Multi-party longitudinal logistic regression privacy protection method and device based on secret sharing

Abstract

The invention discloses a multi-party longitudinal logistic regression privacy protection method and device based on secret sharing, an assistant is responsible for generating a large number of Bever triads before a protocol starts and distributing the Bever triads to participants, and firstly, model weights are distributed to the participants by using a secret sharing technology; then each participant cooperatively calculates a forward propagation value cross term; then, obtaining an output value based on respective mastered data sets, model weight fragments and secret fragments of forward propagation value cross terms, and calculating a final forward propagation value by the active party; then, the active party and the passive party cooperatively calculate model gradient cross terms, obtain secret fragments of the model gradient cross terms, and locally update model weight fragments; according to the method, a large number of Bever triads are generated before the protocol is started and are distributed to the participants, so that the communication traffic and the calculation amount can be transferred before the protocol is started, and the secret fragmentation can be carried out among a plurality of participants, so that the method can be expanded to three or more participants.

Description

Technical Field

[0001] The present invention relates to the field of information security technology, and in particular to a multi-party longitudinal logistic regression privacy protection method and device based on secret sharing, which can be used for safe and practical logistic regression analysis. Background Art

[0002] With the increasing attention paid to data privacy, how to build machine learning (ML) models on different data sources and protect privacy is becoming more and more popular. Vertical federated learning (VFL) considers the situation where different parties jointly train ML models on partitioned features, which is suitable for many real-world cross-enterprise collaborations. Logistic regression (LR) is a classic algorithm in machine learning. Due to the simplicity of LR and its effectiveness in many binary classification tasks, model training or inference with privacy protection of logistic regression based on vertically distributed datasets has attracted widespread attention.

[0003] Existing privacy-preserving solutions for longitudinal logistic regression can be divided into two categories based on how private features are handled. The first category leverages secure multi-party computation (MPC) techniques such as homomorphic encryption (HE) and secret sharing (SS) to achieve complete privacy guarantees. Typically, they leverage data outsourcing techniques widely used in database services to outsource datasets to non-colluding servers for ML training or inference. To maintain privacy, all feature values ​​are converted to HE or SS variables when outsourcing, so that the server cannot know the original values. However, this approach is not suitable for many real-world datasets. For high-dimensional and sparse datasets, the outsourced features become completely dense because the non-zero feature indices are also private. This prevents us from performing sparsification calculations, resulting in performance risks when sparsity is high. The second category follows the split learning paradigm, which does not outsource the original dataset and can therefore handle sparse features well. Typically, each party maintains an underlying model in plaintext that extracts forward activations using its own private features. The activations of each party are exchanged and fed into the top model for prediction. However, the values ​​generated in the underlying model are in plaintext, which can lead to data leakage. This kind of data leakage is prohibited or even illegal in real-world applications.

[0004] In 2022, Fu, Fangcheng and others proposed BlindFL, a secure and practical longitudinal logistic regression privacy protection scheme based on HE and SS (the result "BlindFL: Vertical Federated Machine Learning without Peeking into Your Data." was published in Proceedings of the 2022 International Conference on Management of Data). BlindFL keeps private datasets within each party without outsourcing, and unifies functions through an important component called "Source Layer". By doing so, BlindFL can support sparse input features while achieving security guarantees. Figure 1 As shown in Figure 1, the solution implemented by BlindFL can be decomposed into two parts: forward propagation and back propagation. In the forward propagation, the joint source layer is used as the basic building block to combine the features of the passive party A and the active party B (with the true label value), and its output is represented as Z. Assume that the data sets mastered by A and B are X A , X B , that is, Z=X A W A +X B W B The non-federated submodule in side B, namely the top model, plays the role of classifier / predictor. During back propagation, B calculates the loss function through the final predicted value and the true label, and then back propagates along the top model to obtain the partial derivative of the loss function with respect to Z The model gradient can be further calculated by the chain rule Finally, the model weights of the source layer are updated using the model gradients.

[0005] Specifically, in order to prevent the data sets held by A and B and the label values ​​of B from being leaked, BlindFL prohibits both parties from owning their own or the other party’s underlying model, namely W A and W B During forward propagation, the passive party A is prohibited from revealing the forward propagation value Z. During backward propagation, A and B cannot reveal the other party's model gradient, and A cannot reveal its own model gradient. In order to implement the above restrictions, BlindFL uses SS to convert W A and W B The secret is shared with A and B, and A and B only know W A and W BIn order to calculate the forward propagation value Z, A and B need to homomorphically encrypt one of the shards and send it to the other party, and perform homomorphic scalar multiplication of the obtained ciphertext shard with the plaintext shard they have, and convert the result into a secret sharing form. The above process requires A and B to communicate for many rounds, and finally B can only obtain the forward propagation value Z. The reverse propagation process is similar. After many rounds of communication, A and B can obtain the value used to update W. A and W B Model gradients of shards shards.

[0006] However, the BlindFL solution has insufficient scalability. When expanded to multiple parties, the BlindFL solution needs to be re-customized. In addition, the solution requires too many communication rounds. Especially when expanded to multiple parties, communication between two parties is required, and the time overhead caused by communication is huge. Summary of the invention

[0007] The present invention aims at the technical problem that the privacy protection scheme of the longitudinal logistic regression in the prior art is not easy to be extended to multiple participants and a large number of communication rounds, and proposes a multi-party longitudinal logistic regression privacy protection method and device based on secret sharing. The scheme can be used for secure multi-party longitudinal logistic regression and achieve fewer communication rounds. Since the method of the present invention does not outsource the original data, it is suitable for high-dimensional and sparse data sets. At the same time, each party cannot master any underlying model, so it can achieve the same security guarantee as the above scheme.

[0008] In order to solve the above technical problems, the technical solution adopted by the present invention is as follows:

[0009] In a first aspect, a multi-party longitudinal logistic regression privacy protection method based on secret sharing is provided, which is applied to multiple participants, including active parties and passive parties. The multiple participants each have their own data sets, and the facilitator is responsible for generating a large number of Beaver triples and distributing them to each participant before the start of the protocol. The multi-party longitudinal logistic regression privacy protection method includes:

[0010] Using secret sharing technology to distribute model weights to each participant;

[0011] Each participant collaboratively calculates the forward propagation value cross term based on the assigned model weight shards and triplet shards, and obtains the secret shard of the forward propagation value cross term, where the forward propagation value cross term is part of the forward propagation value calculation;

[0012] Each participant obtains the output value based on their own data set, model weight shard and secret shard of the cross term of the forward propagation value. The passive party sends the obtained output value to the active party. The active party obtains the forward propagation value based on the output value of the passive party and its own output value. The forward propagation value is used as the input of the top-level model to obtain the final prediction value of the model.

[0013] The active party and the passive party collaboratively calculate the model gradient cross term and obtain the secret shard of the model gradient cross term. Each participant locally updates the model weight shard based on the secret shard of the model gradient cross term, where the model gradient cross term is the part of the model gradient that needs to be collaboratively calculated by the active party and the passive party.

[0014] In one embodiment, the model weights are distributed to each participant using secret sharing technology, including:

[0015] The model weight is distributed to each participant through W=U+v, and the model weight shards U and v are held by different participants, where W is the model weight.

[0016] In one embodiment, each participant collaboratively calculates the forward propagation value cross term based on the allocated model weight shards and triplet shards, and obtains the secret shard of the forward propagation value cross term, including:

[0017] Each participant reconstructs the parameters for calculating the cross term based on the allocated model weight shards and triplet shards, and calculates the secret shard of the forward propagation value cross term based on the parameters for calculating the cross term and the triplet shards.

[0018] In one embodiment, each participant reconstructs parameters for calculating cross terms based on the allocated model weight shards and triplet shards, and calculates secret shards of forward propagation value cross terms based on the parameters for calculating cross terms and triplet shards, including:

[0019] Each participant initializes model weight sharding and data set sharding according to the assigned model weights and owned data sets;

[0020] Each participant calculates a shard for calculating the parameters of the cross term based on the initialized model weight shards, data set shards, and triplet shards, and calculates the parameters for calculating the cross term based on the shards for calculating the parameters of the cross term;

[0021] Each participant calculates a secret shard of the forward propagation value cross term based on the parameters used to calculate the cross term and the triplet shard.

[0022] In one implementation, each participant obtains an output value based on their respective data sets, model weight shards, and cross-term shards, including:

[0023] Each participant multiplies the data set they own by the model weight shard and adds it to the shard of the cross term to get the output value.

[0024] In one implementation, the active party obtains a forward propagation value according to the output value of the passive party and its own output value, including:

[0025] The active party adds the output value received from the passive party to its own output value to obtain the forward propagation value.

[0026] In one implementation, the active party and the passive party collaboratively calculate the model gradient cross term and obtain the secret shard of the model gradient cross term, including:

[0027] The active party and the passive party initialize the error value sharding and data set sharding according to the error value between the predicted label and the true label and the data set they have;

[0028] The active party and the passive party calculate the shards used to calculate the model gradient cross terms according to the initialized shards and triplet shards, and calculate the parameters of the model gradient according to the shards used to calculate the model gradient cross terms;

[0029] The active and passive parties compute secret shards of the model gradient cross terms based on the parameters and triplet shards used to compute the model gradients.

[0030] Based on the same inventive concept, the second aspect of the present invention provides a multi-party longitudinal logistic regression privacy protection device based on secret sharing, which is applied to multiple participants, including active parties and passive parties. The multiple participants each have their own data sets. The facilitator is responsible for generating a large number of Beaver triples and distributing them to each participant before the start of the protocol. The multi-party longitudinal logistic regression privacy protection device includes:

[0031] Model weight sharing module, used to distribute model weights to various participants using secret sharing technology;

[0032] A forward propagation value cross term secret shard calculation module, which is used for each participant to collaboratively calculate the forward propagation value cross term according to the allocated model weight shard and triplet shard, and obtain the secret shard of the forward propagation value cross term, wherein the forward propagation value cross term is a part of the forward propagation value calculation;

[0033] The forward propagation value calculation module is used for each participant to obtain the output value according to their own data set, model weight shard and the secret shard of the forward propagation value cross term. The passive party sends the obtained output value to the active party, and the active party obtains the forward propagation value according to the output value of the passive party and its own output value. The forward propagation value can be used as the input of the top-level model to obtain the final prediction value of the model;

[0034] The model weight shard update module is used for the active party and the passive party to collaboratively calculate the model gradient cross term and obtain the secret shard of the model gradient cross term. Each participant locally updates the model weight shard according to the secret shard of the model gradient cross term, wherein the model gradient cross term is a part of the model gradient that needs to be collaboratively calculated by the active party and the passive party.

[0035] Based on the same inventive concept, the third aspect of the present invention provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the multi-party longitudinal logistic regression privacy protection method based on secret sharing described in the first aspect is implemented.

[0036] Based on the same inventive concept, the fourth aspect of the present invention provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the multi-party longitudinal logistic regression privacy protection method based on secret sharing described in the first aspect is implemented.

[0037] Compared with the prior art, the advantages and beneficial technical effects of the present invention are as follows:

[0038] The invention proposes a multi-party longitudinal logistic regression privacy protection method based on secret sharing. The assistant is responsible for generating a large number of Beaver triples and distributing them to each participant before the start of the protocol, and using secret sharing technology to distribute model weights to each participant; in the forward propagation process, the passive party collaboratively calculates the forward propagation value cross term according to the allocated model weight shards and triple shards, and obtains the secret shards of the forward propagation value cross term, then each participant obtains the output value according to the respective data sets, model weight shards and cross term shards, and the active party finally obtains the forward propagation value, during the reverse propagation, the active party and the passive party collaboratively calculate the model gradient cross term, and obtain the secret shards of the model gradient cross term, and each participant locally updates the model weight shard according to the secret shards of the model gradient cross term. Since the method of the invention generates a large number of Beaver triples and distributes them to each participant before the start of the protocol, the communication volume and the calculation volume can be transferred to before the start of the protocol, thereby greatly reducing the calculation volume and the communication volume in the collaborative calculation process, and the secret shards can be performed between multiple participants, so it can be extended to three or more participants. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] 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 or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 Flowchart of the privacy protection scheme for BlindFL longitudinal logistic regression;

[0041] Figure 2 A flowchart of a multi-party longitudinal logistic regression privacy protection method based on secret sharing provided by an embodiment of the present invention;

[0042] Figure 3 A structural diagram of a multi-party longitudinal logistic regression privacy protection device based on secret sharing provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0043] The inventor of this application has found through extensive research and practice that Figure 1 The vertical logistic regression privacy protection scheme described in the paper has the following two problems:

[0044] (1) It is not easy to expand to multiple parties;

[0045] (2) There are many communication rounds.

[0046] The present invention is dedicated to solving the above two problems. The proposed new scheme can be used for secure multi-party longitudinal logistic regression and achieve fewer communication rounds. Since the method of the present invention does not outsource the original data, it is suitable for high-dimensional and sparse data sets. At the same time, each party cannot master any underlying model, so it can achieve the same security guarantee as the above scheme.

[0047] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are 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.

[0048] Embodiment 1

[0049] The present invention discloses a multi-party longitudinal logistic regression privacy protection method based on secret sharing, which is applied to multiple participants, including active parties and passive parties. Multiple participants have their own data sets. The facilitator is responsible for generating a large number of Beaver triples and distributing them to each participant before the protocol starts. See Figure 2 , the method comprising:

[0050] S1: Use secret sharing technology to distribute model weights to each participant;

[0051] S2: Each participant collaboratively calculates the forward propagation value cross term based on the assigned model weight shards and triplet shards, and obtains the secret shard of the forward propagation value cross term, where the forward propagation value cross term is part of the forward propagation value calculation;

[0052] S3: Each participant obtains an output value based on their own data set, model weight shard, and secret shard of the cross term of the forward propagation value. The passive party sends the obtained output value to the active party. The active party obtains the forward propagation value based on the output value of the passive party and its own output value. The forward propagation value is used as the input of the top-level model to obtain the final prediction value of the model.

[0053] S4: The active party and the passive party collaboratively calculate the model gradient cross term and obtain the secret shard of the model gradient cross term. Each participant locally updates the model weight shard based on the secret shard of the model gradient cross term, where the model gradient cross term is the part of the model gradient that needs to be collaboratively calculated by the active party and the passive party.

[0054] Specifically, step S1 is the initialization step, in which the participants share the model weights through secret sharing technology. S2 and S3 belong to the forward propagation process. S2 calculates the forward propagation value cross terms through the collaboration of the participants and obtains the secret shards of the forward propagation value cross terms. S3 obtains the output value based on the data sets, model weight shards and forward propagation value cross terms owned by each participant, and the forward propagation value is calculated by the active party. S4 belongs to the back propagation process. The active party and the passive party collaboratively calculate the model gradient cross terms and obtain the secret shards of the model gradient cross terms. Each participant locally updates the model weight shards according to the secret shards of the model gradient cross terms.

[0055] Logistic regression is a classic algorithm in machine learning. Inspired by the paper, this paper proposes a multi-party longitudinal logistic regression privacy protection method based on secret sharing. This method can be easily extended to three parties or more, achieving lower communication rounds while ensuring security. Beaver triples are crucial as the basic components of this solution. The following first briefly describes the algorithm flow of Beaver triples, and then takes three participants as an example to describe the principle of the multi-party longitudinal logistic regression privacy protection algorithm based on secret sharing.

[0056] [*] indicates the state after the secret is shared, such as [a] indicates that secret a has been shared with the participants through the secret sharing function. For two-party secret sharing, two shard values ​​[a] are recorded. o and Right now

[0057] The Beaver triplet algorithm is described as follows:

[0058] Before the protocol starts, a random triple is generated in advance: [a], [b], [c]. c satisfies c = a·b. Assuming that secrets x and y have been shared, each participant now needs to calculate [x·y], which is the secret shard of the product of x and y.

[0059] (1) All participants calculate and make public [α], [α] = [x] - [a].

[0060] (2) All participants calculate and make public [β], [β] = [y] - [b].

[0061] (3) All participants calculate [z], where [z] o =[c]+α·[b]+β·[a],

[0062] Since [a], [b], [c] are pre-generated and distributed triplets, participants only need to locally calculate [α] = [x] - [a] and [β] = [y] - [b] during multiplication, and make the calculation results [α] and [β] public. All participants can independently calculate α and β through the secret reconstruction function. Beaver triples are consumable, and each multiplication consumes a Beaver triple. By using pre-calculated Beaver triples, the communication and computational workload can be shifted to before the start of the protocol.

[0063] It should be noted that [x·y] represents the secret shard of x·y. For simplicity, the result of the calculation of [x]-[a] is recorded as [α], [β], [z]. The triples mastered by each participant are the shards of a, b, c ([a], [b] and [c]). Because they have the relationship c=a·b, they are called triples. [z] represents the general secret sharing shard, [z] o , The subscripts indicate the shards held by each participant. For example, if z is shared between two parties and z=6, the shard [z] of the secret shared by participant 1 is recorded as [z] o = 1, the secret sharing shard [z] of z held by participant 2 is recorded as In addition, [a] o and The difference is that the two shards are controlled by different parties.

[0064] In the specific implementation process, the privacy protection method of multi-party longitudinal logistic regression based on secret sharing is described as follows:

[0065] Assume that there are three participants in the training of the model, and the participants are recorded as P1, P2, and P3, where P1 and P2 are passive parties and P3 is the active party. The data sets they have are X1, X2, and X3 respectively. The existing assistant P0 is responsible for generating a large number of Beaver triples and distributing them to the participants before the protocol starts. In plain text mode, the underlying model weights of each participant are recorded as W1, W2, and W3 respectively.

[0066] 1. Initialization

[0067] In order to ensure that each party cannot master any underlying model, that is, each party cannot have the plain text model weights W1, W2, and W3, the model weights are first distributed to each participant using SS (secret sharing) technology. In specific implementation, the model weights can be secretly shared through W = U + v, where the slices U and v of the model weights are held by different participants, so they do not know what W is. To achieve this goal, at initialization, P i Initialize U i and V i-1 , i={1, 2, 3}.

[0068] After initialization, the data held by each participant are as follows:

[0069] Participant P1: U1, V3, X1;

[0070] Participant P2: U2, V1, X2;

[0071] Participant P3: U3, V2, X3;

[0072] 2. Forward Propagation

[0073] Goal: Active party P3 grasps the value of Z. The value of Z is the forward activation value of the underlying model.

[0074] Z=X1W1+X2W2+X3W3=X1(U1+V1)+X2(U2+V2)+X3(U3+V3)

[0075] =X1U1+X1V1+X2U2+X2V2+X3U3+X3V3

[0076] Among them, X1U1, X2U2, and X3U3 can be calculated locally by P1, P2, and P3, while X1V1, X2V2, and X3V3 are cross-terms and require the mastery of X i and V i The two parties use Beaver triples to collaboratively compute, and the final result is that both parties can obtain the secret shard of the cross term.

[0077] Take the cross term X1V1 as an example. X1 is owned by P1. P1 initializes the shard of X1 as [X1]=X1. P2 initializes the shard of X1 as [X1]=0. V1 is owned by P2. P2 initializes the shard of V1 as [V1]=V1. P1 initializes the shard of V1 as [V1]=0. P0 has generated and assigned random triples to P1 and P2 before the protocol starts, denoted as [a], [b], [c]. P1 and P2 respectively calculate and publish [α]=[X1]-[a], [β]=[V1]-[b]. After P1 and P2 reconstruct α and β, they can calculate [X1V1] locally. Similarly, the shards of other cross terms can be calculated.

[0078] The specific calculation steps of the above process (participants P1 and P2 collaboratively calculate the forward propagation value cross term based on the assigned model weight shards and triplet shards, and obtain the secret shards of the forward propagation value cross term) are as follows:

[0079] (1) P1 initializes [X1]o = X1, P2 Initialization [V1] o =V1. They are model weight sharding and dataset sharding respectively)

[0080] (2) P1 calculates [α] o =[X1] o -[a] o , And open, P2 calculation And make it public.

[0081] (3) P1 and P2 are reconstructed (α, β are parameters used to calculate cross terms)

[0082] (4) P1 local computing [X1V1] o =[c] o +α·[b] o +β·[a] o P2 local computing Secret shard of the cross term of the forward propagation value.

[0083] The correctness proof is as follows:

[0084]

[0085] After the above local calculations, the data held by each participant are as follows:

[0086] Party P1: U1, V3, X1, X1U1, [X1V1] o ,

[0087] Participant P2: U2, V1, X2, X2U2, [X2V2] o ;

[0088] Participants P3: U3, V2, X3, X3U3, [X3V3] o ;

[0089] Each participant P i Calculate separately The result is recorded as Z i , then P1 and P2 send Z1 and Z2 to P3 respectively, and P3 locally calculates Z = Z1 + Z2 + Z3 = X1W1 + X2W2 + X3W3. Z is used as the input of the top-level model to calculate the final prediction value of the model. The top-level model is unique to P3 and does not require collaborative calculations by all participants.

[0090] The specific algorithm of the above process is as follows:

[0091] (1) P1 local computing P2 Local Computing P3 Local Computing

[0092] (2) P1 sends Z1 to P3, and P2 sends Z2 to P3.

[0093] (3) P3 calculates Z=Z1+Z2+Z3.

[0094] 3. Back Propagation

[0095] Target: P1 updates U1, V3, P2 updates U2, V1, P3 updates U3, V2.

[0096] For the participant P i , the model gradient in plain text form can be obtained by Calculation. Because P3 obtains the value of Z during the forward propagation, P3 can calculate locally in Represents the partial derivative of the loss function with respect to Z. Since Unique to P3, similar to the method of calculating cross terms during forward propagation, P1 and P3 need to collaboratively calculate cross terms and obtain its secret shard, P2 and P3 need to collaboratively calculate the cross term In addition, since P2 is the party that has the P1 model weight shard V1, P3 needs to Sent to P2. Error value sharding and data set sharding are equivalent to and X i The error value is equivalence.

[0097] The specific algorithm of the above process is as follows:

[0098] (1) P3 local computing

[0099] (2) P1 and P3 are calculated separately and and A secret shard of the model gradient cross-term computed for P1 and P3.

[0100] (3) P2 and P3 are calculated separately and and A secret shard of the model gradient cross-term computed for P2 and P3.

[0101] (4) P3 will Send to P2.

[0102] After the above local calculations and one round of communication, the data obtained by each participant during back propagation are as follows:

[0103] Participants P1: U1, V3, X1,

[0104] Participant P2: U2, V1, X2,

[0105] Participants P3: U3, V2, X3,

[0106] Assuming the learning rate is μ, each party locally updates the model weight shard:

[0107] V′3=V3; U′1 and V′3 are the model weight shards after P1 is updated;

[0108] U′2 and V′1 are the updated model weight shards of P2;

[0109] : U′3 and V′2 are the model weight shards after P3 update.

[0110] Taking W1 as an example, in plain text form, Update W1. The correctness of the above update result is proved as follows:

[0111]

[0112] Embodiment 2

[0113] Based on the same inventive concept, this embodiment discloses a multi-party longitudinal logistic regression privacy protection device based on secret sharing, which is applied to multiple participants, including active parties and passive parties. The multiple participants each have their own data sets. The facilitator is responsible for generating a large number of Beaver triples and distributing them to each participant before the protocol starts. See Figure 3 , the multi-party longitudinal logistic regression privacy protection device includes:

[0114] A model weight sharing module 301 is used to distribute model weights to each participant using secret sharing technology;

[0115] A forward propagation value cross term secret shard calculation module 302 is used for each participant to collaboratively calculate the forward propagation value cross term according to the allocated model weight shard and triplet shard, and obtain the secret shard of the forward propagation value cross term, wherein the forward propagation value cross term is a part of the forward propagation value calculation;

[0116] The forward propagation value calculation module 303 is used for each participant to obtain an output value according to the data set, model weight shard and secret shard of the cross term of the forward propagation value owned by each party, wherein the passive party sends the obtained output value to the active party, and the active party obtains the forward propagation value according to the output value of the passive party and its own output value, wherein the forward propagation value can be used as the input of the top-level model to obtain the final prediction value of the model;

[0117] The model weight shard update module 304 is used for the active party and the passive party to collaboratively calculate the model gradient cross term and obtain the secret shard of the model gradient cross term. Each participating party locally updates the model weight shard according to the secret shard of the model gradient cross term, wherein the model gradient cross term is a part of the model gradient that needs to be collaboratively calculated by the active party and the passive party.

[0118] Since the device introduced in the second embodiment of the present invention is a device used to implement the multi-party longitudinal logistic regression privacy protection method based on secret sharing in the first embodiment of the present invention, based on the method introduced in the first embodiment of the present invention, the person skilled in the art can understand the specific structure and deformation of the device, so it is not repeated here. All devices used in the method in the first embodiment of the present invention belong to the scope of protection of the present invention.

[0119] Embodiment 3

[0120] Based on the same inventive concept, the present invention further provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the method described in the first embodiment is implemented.

[0121] Since the computer-readable storage medium introduced in the third embodiment of the present invention is the computer-readable storage medium used to implement the multi-party longitudinal logistic regression privacy protection method based on secret sharing in the first embodiment of the present invention, based on the method introduced in the first embodiment of the present invention, those skilled in the art can understand the specific structure and deformation of the computer-readable storage medium, so it is not repeated here. All computer-readable storage media used in the method of the first embodiment of the present invention belong to the scope of protection of the present invention.

[0122] Embodiment 4

[0123] The present invention also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the method described in Embodiment 1 when executing the program.

[0124] Since the computer device introduced in the fourth embodiment of the present invention is a computer device used to implement the multi-party longitudinal logistic regression privacy protection method based on secret sharing in the first embodiment of the present invention, based on the method introduced in the first embodiment of the present invention, the person skilled in the art can understand the specific structure and deformation of the computer device, so it is not repeated here. All computer devices used in the method of the first embodiment of the present invention belong to the scope of protection of the present invention.

[0125] Those skilled in the art will appreciate that 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.

[0126] 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.

[0127] Although preferred embodiments of the present invention have been described, additional changes and modifications may be made to these embodiments by those skilled in the art once the basic creative concepts are known. Therefore, the appended claims are intended to be interpreted as including the preferred embodiments and all changes and modifications that fall within the scope of the present invention. Obviously, those skilled in the art may make various changes and modifications to the embodiments of the present invention without departing from the spirit and scope of the embodiments of the present invention. Thus, if these modifications and variations of the embodiments of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include these modifications and variations.

Claims

1. A privacy protection method for multi-party longitudinal logistic regression based on secret sharing, characterized in that: Applied to multiple participants, including active parties and passive parties, each of which has its own data set. The facilitator is responsible for generating a large number of Beaver triples and distributing them to each participant before the start of the protocol. The multi-party longitudinal logistic regression privacy protection method includes: Using secret sharing technology to distribute model weights to each participant; Each participant collaboratively calculates the forward propagation value cross term based on the assigned model weight shards and triplet shards, and obtains the secret shard of the forward propagation value cross term, where the forward propagation value cross term is part of the forward propagation value calculation; Each participant obtains the output value based on their own data set, model weight shard and secret shard of the cross term of the forward propagation value. The passive party sends the obtained output value to the active party. The active party obtains the forward propagation value based on the output value of the passive party and its own output value. The forward propagation value is used as the input of the top-level model to obtain the final prediction value of the model. The active party and the passive party collaboratively calculate the model gradient cross term and obtain the secret shard of the model gradient cross term. Each participant locally updates the model weight shard based on the secret shard of the model gradient cross term, where the model gradient cross term is the part of the model gradient that needs to be collaboratively calculated by the active party and the passive party.

2. The multi-party longitudinal logistic regression privacy protection method based on secret sharing as claimed in claim 1, characterized in that: The model weights are distributed to each participant using secret sharing technology, including: The model weight is distributed to each participant through W=U+V, and the model weight shards U and V are held by different participants, where W is the model weight.

3. The multi-party longitudinal logistic regression privacy protection method based on secret sharing as claimed in claim 1, characterized in that: Each participant collaboratively calculates the forward propagation value cross term based on the assigned model weight shards and triplet shards, and obtains the secret shards of the forward propagation value cross term, including: Each participant reconstructs the parameters for calculating the cross term based on the allocated model weight shards and triplet shards, and calculates the secret shard of the forward propagation value cross term based on the parameters for calculating the cross term and the triplet shards.

4. The multi-party longitudinal logistic regression privacy protection method based on secret sharing as claimed in claim 3, characterized in that: Each participant reconstructs the parameters for calculating the cross term based on the allocated model weight shards and triplet shards, and calculates the secret shards of the forward propagation value cross term based on the parameters for calculating the cross term and triplet shards, including: Each participant initializes model weight sharding and data set sharding according to the assigned model weights and owned data sets; Each participant calculates a shard for calculating the parameters of the cross term based on the initialized model weight shards, data set shards, and triplet shards, and calculates the parameters for calculating the cross term based on the shards for calculating the parameters of the cross term; Each participant calculates a secret shard of the forward propagation value cross term based on the parameters used to calculate the cross term and the triplet shard.

5. The multi-party longitudinal logistic regression privacy protection method based on secret sharing as claimed in claim 1, characterized in that: Each participant obtains the output value based on their own data set, model weight sharding and cross-term sharding, including: Each participant multiplies the data set they own by the model weight shard and adds it to the shard of the cross term to get the output value.

6. The multi-party longitudinal logistic regression privacy protection method based on secret sharing as claimed in claim 1, characterized in that: The active party obtains the forward propagation value based on the output value of the passive party and its own output value, including: The active party adds the received output value of the passive party to its own output value to obtain the forward propagation value.

7. The multi-party longitudinal logistic regression privacy protection method based on secret sharing as claimed in claim 1, characterized in that: The active party and the passive party collaboratively calculate the model gradient cross term and obtain the secret shard of the model gradient cross term, including: The active party and the passive party initialize the error value sharding and data set sharding according to the error value between the predicted label and the true label and the data set they have; The active party and the passive party calculate the shards used to calculate the model gradient cross terms according to the initialized shards and triplet shards, and calculate the parameters of the model gradient according to the shards used to calculate the model gradient cross terms; The active and passive parties compute secret shards of the model gradient cross terms based on the parameters and triplet shards used to compute the model gradients.

8. A multi-party longitudinal logistic regression privacy protection device based on secret sharing, characterized in that: Applied to multiple participants, including active parties and passive parties, each of which has its own data set. The facilitator is responsible for generating a large number of Beaver triples and distributing them to each participant before the start of the protocol. The multi-party longitudinal logistic regression privacy protection device includes: Model weight sharing module, used to distribute model weights to various participants using secret sharing technology; A forward propagation value cross term secret shard calculation module, which is used for each participant to collaboratively calculate the forward propagation value cross term according to the allocated model weight shard and triplet shard, and obtain the secret shard of the forward propagation value cross term, wherein the forward propagation value cross term is a part of the forward propagation value calculation; The forward propagation value calculation module is used for each participant to obtain the output value according to their own data set, model weight shard and the secret shard of the forward propagation value cross term. The passive party sends the obtained output value to the active party, and the active party obtains the forward propagation value according to the output value of the passive party and its own output value. The forward propagation value can be used as the input of the top-level model to obtain the final prediction value of the model; The model weight shard update module is used for the active party and the passive party to collaboratively calculate the model gradient cross term and obtain the secret shard of the model gradient cross term. Each participant locally updates the model weight shard according to the secret shard of the model gradient cross term, wherein the model gradient cross term is a part of the model gradient that needs to be collaboratively calculated by the active party and the passive party.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the multi-party longitudinal logistic regression privacy protection method based on secret sharing as described in any one of claims 1 to 7 is implemented.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, the multi-party longitudinal logistic regression privacy protection method based on secret sharing as described in any one of claims 1 to 7 is implemented.

Images