rank conservation mechanism, efficient and secure matrix hybrid multiplication calculation in a semi-honest environment is achieved, and the problems of large overhead of ciphertext space calculation and communication and the risk of data leakage privacy are solved. , improves computing accuracy and privacy security.

CN117313121BActive Publication Date: 2024-11-29BEIHANG UNIV
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
CN202311224160.X
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2023-09-21
Publication Date
2024-11-29
Estimated Expiration
2043-09-21

AI Technical Summary

Technical Problem

The existing technology has greatly increased the complexity of ciphertext space and doubled communication overhead in the secure two-party matrix hybrid multiplication calculation due to the introduction of homomorphic encryption and inadvertent transmission technologies, and has failed to effectively solve the original data privacy risks caused by data leakage. Moreover, ciphertext calculation is limited by the problem of the numerical accuracy of floating-point calculations resulting in the fixed-length of bits.

Method used

The basic protocol of safe two-party matrix multiplication 2PMP is adopted to propose an end-to-end parallel hybrid multiplication calculation scheme suitable for semi-honest environments. Combined with the maximum float64-bit accuracy calculation, it can support safe two-party matrix addition and decomposition technology based on the rank conservation mechanism. Implement efficient and secure matrix hybrid multiplication calculations, avoiding the dependence of third-party cloud services.

Benefits of technology

It realizes efficient and secure matrix hybrid multiplication calculation, solves the problem of large overhead of ciphertext space calculation and communication, improves calculation accuracy, ensures the privacy and security of the original data, and reduces the privacy risks brought by data leakage.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN117313121B_ABST
    Figure CN117313121B_ABST
Patent Text Reader

Abstract

The present invention discloses a privacy computing method, system and medium for secure two-party matrix hybrid multiplication, which relates to the field of information security, including: a first computing participant and a second computing participant perform secure two-party matrix multiplication calculation based on a secure two-party matrix multiplication calculation protocol, based on private matrix A1 and private matrix B2, based on private matrix A2 and private matrix B1, and randomly split each matrix multiplication calculation result into two pieces of random data items; at the same time, the first computing participant performs private matrix multiplication based on private matrix A1 and private matrix A2, and the second computing participant performs private matrix multiplication based on private matrix B1 and private matrix B2 respectively; the first computing participant and the second computing participant obtain the final result matrix based on the corresponding two pieces of random data items and the private matrix multiplication result, respectively, and send it to the computing requester to calculate the privacy computing result. The present invention can greatly improve the privacy computing accuracy of secure two-party matrix hybrid multiplication.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the field of information security, and in particular to a privacy computing method, system and medium for secure bidirectional matrix hybrid multiplication. Background Art

[0002] With the continuous innovation and application of artificial intelligence and big data technologies, the world has officially entered the "data-driven" era. For countries and enterprises, data has become an important strategic resource. However, the big data era is inseparable from the open sharing of data. How to achieve the "availability but invisibility" of data, break the data silos and realize the interconnection and integrated analysis of data has become an urgent problem to be solved. The emergence of privacy computing technology, under the premise of effectively protecting the privacy of original data, not only realizes the safe circulation of data, but also effectively guarantees the separation of data ownership and usage rights. Privacy computing technology is widely used in application scenarios such as joint mining of big data and joint modeling of machine learning. The model calculation process involved in these related works widely contains compound operations of multi-party data mixing, superposition and multiplication. For example, in the common two-party secure linear regression problem, solving the regression coefficient β = (X T X) -1 X T The intermediate calculation process of Y (A 1 +B 1 )·(A 2 Y+B 2 Y) involves the problem of secure mixed multiplication of two matrices.

[0003] In the prior art, Benjamin and Atallah constructed a matrix hybrid computing outsourcing solution that can be verified by multiple servers based on homomorphic encryption technology. The user's computing complexity is O(n 2 t 2 ) where n and t are the order and threshold of the matrix, respectively. Based on different cryptographic difficulty assumptions (such as the co-CDH difficulty assumption and the deterministic linear difficulty assumption) and secret sharing techniques, Fiore et al. constructed a publicly verifiable amortized computation model for high-order polynomial and matrix multiplication.

[0004] The existing technologies have the following defects: (1) Although the homomorphic encryption computing scheme of Benjamin and Atallah effectively protects the privacy of input data and the verifiability of computing, the introduction of homomorphic encryption technology leads to a significant reduction in computing efficiency, and there is a risk that outsourced third-party cloud service providers are vulnerable to collusion attacks. (2) Although the secret sharing scheme of Fiore et al. improves the overall computing security, it involves a large number of public key cryptography tools, which causes a significant increase in computing overhead and a lack of computing accuracy due to the limited ring domain involved in the secret sharing scheme. Summary of the invention

[0005] The purpose of the present invention is to provide a privacy computing method, system and medium for secure two-party matrix hybrid multiplication, which can greatly improve the privacy computing accuracy of secure two-party matrix hybrid multiplication.

[0006] To achieve the above object, the present invention provides the following solutions:

[0007] The present invention provides a privacy computing method for secure two-party matrix hybrid multiplication, involving two computing participants, the first computing participant holds a private matrix A 1 and private matrix A 2 , the second computing participant holds a private matrix B 1 and private matrix B 2 , the method comprising:

[0008] The first computing participant and the second computing participant respectively input the private matrix A based on the secure two-party matrix multiplication computing protocol. 1 and the private matrix B 2 Perform a secure two-way matrix multiplication calculation to obtain a first matrix multiplication calculation result, and apply an obfuscation technique to the first matrix multiplication calculation result to randomly split it into matrices V a1 and the matrix V b1 , and sent to the first computing participant and the second computing participant respectively; the matrix V a1 and the matrix V b1 Satisfaction: V a1 +V b1 =A 1 ×B 2 ;

[0009] The first computing participant and the second computing participant respectively input the private matrix A based on the secure two-party matrix multiplication computing protocol. 2 and the private matrix B 1 Perform a secure two-way matrix multiplication calculation to obtain a second matrix multiplication calculation result, and apply an obfuscation technique to the second matrix multiplication calculation result to randomly split it into matrices V a2 and the matrix V b2 , and sent to the first computing participant and the second computing participant respectively; the matrix V a2 and the matrix V b2 Satisfaction: V a2 +V b2 =B 1 ×A 2 ;

[0010] The first computing participant performs a private matrix multiplication to obtain a matrix V a0 ; The matrix Va0 Satisfaction: V a0 =A 1 ×A 2 ;

[0011] The second computing participant performs a private matrix multiplication to obtain a matrix V b0 ; The matrix V b0 Satisfaction: V b0 =B 1 ×B 2 ; Step S4 is performed in parallel with step S1;

[0012] The first computing participant converts the matrix V a0 , the matrix V a1 and the matrix V a2 Perform local private calculations to obtain the matrix V a ;

[0013] The second computing participant converts the matrix V b0 , the matrix V b1 and the matrix V b2 Perform local private calculations to obtain the matrix V b ;

[0014] The first computing participant and the second computing participant respectively convert the matrix V a and the matrix V b The calculation requester is sent to the secure two-party mixed multiplication calculation, and the calculation requester calculates the privacy calculation result.

[0015] The present invention also provides a privacy computing system for secure two-party matrix mixed multiplication, involving two computing participants, the first computing participant holds a private matrix A 1 and private matrix A 2 , the second computing participant holds a private matrix B 1 and private matrix B 2 , the system comprising:

[0016] The first matrix multiplication calculation module is used for the first calculation participant and the second calculation participant to input the private matrix A respectively based on the secure two-party matrix multiplication calculation protocol. 1 and the private matrix B 2 Perform a secure two-way matrix multiplication calculation to obtain a first matrix multiplication calculation result, and apply an obfuscation technique to the first matrix multiplication calculation result to randomly split it into matrices V a1 and the matrix V b1 , and sent to the first computing participant and the second computing participant respectively; the matrix V a1 and the matrix V b1Satisfaction: V a1 +V b1 =A 1 ×B 2 ;

[0017] The second matrix multiplication calculation module is used for the first calculation participant and the second calculation participant to input the private matrix A respectively based on the secure two-party matrix multiplication calculation protocol. 2 and the private matrix B 1 Perform a secure two-way matrix multiplication calculation to obtain a second matrix multiplication calculation result, and apply an obfuscation technique to the second matrix multiplication calculation result to randomly split it into matrices V a2 and the matrix V b2 , and sent to the first computing participant and the second computing participant respectively; the matrix V a2 and the matrix V b2 Satisfaction: V a2 +V b2 =B 1 ×A 2 ;

[0018] The first local private computing module is used for the first computing participant to perform private matrix multiplication to obtain the matrix V a0 ; The matrix V a0 Satisfaction: V a0 =A 1 ×A 2 ;

[0019] The second local private computing module is used for the second computing participant to perform private matrix multiplication to obtain the matrix V b0 ; The matrix V b0 Satisfaction: V b0 =B 1 ×B 2 ;

[0020] The first matrix calculation module is used for the first calculation participant to convert the matrix V a0 , the matrix V a1 and the matrix V a2 Perform local private calculations to obtain the matrix V a ;

[0021] The second matrix calculation module is used for the second calculation participant to convert the matrix V b0 , the matrix V b1 and the matrix V b2 Perform local private calculations to obtain the matrix V b ;

[0022] The privacy computing result output module is used for the first computing participant and the second computing participant to respectively output the matrix V a and the matrix V b The calculation requester is sent to the secure two-party mixed multiplication calculation, and the calculation requester calculates the privacy calculation result.

[0023] The present invention also provides a computer-readable storage medium storing a computer program, characterized in that when the computer program is executed by a processor, it implements the privacy computing method for secure bidirectional matrix mixed multiplication.

[0024] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:

[0025] The present invention provides a privacy computing method, system and medium for secure two-party matrix hybrid multiplication. The present invention proposes an end-to-end parallel hybrid multiplication computing scheme suitable for a semi-honest environment based on the secure two-party matrix multiplication basic protocol 2PMP, which solves the problem of large ciphertext space computing and communication overhead caused by the introduction of homomorphic encryption and oblivious transmission technology in the prior art. Involving the key steps S1 and S3 for ensuring the stability of the precision of the hybrid operation, the calculation precision of the present invention lies in the use of the 2PMP secure two-party multiplication computing protocol that can support float64-bit precision computing at the highest, solving the problem of loss of numerical precision of floating-point number calculations due to the limitation of the fixed length of the digits in the ciphertext calculation in the prior art. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0027] Figure 1 A schematic diagram of a secure two-way matrix mixed multiplication calculation problem provided in the first embodiment of the present invention;

[0028] Figure 2 A schematic diagram of a secure two-way matrix multiplication calculation problem provided in the first embodiment of the present invention;

[0029] Figure 3 A flowchart of a secure two-party matrix multiplication calculation protocol provided in Example 1 of the present invention;

[0030] Figure 4 A flow chart of a secure two-party matrix hybrid multiplication calculation protocol for a semi-honest environment provided in the first embodiment of the present invention;

[0031] Figure 5A schematic diagram of a secure multi-party computing data obfuscation technology provided in Embodiment 1 of the present invention;

[0032] Figure 6 A flow chart of a privacy computing method for secure bi-matrix hybrid multiplication provided in the first embodiment of the present invention;

[0033] Figure 7 A specific flow chart of step S1 in a privacy computing method for secure bi-matrix hybrid multiplication provided in Embodiment 2 of the present invention;

[0034] Figure 8 A schematic diagram of a secure two-party matrix addition decomposition technology based on a rank conservation mechanism provided in Embodiment 2 of the present invention;

[0035] Fig. 9 A specific flow chart of step S2 in a privacy computing method for secure bi-matrix hybrid multiplication provided in Embodiment 2 of the present invention;

[0036] Fig.10 A flow chart of a secure two-party matrix hybrid multiplication calculation protocol for a privacy leakage environment corresponding to the specific example provided in the second embodiment of the present invention. DETAILED DESCRIPTION

[0037] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0038] In the prior art, Zhen and Jia et al. constructed a secure two-party matrix hybrid computing protocol that can be used to solve n-order matrix equations by citing the oblivious transfer protocol. However, Zhen et al.'s scheme based on oblivious transfer involves a large number of rounds of communication interactions, so for the mixed multiplication tasks of large-scale data matrices, there are problems of high communication cost and low computational efficiency. In addition, the existing technical solutions mentioned in the scheme are all based on the cryptographic technology stack, and can guarantee security in a semi-honest environment, but for the problem of secure two-party matrix calculation, they have not solved the problem of original data privacy risk that may be caused by the leakage of output results.

[0039] The present invention mainly focuses on the problem of secure two-way matrix mixed multiplication calculation, and realizes an efficient, secure, reliable and highly coupled basic privacy computing method and device. Therefore, based on the above objectives, the technical problems to be solved by the present invention are as follows:

[0040] 1) Existing solutions to the problem of secure two-way matrix mixed multiplication calculation mostly use cryptographic technology stacks such as homomorphic encryption, oblivious transfer, and secret sharing to serially execute each corresponding part after the mixed multiplication decomposition, which greatly increases the computational complexity of the ciphertext space and doubles the communication overhead.

[0041] 2) Existing protocols for secure two-party matrix mixed multiplication are mostly based on semi-honest environments, and do not consider the security issues of the results caused by the full rank of the input matrix in a data leakage environment.

[0042] 3) Existing solutions to the problem of secure two-party matrix mixed multiplication calculations mostly rely on outsourced cloud service computing systems. If the credibility of the third-party cloud service computing nodes is not high or they are attacked by malicious nodes, the secret key may be leaked, further causing the security risk of privacy leakage of the original data party.

[0043] 4) The existing solutions to the problem of secure two-way matrix mixed multiplication calculations use large prime number encryption, which increases the number of ciphertext bits in the calculation space. At the same time, the limitation of the fixed length of ciphertext bits directly causes the loss of numerical calculation precision, which affects the reliability of the calculation results.

[0044] Purpose of the present invention:

[0045] 1) The present invention uses the secure two-party matrix multiplication basic protocol 2PMP to implement hybrid multiplication calculation schemes in two different scenarios under end-to-end parallel computing modes, solving the problem of large ciphertext space calculation and communication overhead caused by the introduction of homomorphic encryption and oblivious transmission technology in the prior art, and realizing a parallel, secure, reliable and secure two-party matrix hybrid multiplication calculation scheme without the need to introduce third-party cloud services. (For question 13 above)

[0046] 2) The present invention proposes a rank conservation mechanism in matrix additive decomposition and a secure two-way matrix additive decomposition technology based on the rank conservation mechanism, which solves the problem that the existing technical solutions fail to essentially realize the autonomous and controllable decomposition process of the rank of the input matrix. (For question 1234)

[0047] 3) The present invention uses a secure two-way matrix multiplication basic protocol and a secure two-way matrix addition decomposition technology based on a rank conservation mechanism to implement a hybrid multiplication calculation scheme in an eight-link parallel computing mode, solving the original data privacy risk caused by data result leakage that the existing technical solutions have not been able to solve. (For question 1234)

[0048] 4) The present invention adopts a 2PMP secure two-way matrix multiplication calculation protocol that can support float64-bit precision calculation and has result reliability verification, which solves the problem of loss of numerical precision of floating-point calculation in the prior art due to the bottleneck of ciphertext calculation being limited by the fixed length of bits, and realizes a secure two-way matrix hybrid multiplication calculation protocol that can support high-precision floating-point matrix calculation. (For the above question 12)

[0049] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0050] Embodiment 1

[0051] This embodiment provides a privacy computing method for secure two-party matrix hybrid multiplication, involving two computing participants. The first computing participant holds a private matrix A 1 and private matrix A 2 , the second computing participant holds a private matrix B 1 and private matrix B 2 The method is applicable to a semi-honest computing environment; the semi-honest model (Semi-Honest AdversariesSecurity) is a specific protocol that assumes that all parties involved in the computing process will honestly participate in the privacy computing and strictly follow each step of the process. However, there is a risk that some corrupted parties may try to infer the privacy of other parties through the intermediate or final results of the protocol execution process.

[0052] In order to more clearly understand the solution of this embodiment, the following related definitions are first introduced below:

[0053] Privacy-Preserving Computation refers to a series of information technologies that analyze and calculate data without leaking the original data. This ensures that data is “available but invisible” during its circulation and integration.

[0054] Secure Two-Party Matrix Hybrid Multiplication Protocol (2PHMP) problem definition: Figure 1 As shown in the figure, it is known that there are two computing participants, Alice and Bob, who are independent and do not trust each other. Alice holds a set of private data matrices A that are only stored in her own computing node. 1 ∈R m×t and A 2 ∈R t×n , Bob holds a set of private data matrices B stored only in his own computing nodes1 ∈R m×t and B 2 ∈R t×n The two parties jointly execute a two-party matrix mixed multiplication protocol f((A 1 , A 2 ), (B 1 , B 2 ))=(A 1 +B 1 )×(A 2 +B 2 )=V a +V b , and finally each computing participant node obtains its corresponding output matrix V a 、V b ∈R m×n , and sent to the calculation requester to aggregate and obtain the expected result of the two-party matrix mixed multiplication calculation. During the calculation process, each participating node can only obtain the input and output information involved in its own calculation process, and cannot obtain the intermediate settlement results and private data information held by other participants.

[0055] The secure two-party matrix multiplication protocol (2PMP) problem definition: It is known that there are two computing parties, Alice and Bob, who are independent and do not trust each other. Alice holds a private data matrix A with a dimension of n×s and is only stored in her own computing node, and Bob holds a private data matrix B with a dimension of s×m. The two participants hope to jointly execute a secure matrix multiplication protocol to achieve f(A, B) = AB = V a +V b Finally, each computing participant node obtains its own output matrix V with dimensions of n×m a 、V b , and sent to the calculation requester to aggregate and obtain the expected two-way matrix product result. During the calculation process, each participant node can only understand its own input and output information, and cannot obtain the intermediate settlement results and holding data information of other participants. The formal description of the problem is shown in Figure 2 The secure two-party multiplication protocol 2PMP involved can be replaced by existing cryptographic tools, such as secret sharing, oblivious transmission, obfuscated circuits, fully homomorphic encryption and other technologies.

[0056] The protocol flow of the secure two-party matrix multiplication protocol (2PMP) is as follows: Figure 3 As shown:

[0057] Step 1: The auxiliary computing node, also known as the commodity service node CS (CommodityServer), generates two sets of random matrix pairs, which are in the form of random matrices R with dimensions n×s a , a random matrix R with dimension s×mb and two random matrices r of dimension n×m a 、r b These random matrices must strictly satisfy the following constraints: a +r b =R a ·R b Then the CS auxiliary node converts the random matrix pair (R a , r a ) is sent to the participant Alice’s computing node, and the random matrix pair (R b , r b ) is sent to the computing node of participant Bob.

[0058] Step 2: Upon receiving the corresponding random matrix pair (R a , r a ) after which the node internally calculates And send it to the participating node Bob.

[0059] Step 3: After receiving the corresponding random matrix pair (R b , r b ) after which the node internally calculates And send it to the participating node Alice.

[0060] Step 4: Participant Bob's node receives the matrix sent from Alice's node After that, a random matrix V is generated secretly inside b ∈R n×m , and secretly compute the matrix locally Sent to Alice node.

[0061] Step 5: After receiving T, the participant Alice's node calculates the matrix locally

[0062] Step 6: Participants Alice and Bob respectively generate their corresponding final obfuscated split results V a 、V b Sent to the two parties requesting matrix multiplication, which will summarize and obtain the final product AB = V a +V b .

[0063] It is easy to verify that:

[0064]

[0065] Secure Multi-party Computation Data Obfuscation Technology (SMDOT): For the vast majority of multi-party computations, the process of achieving secure computation usually involves more than one step. Inevitably, we need to face a problem: how to ensure the security of intermediate results. For example, when the product of two matrices A×B is used as the intermediate result of the calculation, whether the participating node Alice or the node Bod obtains the result of the final matrix A×B, it is possible to infer the data information of the other party. Therefore, in the privacy computing process, not only the security of the input original data must be guaranteed, but also the security of the intermediate value. To solve this problem, this study proposes a data obfuscation encryption technology, which decomposes any multi-objective operation into a new multi-objective addition to obfuscate the result of the calculation of the intermediate value. In order to make it easier to explain its principle, this study takes the most basic two-party operation type as an example. The principle is as follows Figure 4 shown.

[0066] Assume S k =F k (A i , B i ), where F k is the target calculation function, corresponding to Figure 5 SMDOT (Secure Multi-Party Data Obfuscation Techniques), A i is private data belonging to organization Alice, B i It is the private data of the organization Bob. When executing each sub-protocol of the multi-party secure computation protocol, the intermediate result S k will strictly follow the following constraints: Alice only knows her own calculation result A k and Bob only knows B k , and A k +B k =S k . Formula[A i :B i ]→[A k :B k |A k +B k =F k (A i , B i )] represents the process of transferring the intermediate value. During the whole process, Alice and Bod are not allowed to exchange each other’s data information, including the calculation of the intermediate result after splitting A k and B kAs long as the intermediate value is split into two random data items at each step of the calculation, it can be guaranteed that no party can reverse the original data item from the obfuscated encrypted data, making the entire privacy computing process more secure.

[0067] Specifically, Figure 6 As shown, the method includes:

[0068] Step S1: The first computing participant (i.e., computing participant Alice's node) and the second computing participant (i.e., computing participant Bob's node) respectively input the private matrix A based on the secure two-party matrix multiplication computing protocol 2PMP 1 ∈R m×t and the private matrix B 2 ∈R t×n Perform a secure two-party matrix multiplication calculation to obtain a first matrix multiplication calculation result. After the 2PMP protocol calculation is completed, the first matrix multiplication calculation result is randomly split into matrices V by applying obfuscation technology a1 ∈R m×n and the matrix V b1 ∈R m×n , and sent to the first computing participant and the second computing participant respectively; the matrix V a1 and the matrix V b1 Satisfaction: V a1 +V b1 =A 1 ×B 2 .

[0069] Step S2: The first computing participant and the second computing participant respectively input the private matrix A based on the secure two-party matrix multiplication protocol 2PMP 2 ∈R m×t and the private matrix B 1 ∈R t×n Perform a secure two-way matrix multiplication calculation to obtain a second matrix multiplication calculation result. After the 2PMP protocol calculation is completed, the second matrix multiplication calculation result is randomly split into matrices V by applying obfuscation technology a2 ∈R m×n and the matrix V b2 ∈R m×n , and sent to the first computing participant and the second computing participant respectively; the matrix V a2 and the matrix V b2 Satisfaction: V a2 +V b2 =B 1 ×A 2 ; Step S2 is executed in parallel with step S1.

[0070] Step S3: The first computing participant performs a private matrix multiplication locally to obtain a matrix V a0 ; and after the calculation is completed, the result is stored in the private storage space inside Alice's local node. The matrix V a0 Satisfaction: V a0 =A 1 ×A 2 The step S3 is performed in parallel with the step S1.

[0071] Step S4: The second computing participant performs a private matrix multiplication locally to obtain the matrix V b0 ; and after the calculation is completed, the result is stored in the private storage space inside Bob's local node; the matrix V b0 Satisfaction: V b0 =B 1 ×B 2 The step S4 is performed in parallel with the step S1.

[0072] Step S5: The first computing participant converts the matrix V a1 and the matrix V a2 Perform secret aggregation and perform private computation V locally a =V a0 +V a1 +V a2 , where V a ∈R m×n .

[0073] Step S6: The second computing participant converts the matrix V b1 and the matrix V b2 Perform secret aggregation and perform private computation V locally b =V b0 +V b1 +V b2 , where V b ∈R m×n .

[0074] Step S7: The first computing participant and the second computing participant respectively convert the matrix V a and the matrix V b Sent to the calculation requester of the secure two-party mixed multiplication calculation, which summarizes the final calculation result (A 1 +B 1 )×(A 2 +B 2 )=V a +V b .

[0075] Obviously, it is easy to verify that:

[0076] V a +V b =(V a0 +V a1 +V a2 )+(V b0 +V b1 +V b2 )=V a0 +V b0 +(V a1 +V b2 )+(V a2 +V b1 )=(A 1 ×A 2 )+(B 1 ×B 2 )+(A 1 ×B 2 )+(B 1 ×A 2 )=(A 1 +B 1 )×(A 2 +B 2 ).

[0077] In this embodiment, the present invention proposes an end-to-end parallel hybrid multiplication calculation scheme suitable for a semi-honest environment based on the secure two-party matrix multiplication basic protocol 2PMP, and adopts parallel calculation of steps S1 to S4 to solve the problems of large ciphertext space calculation and communication overhead caused by the introduction of homomorphic encryption and oblivious transmission technology in the prior art.

[0078] In this embodiment, the key steps S1 and S3 are involved to ensure the stability of the precision of the mixed operation. The calculation precision of the present invention lies in the use of the 2PMP secure two-party multiplication calculation protocol that can support float64-bit precision calculation at the highest, which solves the problem of loss of numerical precision in floating-point calculation due to the limitation of the fixed length of the digits in the ciphertext calculation in the prior art.

[0079] Embodiment 2

[0080] The difference from the first embodiment is that the private matrix A in step S1 of the first embodiment 2 and the private matrix B 1 The secure two-way matrix multiplication calculation performed and the private matrix A in step S2 1 and the private matrix B 2 A secure two-way matrix addition decomposition strategy based on rank conservation mechanism is introduced in the secure two-way matrix multiplication calculation performed.

[0081] A semi-honest computing environment is an ideal situation, which has high requirements for the network communication environment and the reliability of computing nodes. In real engineering application scenarios, even if the participating nodes do not disclose their computing information to each other, once there is data leakage in the intermediate process or the final output result shard matrix, it may bring privacy exposure security risks to the input data information of the participating computing protocol. Its privacy risk is mainly caused by the matrix data structure at the input end of the secure two-party matrix multiplication protocol 2PMP based on the multi-party computing framework. Specifically, for any input binary matrix multiplication operation F(A, B) = A×B = V a +V b =M, once the output sub-matrix V a or V b If there is a malicious node attack that leads to data exposure, the final multiplication result M=A×B will be directly exposed. Furthermore, based on the calculation result of the final two-party matrix multiplication and the input matrix of the calculation, the participating party can infer the input data of the other party by solving the matrix equation AX=M or XB=M, for example, when one of the inputs is a reversible matrix. In order to deal with the problem of secure two-party matrix mixed multiplication in the above-mentioned problem scenario, this embodiment provides a privacy computing method for secure two-party matrix mixed multiplication, involving two computing participants. The first computing participant holds a private matrix A. 1 and private matrix A 2 , the second computing participant holds a private matrix B 1 and private matrix B 2 The method is applicable to scenarios with privacy leakage risks. The privacy risk is mainly caused by the matrix data structure at the input end of the secure two-party matrix multiplication protocol 2PMP based on the multi-party computing framework. Specifically, for any input real number field, the binary matrix multiplication operation F(A, B) = A × B = V a +V b =M, once the output sub-matrix V a or V b If any of the nodes are attacked by a malicious party and the data is exposed, the final multiplication result M=A×B will be directly exposed. Furthermore, the participating parties can infer the input data of the other party by solving the matrix equation AX=M or XB=M based on the calculation result of the final two-party matrix multiplication and the input matrix they participate in the calculation, for example, when one of the inputs is a reversible matrix. In order to deal with the problem of secure two-party matrix mixed multiplication in the above problem scenario, this paper designs a secure two-party matrix mixed multiplication calculation protocol for privacy leakage risk scenarios.

[0082] Specifically, Figure 6 As shown, the method includes:

[0083] Step S1: The first computing participant and the second computing participant respectively input the private matrix A based on the secure two-party matrix multiplication computing protocol 1 and the private matrix B 2 Perform a secure two-way matrix multiplication calculation to obtain a first matrix multiplication calculation result, and apply an obfuscation technique to the first matrix multiplication calculation result to randomly split it into matrices V a1 and the matrix V b1 , and sent to the first computing participant and the second computing participant respectively; the matrix V a1 and the matrix V b1 Satisfaction: V a1 +V b1 =A 1 ×B 2 .

[0084] Among them, Figure 7 As shown, step S1 specifically includes:

[0085] Step S11: The first computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to convert the private matrix A 1 Decompose the private matrix B2 into multiple sub-matrices; the second computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix B2 into multiple sub-matrices; the private matrix The private matrix The rank of the private matrix A1 is equal to the rank of the corresponding submatrices A 1i The private matrix B 2 The rank is equal to the corresponding submatrix B 2j The sum of the ranks.

[0086] Secure Bipartite Matrix Additive Decomposition Technology (RS2MADT) based on Rank Conservation Mechanism:

[0087] The key to matrix decomposition is to find a specific set of matrix sequences A 1 , A 2 …A n ∈R m×n , so that the intersection of the row space and column space of any binary matrix in the sequence is an empty set dimension(C(A i )∩C(A j ))=dimension(R(A i )∩R(A j ))=0, and A=A 1 +A 2 +...+A nThen this set of sequence matrices that satisfy the linear independence of row and column space is the submatrix sequence that satisfies rank conservation of the original matrix A. As for the need for secure two-way matrix addition decomposition, its ultimate goal is to transform the addition of two matrices into the addition of two sets of non-full rank matrix sequences without exposing each other's data information, such as Figure 8 As shown, the realization The formal transformation from binary operation to multi-order operation is used to solve the security risks caused by data leakage in the two-party mixed multiplication protocol.

[0088] For any matrix M∈R m×n There must exist and satisfy the rank r(M) = r ≥ 1. Since the goal of additive decomposition is to decompose a matrix into a series of submatrices whose sum of ranks is additively invariant to the rank of the original matrix, there is no need to decompose a matrix of rank 1. For any (m, n)-dimensional matrix M with a rank greater than 1, it can be simplified to a standard matrix through elementary row and column transformations. This standard matrix is ​​completely determined by three parameters: m, n, and r, where r is the rank of the original matrix. For the standard matrix F, assuming m < n, r(F) = r(E r )=r(M)=r≤min(m,n)=n, so the standard matrix F can be decomposed into 2≤N≤n submatrices F 1 、F 2 ,…F N , and these sub-matrices are all diagonal sparse matrices consisting of 0 elements and 1 elements, for example in I ri It is the sparse diagonal matrix F with rank ri elements 1 obtained by partitioning the standard matrix F according to the non-overlapping rows and columns. i The non-zero sub-blocks of F i With F j When (i≠j), the constraint that the row space and column space do not intersect each other is strictly satisfied. Therefore, for any matrix M in the real number field with dimension (m, n), its standard matrix F can be decomposed into a finite number of sub-matrices that conform to additive rank conservation. And since for any matrix F i For example, performing elementary transformations of rows and columns will not change the rank of the matrix, nor will it change the spatial distribution of the matrix. There is always the equation r(F i )=r(P•F i •Q) holds. Therefore, for any input matrix M, there must exist a set of reversible matrices P and Q whose row transformation matrices and column transformation matrices corresponding to their elementary transformations satisfy PMQ = F. Correspondingly, for the pairwise spatially linearly independent standard matrix sequence F 1、F 2 ,…F N In general, we can obtain a set of matrix sequences M that correspond one-to-one with the standard sequence matrix by performing elementary inverse transformation on the above matrix. i =P -1 ·F i Q -1 (i=1,2,…,N). Since elementary transformation is reversible, and for any two standard submatrices, the distribution of their row and column spaces will not change with elementary row and column transformation, so for any two inverse transformed matrices M i and the matrix M j In terms of row and column space, the distribution characteristics of the standard matrix strictly satisfy dimension(C(M i )∩C(M j ))=dimension(R(M i )∩ R(M j ))=0, and M 1 +M 2 +…+M N =∑M i =∑P -1 •F i •Q -1 =P -1 •∑F i •Q -1 =P -1 FQ -1 = M. Therefore, for any matrix with rank greater than 1, a set of sequence matrices can be found as its submatrices that meet the rank conservation after decomposition.

[0089] The secure two-party matrix addition decomposition technique based on the rank conservation mechanism realizes the decomposition of the original matrix on the one hand, which has the effect of random information confusion after the matrix data is split, and on the other hand, it realizes the rank reduction of the original matrix. The "rank reduction" here specifically refers to the phenomenon that after the original matrix is ​​decomposed into sub-matrices, the rank of each sub-matrix is ​​smaller than the rank of the original matrix. Unlike rank conservation, rank reduction studies the process of individual property changes from the original matrix to the sub-matrix as a point-to-point, while rank conservation studies the systematic process of the entire addition calculation satisfying the invariance of the overall rank before and after the decomposition. Based on the above design ideas and research purposes, this paper proposes a secure two-party matrix addition decomposition strategy (RS2MADT, Rank-Preserving Secure2-PartyMatrix Adition Decomposition Technique) based on the rank conservation mechanism.

[0090] To convert the private matrix A 1The secure two-party matrix addition decomposition strategy based on the rank conservation mechanism is used as an example to decompose multiple sub-matrices. Therefore, in step S11, the first computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix A 1 Decompose multiple sub-matrices, including:

[0091] (1) The first computing participant determines the private matrix A 1 rank, and determine the private matrix A 1 Is the rank of 1? If the private matrix A 1 The rank is 1, which means there is only one basis vector in the row space and column space. In this case, no matrix decomposition is performed and the matrix A is kept 1 Output as is; the private matrix A 1 No matrix decomposition is performed.

[0092] (2) If the private matrix A 1 The rank of is not 1, then for the private matrix A 1 Perform elementary row transformation (using Guass-Jordan elimination method to achieve elementary row transformation) to obtain the private matrix A 1 The row echelon matrix and the non-singular matrix P a ; Using the same method to perform elementary column transformation on the row echelon matrix to obtain the private matrix A 1 The standard form matrix F and the non-singular matrix Q a .

[0093] (3) According to the private matrix A 1 The rank of the private matrix A is divided into rows and columns without overlapping blocks. 1 The preset number N of matrix decompositions 1 , decompose the standard matrix F into N 1 A standard submatrix, F 1 、F 2 ,…F N ∈R m×n ; Each of the standard submatrices; Each of the sparse diagonal matrices is a sparse diagonal matrix consisting of 0 elements and 1 elements. 2≤N 1 ≤n; Among them I ri (i=1,2,…,N 1 ) is the standard submatrix F obtained after partitioning i The non-zero sub-blocks of .

[0094] (4) According to the non-singular matrix P a and the non-singular matrix Q a , for each of the standard submatrices Fi Do a reversible transformation and get N 1 The sub-matrix

[0095] Use the above steps (1) to (4) to calculate the private matrix B 2 Do similar processing and convert the private matrix B into 2 Decompose into T 1 The final target submatrix can be obtained by linearly independent submatrices in row and column space.

[0096] Step S12: For each first sub-matrix product combination, the first computing participant and the second computing participant respectively input the sub-matrix A in the first sub-matrix product combination based on the secure two-party matrix multiplication computing protocol. 1i and the submatrix B 2j Perform a secure two-way matrix multiplication calculation to obtain the first matrix multiplication calculation result corresponding to each first sub-matrix product combination; apply an obfuscation technique to each first matrix multiplication calculation result to randomly split it into matrices V a1e and the matrix V b1e ; The private matrix A 1 Each of the sub-matrices and the private matrix B 2 Each of the sub-matrices is randomly combined to obtain N 1 ×T 1 The first sub-matrix product combination; i = 1, 2, ..., N 1 ; j = 1, 2, ..., T 1 ; e = 1, 2, ..., N 1 ×T 1 .

[0097] The step S12 corresponding to each first sub-matrix product combination may be executed in parallel.

[0098] Step S13: multiply each of the first sub-matrices by the corresponding matrix V a1e The matrix V is obtained a1 , according to the matrix V corresponding to each of the first sub-matrix product combinations b1e The matrix V is obtained b1 .

[0099] Step S2: The first computing participant and the second computing participant respectively input the private matrix A based on the secure two-party matrix multiplication computing protocol 2 and the private matrix B 1 Perform a secure two-way matrix multiplication calculation to obtain a second matrix multiplication calculation result, and apply an obfuscation technique to the second matrix multiplication calculation result to randomly split it into matrices Va2 and the matrix V b2 , and sent to the first computing participant and the second computing participant respectively; the matrix V a2 and the matrix V b2 Satisfaction: V a2 +V b2 =B 1 ×A 2 ; Step S2 and step S1 can be executed in parallel.

[0100] Among them, Fig. 9 As shown, step S2 specifically includes:

[0101] Step S21: The first computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to convert the private matrix A 2 Decompose the private matrix B into multiple sub-matrices; the second computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix B 1 Decompose into multiple sub-matrices; the private matrix The private matrix The private matrix A 2 The rank of each submatrix A is equal to 2s The private matrix B 2 The rank of each submatrix B is equal to 1f The sum of the ranks.

[0102] In step S21, the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism is used to decompose the private matrix A 2 and private matrix B 1 The specific decomposition process involved in decomposing multiple sub-matrices can refer to step (1) to step (4) in step S11, and the specific process will not be repeated. Step S21 and step S11 can be executed in parallel.

[0103] Step S22: For each second sub-matrix product combination, the first computing participant and the second computing participant respectively input the sub-matrix A in the second sub-matrix product combination based on the secure two-party matrix multiplication computing protocol. 2s and the submatrix B 1f Perform a secure two-way matrix multiplication calculation to obtain the second matrix multiplication calculation result corresponding to each second sub-matrix product combination; apply an obfuscation technique to each second matrix multiplication calculation result to randomly split it into matrices V a2g and the matrix V b2g ; The private matrix A 2 Each of the sub-matrices and the private matrix B 2Each of the second sub-matrices is randomly combined to obtain N 2 xT 2 The product combination of the second sub-matrices; s = 1, 2, ..., N 2 ; f = 1, 2, ..., T 2 ; g = 1, 2, ..., N 2 ×T 2 Step S22 and step S12 may be performed in parallel.

[0104] The step S22 corresponding to each second sub-matrix product combination may be executed in parallel.

[0105] Step S23: multiply the matrix V corresponding to each of the second sub-matrices by product combination a2g The matrix V is obtained a2 , according to the matrix V corresponding to each second sub-matrix product combination b2g The matrix V is obtained b2 Step S23 and step S13 may be performed in parallel.

[0106] Step S3: The first computing participant performs a private matrix multiplication to obtain a matrix V a0 ; The matrix V a0 Satisfaction: V a0 =A 1 ×A 2 The step S3 and the step S1 may be performed in parallel.

[0107] Step S4: The second computing participant performs a private matrix multiplication to obtain a matrix V b0 ; The matrix V b0 Satisfaction: V b0 =B 1 ×B 2 The step S4 and the step S1 may be performed in parallel.

[0108] Step S5: The first computing participant converts the matrix V a0 , the matrix V a1 and the matrix V a2 Perform local private calculations to obtain the matrix V a .

[0109] Step S6: The second computing participant converts the matrix V b0 , the matrix V b1 and the matrix V b2 Perform local private calculations to obtain the matrix V b .

[0110] Step S7: The first computing participant and the second computing participant respectively convert the matrix V a and the matrix V b The calculation requester is sent to the secure two-party mixed multiplication calculation, and the calculation requester calculates the privacy calculation result.

[0111] The present invention adopts the rank conservation mechanism in the matrix addition decomposition proposed in steps (1) to (4) in the first step S11 and the secure two-party matrix addition decomposition technology based on the rank conservation mechanism to solve the security problem of binary matrix multiplication when the input matrix is ​​full rank, which the existing technical solutions cannot solve. A hybrid multiplication calculation scheme in a multi-link 2PMP parallel computing mode is implemented by using the secure two-party matrix multiplication basic protocol and the secure two-party matrix addition decomposition technology based on the rank conservation mechanism, which solves the original data privacy risk caused by the leakage of data results that the existing technical solutions cannot solve.

[0112] In order to enable those skilled in the art to more clearly understand the privacy computing method of this embodiment based on the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism, as shown in Fig.10 As shown, the decomposition of two sub-matrices is used as a specific example to illustrate:

[0113] Step 1: Participant Alice and participant Bob combine the secure two-party matrix addition decomposition strategy (RS2MADT) based on rank conservation mechanism proposed in 4.4 to convert the initial input matrix A 1 ∈R m×t and B 1 ∈R m×t According to steps (1) to (4) of RS2MADT in step S11, it is decomposed into two sub-matrices A with non-full column rank 11 ∈R m×t , A 12 ∈R m×t and B 11 ∈R m×t , B 12 ∈R m×t , and is stored in the local private database as a new input matrix after random splitting. The decomposition process satisfies the relationship A 1 +B 1 =(A 11 +A 12 )+(B 11 +B 12 ), A 11 +A 12 =A 1 , B 11 +B 12 =B 1 From the rank conservation property of matrix additive decomposition RMADM, we know that the rank of the matrix before and after decomposition satisfies r(A1 )=r(A 11 )+r(A 12 ), r(B 1 )=r(B 11 )+r(B 12 ), and the decomposed submatrix will satisfy r(A 11 )<t,r(A 12 )<t,r(B 11 )<t,r(B 12 )<t columns are not full rank relations.

[0114] Step 2: Synchronous with step 1, the participant Alice node and the participant Bob node convert the initial input matrix A into 2 ∈R t×n and B 2 ∈R t×n According to steps 1 to 5 of RS2MADT, it is decomposed into two sub-matrices A with non-full row rank 21 ∈R t×n , A 22 ∈R t×n and B 21 ∈R t×n , B 22 ∈R t×n , and is stored in the local private database as a new input matrix after random splitting. The decomposition process satisfies the relationship A 2 +B 2 =(A 21 +A 22 )+(B 21 +B 22 ), A 21 +A 22 =A 2 , B 21 +B 22 =B 2 From the rank conservation property of matrix additive decomposition RMADM proposed in 4.3, we know that the rank of the matrix before and after decomposition satisfies r(A 2 )=r(A 21 )+r(A 22 ), r(B 2 )=r(B 21 )+r(B 22 ), and the decomposed submatrix will satisfy r(A 21 )<t,r(A 22 )<t,r(B 21 )<t,r(B 22 )<t rows with non-full rank.

[0115] Step 3: Participant Alice’s node performs private matrix multiplication calculation V locallya0 =A 1 ×A 2 , and after the calculation is completed, the private matrix V a0 ∈R m×n Stored in a local private database.

[0116] Step 4: Synchronous with step 3, the participant Bob's node performs the private matrix multiplication calculation V locally b0 =B 1 ×B 2 , and after the calculation is completed, the private matrix V b0 ∈R m×n Stored in a local private database.

[0117] Step 5: Participant Alice’s node and participant Bob’s node input their converted private matrices A based on the secure two-party matrix multiplication protocol 2PMP. 11 ∈R m×t and B 21 ∈R t×n The first round of secure two-way matrix multiplication calculation is performed. After the first 2PMP module calculation is completed, the intermediate result of this round of calculation will be randomly split into matrices V according to the random obfuscation technology SMDOT. a1 、V b1 ∈R m×n They are sent to the participant Alice’s node and the participant Bob’s node respectively, and these two private output matrices satisfy the relation V a1 +V b1 =A 11 ×B 21 =M 1 .

[0118] Step 6: At the same time, the participant Alice node and the participant Bob node input their respective converted private matrices A based on the secure two-party matrix multiplication calculation protocol 2PMP 11 ∈R m×t and B 22 ∈R t×n The second round of secure two-way matrix multiplication is performed in parallel with step 5 above. After the second 2PMP module is completed, the intermediate result of this round of calculation will be randomly split into matrices V according to the random obfuscation technology SMDOT a2 、V b2 ∈R m×n They are sent to the participant Alice’s node and the participant Bob’s node respectively, and these two private output matrices satisfy the relation V a2 +V b2 =A 11 ×B 22 =M 2 .

[0119] Step 7: Participant Alice’s node and participant Bob’s node input their converted private matrices A based on the secure two-party matrix multiplication protocol 2PMP. 12 ∈R m×t and B 21 ∈R t×n The third round of secure two-way matrix multiplication is performed in parallel with the above steps 5-6. After the third 2PMP module is completed, the intermediate result of this round of calculation will be randomly split into matrices V according to the random obfuscation technology SMDOT a3 、V b3 ∈R m×n They are sent to the participant Alice’s node and the participant Bob’s node respectively, and these two private output matrices satisfy the relation V a3 +V b3 =A 12 ×B 21 =M 3 .

[0120] Step 8: Participant Alice’s node and participant Bob’s node input their converted private matrices A based on the secure two-party matrix multiplication protocol 2PMP. 12 ∈R m×t and B 22 ∈R t×n The fourth round of secure two-way matrix multiplication is performed in parallel with the above steps 5 to 7. After the fourth 2PMP module is completed, the intermediate result of this round of calculation will be randomly split into matrices V according to the random obfuscation technology SMDOT. a4 、V b4 ∈R m×n They are sent to the participant Alice’s node and the participant Bob’s node respectively, and these two private output matrices satisfy the relation V a4 +V b4 =A 12 ×B 22 =M 4 .

[0121] Step 9: Participant Alice’s node and participant Bob’s node input their converted private matrices A based on the secure two-party matrix multiplication protocol 2PMP. 21 ∈R m×t and B 11 ∈R t×n The fifth round of secure two-way matrix multiplication is performed in parallel with the above steps 5 to 8. After the fifth 2PMP module is completed, the intermediate result of this round of calculation will be randomly split into matrices V according to the random obfuscation technology SMDOT. a5 、V b5 ∈R m×nThey are sent to the participant Alice’s node and the participant Bob’s node respectively, and these two private output matrices satisfy the relation V a5 +V b5 =B 11 ×A 21 =M 5 .

[0122] Step 10: Participant Alice’s node and participant Bob’s node input their converted private matrices A based on the secure two-party matrix multiplication protocol 2PMP. 22 ∈R m×t and B 11 ∈R t×n The sixth round of secure two-way matrix multiplication is performed in parallel with the above steps 5 to 9. After the sixth 2PMP module is completed, the intermediate result of this round of calculation will be randomly split into matrices V according to the random obfuscation technology SMDOT a6 、V b6 ∈R m×n They are sent to the participant Alice’s node and the participant Bob’s node respectively, and these two private output matrices satisfy the relation V a6 +V b6 =B 11 ×A 22 =M 6 .

[0123] Step 11: Participant Alice’s node and participant Bob’s node input their converted private matrices A based on the secure two-party matrix multiplication protocol 2PMP. 21 ∈R m×t and B 12 ∈R t×n The seventh round of secure two-way matrix multiplication is performed in parallel with the above steps 5 to 10. After the seventh 2PMP module is completed, the intermediate result of this round of calculation will be randomly split into matrices V according to the random obfuscation technology SMDOT a7 、V b7 ∈R m×n They are sent to the participant Alice’s node and the participant Bob’s node respectively, and these two private output matrices satisfy the relation V a7 +V b7 =B 12 ×A 21 =M 7 .

[0124] Step 12: Participant Alice’s node and participant Bob’s node input their converted private matrices A based on the secure two-party matrix multiplication protocol 2PMP. 22 ∈R m×t and B 12 ∈Rt×n The eighth round of secure two-way matrix multiplication is performed in parallel with the above steps 5 to 11. After the eighth 2PMP module is completed, the intermediate result of this round of calculation will be randomly split into matrices V according to the random obfuscation technology SMDOT a8 、V b8 ∈R m×n They are sent to the participant Alice’s node and the participant Bob’s node respectively, and these two private output matrices satisfy the relation V a8 +V b8 =B 12 ×A 22 =M 8 .

[0125] Step 13: Alice's node randomly splits the intermediate calculation result V of the 2PMP matrix multiplication protocol executed in parallel for the first eight rounds of steps 5 to 12. a1 、V a2 、V a3 、V a4 、V a5 、V a6 、V a7 、V a8 and the calculation result V in step 3 a0 Perform secret aggregation and perform private computation V locally a =∑ 0≤i≤8 [V ai ]=V a0 +V a1 +V a2 +V a3 +V a4 +V a5 +V a6 +V a7 +V a8 , where V a ∈R m×n .

[0126] Step 14: The node of participant Bob randomly splits the intermediate calculation result V of the 2PMP matrix multiplication protocol executed in parallel for the first eight rounds of steps 5 to 12. b1 、V b2 、V b3 、V b4 、V b5 、V b6 、V b7 、V b8 and the calculation result V in step 4 b0 Perform secret aggregation and perform private computation V locally b =∑ 0≤j≤8 [V bj ]=V b0 +Vb1 +V b2 +V b3 +V b4 +V b5 +V b6 +V b7 +V b8 , where V b ∈R m ×n .

[0127] Step 15: Participant Alice's node and participant Bob's node respectively aggregate the intermediate calculation results of steps 13 to 14 to obtain the private matrix V a 、V b ∈R m×n Sent to the requester of the secure two-party mixed multiplication calculation, which then summarizes the final calculation result (A 1 +B 1 )×(A 2 +B 2 )=V a +V b Obviously, it is easy to verify that:

[0128] V a +V b =∑ 0≤i≤8 [V ai ]+∑ 0≤j≤8 [V bj ]=(V a0 +V a1 +V a2 +V a3 +V a4 +V a5 +V a6 +V a7 +V a8 )+(V b0 +V b1 +V b2 +V b3 +V b4 +V b5 +V b6 +V b7 +V b8 )=(V a0 +V b0 )+(V a1 +V b1 )+(V a2 +V b2 )+(V a3 +V b3 )+(V a4 +V b4 )+(V a5 +V b5)+(V a6 +V b6 )+(V a7 +V b7 )+(V a8 +V b8 )=A 1 A 2 +B 1 B 2 +A 11 B 21 +A 11 B 22 +A 12 B 21 +A 12 B 22 +B 11 A 21 +B 11 A 22 +B 12 A 21 +B 12 A 22 =(A 11 +A 12 )(A 21 +A 22 )+(B 11 +B 12 )(B 21 +B 22 )+(A 11 +A 12 )(B 21 +B 22 )+(B 11 +B 12 )(A 21 +A 22 )=A 1 B 2 +B 1 A 2 +A 1 A 2 +B 1 B 2 =(A 1 +B 1 )×(A 2 +B 2 )。

[0129] In the present embodiment, in the specific example of decomposing two sub-matrices, the present invention proposes an end-to-end parallel hybrid multiplication calculation scheme suitable for result leakage environment based on the secure two-party matrix multiplication basic protocol 2PMP, and parallelized calculation steps 3 to 12, to solve the problem of large ciphertext space calculation and communication overhead caused by the introduction of homomorphic encryption and inadvertent transmission technology in the prior art. Steps 1 to 2 in the specific example use the secure two-party matrix multiplication basic protocol and the secure two-party matrix addition decomposition technology based on the rank conservation mechanism to realize a hybrid multiplication calculation scheme under the eight-link 2PMP parallel computing mode, solving the original data privacy risk caused by the leakage of data results that the prior art solutions have not solved. Steps 2 to 12 in the specific example involve key steps to ensure the stability of the precision of secure two-party hybrid multiplication. The calculation precision of the invention lies in the use of the 2PMP secure two-party multiplication calculation protocol that can support float64-bit precision calculation at the highest, solving the problem of loss of numerical precision of floating-point calculation due to the limitation of the fixed length of the digits in the ciphertext calculation in the prior art.

[0130] In combination with the privacy computing for secure two-way matrix mixed multiplication proposed in Example 1 and Example 2, the advantages over the prior art are explained:

[0131] 1) Based on the secure two-party matrix multiplication basic protocol 2PMP, the present invention proposes two sets of end-to-end parallel hybrid multiplication calculation schemes suitable for semi-honest environments and result leakage environments, which solve the problems of large ciphertext space calculation and communication overhead caused by the introduction of homomorphic encryption and oblivious transmission technology in the prior art, and realizes a dual-link parallel secure two-party matrix hybrid multiplication calculation scheme that is independent of third-party cloud service providers and has a higher security level, and an eight-link parallel secure two-party matrix hybrid multiplication calculation scheme in scenarios with lower security level requirements.

[0132] 2) The present invention adopts the rank conservation mechanism in matrix additive decomposition that was first proposed and the secure two-party matrix additive decomposition technology based on the rank conservation mechanism to solve the security problem of binary matrix multiplication when the input matrix is ​​full rank, which the existing technical solutions cannot solve, and realizes a random decomposition method with autonomous and controllable rank of the input matrix.

[0133] 3) The present invention adopts a secure two-party matrix multiplication basic protocol and a secure two-party matrix addition decomposition technology based on a rank conservation mechanism to implement a hybrid multiplication calculation scheme under an eight-link 2PMP parallel computing mode, thereby solving the original data privacy risk caused by the leakage of data results that cannot be solved by existing technical solutions, and realizing an efficient and parallel secure two-party matrix hybrid multiplication calculation protocol under a privacy leakage environment.

[0134] 4) The present invention adopts the idea of ​​random obfuscation encryption combined with the high-precision 2PMP secure two-party multiplication protocol to construct two secure two-party hybrid multiplication calculation protocols under different scenarios. Since the 2PMP protocol itself can support up to float64-bit precision, the matrix calculation type it faces is not limited to integers. Compared with the limitations of the obfuscated circuit scheme and the homomorphic encryption scheme that can only process fixed-length decimals and integer types, the applicable scenarios of secure matrix multiplication are richer, and the calculation results have the same order of magnitude of numerical calculation accuracy as the centralized calculation method.

[0135] Embodiment 3

[0136] This embodiment provides a privacy computing system for secure two-party matrix hybrid multiplication, involving two computing participants. The first computing participant holds a private matrix A 1 and private matrix A 2 , the second computing participant holds a private matrix B 1 and private matrix B 2 , the system comprising:

[0137] The first matrix multiplication calculation module 100 is used for the first calculation participant and the second calculation participant to input the private matrix A respectively based on the secure two-party matrix multiplication calculation protocol. 1 and the private matrix B 2 Perform a secure two-way matrix multiplication calculation to obtain a first matrix multiplication calculation result, and apply an obfuscation technique to the first matrix multiplication calculation result to randomly split it into matrices V a1 and the matrix V b1 , and sent to the first computing participant and the second computing participant respectively; the matrix V a1 and the matrix V b1 Satisfaction: V a1 +V b1 =A 1 ×B 2 .

[0138] The second matrix multiplication calculation module 200 is used for the first calculation participant and the second calculation participant to input the private matrix A respectively based on the secure two-party matrix multiplication calculation protocol. 2 and the private matrix B 1 Perform a secure two-way matrix multiplication calculation to obtain a second matrix multiplication calculation result, and apply an obfuscation technique to the second matrix multiplication calculation result to randomly split it into matrices V a2 and the matrix V b2 , and sent to the first computing participant and the second computing participant respectively; the matrix V a2 and the matrix V b2 Satisfaction: V a2 +Vb2 =B 1 ×A 2 .

[0139] The first local private computing module 300 is used for the first computing participant to perform private matrix multiplication to obtain the matrix V a0 ; The matrix V a0 Satisfaction: V a0 =A 1 ×A 2 .

[0140] The second local private computing module 400 is used for the second computing participant to perform private matrix multiplication to obtain the matrix V b0 ; The matrix V b0 Satisfaction: V b0 =B 1 ×B 2 .

[0141] The first matrix calculation module 500 is used by the first calculation participant to convert the matrix V a0 , the matrix V a1 and the matrix V a2 Perform local private calculations to obtain the matrix V a .

[0142] The second matrix calculation module 600 is used for the second calculation participant to convert the matrix V b0 , the matrix V b1 and the matrix V b2 Perform local private calculations to obtain the matrix V b .

[0143] The privacy calculation result output module 700 is used for the first calculation participant and the second calculation participant to respectively output the matrix V a and the matrix V b The calculation requester is sent to the secure two-party mixed multiplication calculation, and the calculation requester calculates the privacy calculation result.

[0144] Wherein, the first matrix multiplication calculation module specifically includes:

[0145] The first matrix decomposition submodule is used for the first computing participant to combine the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix A 1 Decompose the private matrix B into multiple sub-matrices; the second computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix B 2 Decompose into multiple sub-matrices; the private matrix The private matrix The private matrix A1 The rank is equal to the corresponding submatrix A 1i The private matrix B 2 The rank is equal to the corresponding submatrix B 2j The sum of the ranks.

[0146] The first sub-matrix secure two-party matrix multiplication calculation submodule is used for, for each first sub-matrix product combination, the first calculation participant and the second calculation participant respectively input the sub-matrix A in the first sub-matrix product combination based on the secure two-party matrix multiplication calculation protocol. 1i and the submatrix B 2j Perform a secure two-way matrix multiplication calculation to obtain the first matrix multiplication calculation result corresponding to each first sub-matrix product combination; apply an obfuscation technique to each first matrix multiplication calculation result to randomly split it into matrices V a1e and the matrix V b1e ; The private matrix A 1 Each of the sub-matrices and the private matrix B 2 Each of the sub-matrices is randomly combined to obtain N 1 ×T 1 The first sub-matrix product combination; i = 1, 2, ..., N 1 ; j = 1, 2, ..., T 1 ; e = 1, 2, ..., N 1 ×T 1 .

[0147] The first matrix fusion submodule is used to combine the corresponding matrix V according to each of the first sub-matrices. a1e The matrix V is obtained a1 , according to the matrix V corresponding to each of the first sub-matrix product combinations b1e The matrix V is obtained b1 .

[0148] Wherein, the second matrix multiplication calculation module specifically includes:

[0149] The second matrix decomposition submodule is used for the first computing participant to combine the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix A 2 Decompose into multiple sub-matrices; the second computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix B1 into multiple sub-matrices; the private matrix The private matrix The private matrix A 2 The rank of each submatrix A is equal to 2s The private matrix B 2The rank of each submatrix B is equal to 1f The sum of the ranks.

[0150] The second sub-matrix secure two-party matrix multiplication calculation submodule is used for, for each second sub-matrix product combination, the first calculation participant and the second calculation participant respectively input the sub-matrix A in the second sub-matrix product combination based on the secure two-party matrix multiplication calculation protocol. 2s and the submatrix B 1f Perform a secure two-way matrix multiplication calculation to obtain the second matrix multiplication calculation result corresponding to each second sub-matrix product combination; apply an obfuscation technique to each second matrix multiplication calculation result to randomly split it into matrices V a2g and the matrix V b2g ; The private matrix A 2 Each of the sub-matrices and the private matrix B 2 Each of the sub-matrices is randomly combined to obtain N 2 ×T 2 The sub-matrix product combination; s = 1, 2, ..., N 2 ; f = 1, 2, ..., T 2 ; g = 1, 2, ..., N 2 ×T 2 .

[0151] The second matrix fusion submodule is used to combine the corresponding matrix V according to each product of the second submatrices. a2g The matrix V is obtained a2 , according to the matrix V corresponding to each second sub-matrix product combination b2g The matrix V is obtained b2 .

[0152] The first matrix decomposition submodule specifically includes: a first matrix decomposition unit and a second matrix decomposition unit.

[0153] The first matrix decomposition unit is used by the first computing participant to combine the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix A 1 Decompose into multiple sub-matrices.

[0154] The second matrix decomposition unit is used for the second computing participant to combine the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix B 2 Decompose into multiple sub-matrices.

[0155] The first matrix decomposition unit specifically includes:

[0156] A judgment subunit is used for the first computing participant to determine the private matrix A1 rank, and determine the private matrix A 1 Is the rank of 1? If the private matrix A 1 The rank of is 1, then the private matrix A 1 No matrix decomposition is performed.

[0157] The elementary transformation subunit is used if the private matrix A 1 The rank of is not 1, then for the private matrix A 1 Perform elementary row transformation to obtain the private matrix A 1 The row echelon matrix and the non-singular matrix P a ; Perform elementary column transformation on the row echelon matrix to obtain the private matrix A 1 The standard form matrix F and the non-singular matrix Q a .

[0158] The matrix decomposition subunit is used to decompose the private matrix A 1 The rank of the private matrix A is divided into rows and columns without overlapping blocks. 1 The standard matrix F is decomposed into N 1 standard sub-matrices; each of the standard sub-matrices; each of the sparse diagonal matrices is a sparse diagonal matrix consisting of 0 elements and 1 elements.

[0159] The inverse transformation subunit is used to transform the non-singular matrix P a and the non-singular matrix Q a , perform a reversible transformation on each of the standard submatrices to obtain N 1 The submatrix A 1i .

[0160] The second matrix decomposition unit has the same structure as the second matrix decomposition unit, except that the object processed is the private matrix B. 2 .

[0161] Embodiment 4

[0162] This embodiment provides an electronic device, including a memory and a processor, the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute a privacy computing method for secure bidirectional matrix mixed multiplication of embodiment one or embodiment two.

[0163] Optionally, the above-mentioned electronic device may be a server.

[0164] In addition, an embodiment of the present invention also provides a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements a privacy computing method for secure bidirectional matrix mixed multiplication of embodiment one or embodiment two.

[0165] Embodiments of the present invention may be provided as methods, systems, or computer program products. Therefore, the present invention may take the form of a complete hardware embodiment, a complete software embodiment, or an embodiment combining software and hardware. Moreover, the present invention 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 code.

[0166] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems), and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes 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 a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0167] Each embodiment focuses on the differences from other embodiments, and the same or similar parts between the embodiments can be referred to each other. As for the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part description.

[0168] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A privacy computing method for secure two-party matrix hybrid multiplication, involving two computing participants, a first computing participant holding private matrices A1 and A2, and a second computing participant holding private matrices B1 and B2, characterized in that: The method comprises: The first computing participant and the second computing participant respectively input the private matrix A1 and the private matrix B2 to perform a secure two-party matrix multiplication calculation based on the secure two-party matrix multiplication calculation protocol, obtain a first matrix multiplication calculation result, and apply an obfuscation technique to the first matrix multiplication calculation result to randomly split it into matrices V a1 and the matrix V b1 , and sent to the first computing participant and the second computing participant respectively; the matrix V a1 and the matrix V b1 Satisfaction: V a1 +V b1 =A1×B2; The first computing participant and the second computing participant respectively input the private matrix A2 and the private matrix B1 to perform a secure two-party matrix multiplication calculation based on the secure two-party matrix multiplication calculation protocol, obtain a second matrix multiplication calculation result, and apply an obfuscation technique to the second matrix multiplication calculation result to randomly split it into matrices V a2 and the matrix V b2 , and sent to the first computing participant and the second computing participant respectively; the matrix V a2 and the matrix V b2 Satisfaction: V a2 +V b2 =B1×A2; The first computing participant performs a private matrix multiplication to obtain a matrix V a0 ; The matrix V a0 Satisfaction: V a0 =A1×A2; The second computing participant performs a private matrix multiplication to obtain a matrix V b0 ; The matrix V b0 Satisfaction: V b0 =B1×B2; The first computing participant converts the matrix V a0 , the matrix V a1 and the matrix V a2 Perform local private calculations to obtain the matrix V a ; The second computing participant converts the matrix V b0 , the matrix V b1 and the matrix V b2 Perform local private calculations to obtain the matrix V b ; The first computing participant and the second computing participant respectively convert the matrix V a and the matrix V b The calculation requester is sent to the secure two-party mixed multiplication calculation, and the calculation requester calculates the privacy calculation result.

2. The method according to claim 1, characterized in that The first computing participant and the second computing participant respectively input the private matrix A1 and the private matrix B2 to perform a secure two-party matrix multiplication calculation based on the secure two-party matrix multiplication calculation protocol, obtain a first matrix multiplication calculation result, and apply an obfuscation technique to the first matrix multiplication calculation result to randomly split it into matrices V a1 and the matrix V b1 , specifically including: The first computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix A1 into multiple sub-matrices; the second computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix B2 into multiple sub-matrices; the private matrix The private matrix The rank of the private matrix A1 is equal to the rank of the corresponding submatrices A 1i The rank of the private matrix B2 is equal to the rank of the corresponding sub-matrices B 2j The sum of the ranks of For each first sub-matrix product combination, the first computing participant and the second computing participant respectively input the sub-matrix A in the first sub-matrix product combination based on the secure two-party matrix multiplication computing protocol. 1i and the submatrix B 2j Perform a secure two-way matrix multiplication calculation to obtain the first matrix multiplication calculation result corresponding to each first sub-matrix product combination; apply an obfuscation technique to each first matrix multiplication calculation result to randomly split it into matrices V a1e and the matrix V b1e ; Each of the submatrices of the private matrix A1 is randomly combined with each of the submatrices of the private matrix B2 to obtain N1×T1 product combinations of the first submatrices; i=1, 2, ..., N1; j=1, 2, ..., T1; e=1, 2, ..., N1×T1; The matrix V corresponding to each first sub-matrix product combination a1e The matrix V is obtained a1 , according to the matrix V corresponding to each of the first sub-matrix product combinations b1e The matrix V is obtained b1 .

3. The method according to claim 2, characterized in that The first computing participant and the second computing participant respectively input the private matrix A2 and the private matrix B1 to perform a secure two-party matrix multiplication calculation based on the secure two-party matrix multiplication calculation protocol, obtain a second matrix multiplication calculation result, and apply an obfuscation technique to the second matrix multiplication calculation result to randomly split it into matrices V a2 and the matrix V b2 , including: The first computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix A2 into multiple sub-matrices; the second computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix B1 into multiple sub-matrices; the private matrix The private matrix The rank of the private matrix A2 is equal to the rank of the corresponding submatrices A 2s The rank of the private matrix B2 is equal to the corresponding sub-matrices B 1f The sum of the ranks of For each second sub-matrix product combination, the first computing participant and the second computing participant respectively input the sub-matrix A in the second sub-matrix product combination based on the secure two-party matrix multiplication computing protocol. 2s and the submatrix B 1f Perform a secure two-way matrix multiplication calculation to obtain the second matrix multiplication calculation result corresponding to each second sub-matrix product combination; apply an obfuscation technique to each second matrix multiplication calculation result to randomly split it into matrices V a2g and the matrix V b2g ; Each of the submatrices of the private matrix A2 is randomly combined with each of the second submatrices of the private matrix B2 to obtain N2×T2 product combinations of the second submatrices; s=1, 2, ..., N2; f=1, 2, ..., T2; g=1, 2, ..., N2×T2; The matrix V corresponding to each second sub-matrix product combination a2g The matrix V is obtained a2 , according to the matrix V corresponding to each second sub-matrix product combination b2g The matrix V is obtained b2 .

4. The method according to claim 2, characterized in that: The first computing participant combines the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism to decompose the private matrix A1 into multiple sub-matrices, specifically including: The first computing participant determines the rank of the private matrix A1, and determines whether the rank of the private matrix A1 is 1; if the rank of the private matrix A1 is 1, the private matrix A1 is not subjected to matrix decomposition; If the rank of the private matrix A1 is not 1, an elementary row transformation is performed on the private matrix A1 to obtain a row echelon matrix of the private matrix A1 and a non-singular matrix P a ; Perform elementary column transformation on the row echelon matrix to obtain the standard matrix F and non-singular matrix Q of the private matrix A1 a ; According to the rank of the private matrix A1, the row-column block division method without overlap and the preset number of matrix decompositions of the private matrix A1 are adopted to decompose the standard matrix F into N1 standard sub-matrices; each of the standard sub-matrices is a sparse diagonal matrix composed of 0 elements and 1 elements; According to the non-singular matrix P a and the non-singular matrix Q a , perform a reversible transformation on each of the standard submatrices to obtain N1 submatrices A 1i .

5. A privacy computing system for secure two-party matrix hybrid multiplication, involving two computing participants, a first computing participant holding private matrices A1 and A2, and a second computing participant holding private matrices B1 and B2, characterized in that: The system comprises: The first matrix multiplication calculation module is used for the first calculation participant and the second calculation participant to input the private matrix A1 and the private matrix B2 respectively based on the secure two-party matrix multiplication calculation protocol to perform secure two-party matrix multiplication calculation, obtain the first matrix multiplication calculation result, and apply the obfuscation technology to the first matrix multiplication calculation result to randomly split it into matrices V a1 and the matrix V b1 , and sent to the first computing participant and the second computing participant respectively; the matrix V a1 and the matrix V b1 Satisfaction: V a1 +V b1 =A1×B2; The second matrix multiplication calculation module is used for the first calculation participant and the second calculation participant to input the private matrix A2 and the private matrix B1 respectively to perform a secure two-party matrix multiplication calculation based on the secure two-party matrix multiplication calculation protocol, obtain a second matrix multiplication calculation result, and apply an obfuscation technology to the second matrix multiplication calculation result to randomly split it into matrices V a2 and the matrix V b2 , and sent to the first computing participant and the second computing participant respectively; the matrix V a2 and the matrix V b2 Satisfaction: V a2 +V b2 =B1×A2; The first local private computing module is used for the first computing participant to perform private matrix multiplication to obtain the matrix V a0 ; The matrix V a0 Satisfaction: V a0 =A1×A2; The second local private computing module is used for the second computing participant to perform private matrix multiplication to obtain the matrix V b0 ; The matrix V b0 Satisfaction: V b0 =B1×B2; The first matrix calculation module is used for the first calculation participant to convert the matrix V a0 , the matrix V a1 and the matrix V a2 Perform local private calculations to obtain the matrix V a ; The second matrix calculation module is used for the second calculation participant to convert the matrix V b0 , the matrix V b1 and the matrix V b2 Perform local private calculations to obtain the matrix V b ; The privacy computing result output module is used for the first computing participant and the second computing participant to respectively output the matrix V a and the matrix V b The calculation requester is sent to the secure two-party mixed multiplication calculation, and the calculation requester calculates the privacy calculation result.

6. The system according to claim 5, characterized in that The first matrix multiplication calculation module specifically includes: The first matrix decomposition submodule is used for the first computing participant to decompose the private matrix A1 into multiple sub-matrices in combination with the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism; the second computing participant to decompose the private matrix B2 into multiple sub-matrices in combination with the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism; the private matrix The private matrix The rank of the private matrix A1 is equal to the rank of the corresponding submatrices A 1i The rank of the private matrix B2 is equal to the rank of the corresponding sub-matrices B 2j The sum of the ranks of The first sub-matrix secure two-party matrix multiplication calculation submodule is used for, for each first sub-matrix product combination, the first calculation participant and the second calculation participant respectively input the sub-matrix A in the first sub-matrix product combination based on the secure two-party matrix multiplication calculation protocol. 1i and the submatrix B 2j Perform a secure two-way matrix multiplication calculation to obtain the first matrix multiplication calculation result corresponding to each first sub-matrix product combination; apply an obfuscation technique to each first matrix multiplication calculation result to randomly split it into matrices V a1e and the matrix V b1e ; Each of the submatrices of the private matrix A1 is randomly combined with each of the submatrices of the private matrix B2 to obtain N1×T1 product combinations of the first submatrices; i=1, 2, ..., N1; j=1, 2, ..., T1; e=1, 2, ..., N1×T1; The first matrix fusion submodule is used to combine the corresponding matrix V according to each of the first sub-matrices. a1e The matrix V is obtained a1 , according to the matrix V corresponding to each of the first sub-matrix product combinations b1e The matrix V is obtained b1 .

7. The system according to claim 6, characterized in that The second matrix multiplication calculation module specifically includes: The second matrix decomposition submodule is used for the first computing participant to decompose the private matrix A2 into multiple sub-matrices in combination with the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism; the second computing participant to decompose the private matrix B1 into multiple sub-matrices in combination with the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism; the private matrix The private matrix The rank of the private matrix A2 is equal to the rank of the corresponding submatrices A 2s The rank of the private matrix B2 is equal to the corresponding sub-matrices B 1f The sum of the ranks of The second sub-matrix secure two-party matrix multiplication calculation submodule is used for, for each second sub-matrix product combination, the first calculation participant and the second calculation participant respectively input the sub-matrix A in the second sub-matrix product combination based on the secure two-party matrix multiplication calculation protocol. 2s and the submatrix B 1f Perform a secure two-way matrix multiplication calculation to obtain the second matrix multiplication calculation result corresponding to each second sub-matrix product combination; apply an obfuscation technique to each second matrix multiplication calculation result to randomly split it into matrices V a2g and the matrix V b2g ; Each submatrix of the private matrix A2 is randomly combined with each submatrix of the private matrix B2 to obtain N2×T2 submatrix product combinations; s=1,2,...,N2; f=1,2,...,T2; g=1,2,...,N2×T2; The second matrix fusion submodule is used to combine the corresponding matrix V according to each product of the second submatrices. a2g The matrix V is obtained a2 , according to the matrix V corresponding to each second sub-matrix product combination b2g The matrix V is obtained b2 .

8. The system according to claim 6, characterized in that The first matrix decomposition submodule specifically includes: a first matrix decomposition unit and a second matrix decomposition unit; The first matrix decomposition unit is used by the first computing participant to decompose the private matrix A1 into a plurality of sub-matrices in combination with a secure two-party matrix addition decomposition strategy based on a rank conservation mechanism; The second matrix decomposition unit is used for the second computing participant to decompose the private matrix B2 into a plurality of sub-matrices in combination with the secure two-party matrix addition decomposition strategy based on the rank conservation mechanism; The first matrix decomposition unit specifically includes: A judging subunit, used for the first computing participant to determine the rank of the private matrix A1, and to judge whether the rank of the private matrix A1 is 1; if the rank of the private matrix A1 is 1, the private matrix A1 is not subjected to matrix decomposition; The elementary transformation subunit is used to perform elementary row transformation on the private matrix A1 if the rank of the private matrix A1 is not 1, so as to obtain the row echelon matrix and the non-singular matrix P of the private matrix A1. a ; Perform elementary column transformation on the row echelon matrix to obtain the standard matrix F and non-singular matrix Q of the private matrix A1 a ; A matrix decomposition subunit, for decomposing the standard matrix F into N1 standard sub-matrices according to the rank of the private matrix A1 by adopting a row-column block non-overlapping partitioning method and a preset number of matrix decompositions of the private matrix A1; each of the standard sub-matrices is a sparse diagonal matrix consisting of 0 elements and 1 elements; The inverse transformation subunit is used to transform the non-singular matrix P a and the non-singular matrix Q a , perform a reversible transformation on each of the standard submatrices to obtain N1 submatrices A 1i .

9. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, it implements a privacy computing method for secure bidirectional matrix mixed multiplication as described in any one of claims 1 to 4.

Citation Information

Patent Citations

  • End-to-end efficient and safe two-square matrix inversion privacy computing device and method

    CN116760539A

  • Method and apparatus for jointly performing data processing by two parties

    WO2021197037A1