Secure aggregation method with cache layer against users for partial server exit

Abstract

The application discloses a security aggregation method with a cache layer for users to quit part of servers. The method includes the design of noise random variables on the server side of the cache layer, the encoding of user upload messages, the encoding of shared messages between servers, the decoding of server recovery of unreceived user upload messages, the encoding of communication messages between the simulation center and the servers, and the decoding of the simulation center for received messages. This method can be applied to each user separately for any no more than N-N r server dropouts, any no more than N-N r server dropouts for the simulation center, and any no more than T (T r ) number of server collusion, and the minimum communication rate of the first hop and the second hop.

Description

Technical Field

[0001] This invention relates to the field of information theory, and in particular to a method for dealing with K users, N cache layer servers, and each user dealing with any number of no more than N... r One server falls out, and no more than NN servers can be added. r For each server, the simulation center falls behind and any number of servers not exceeding T (T < N) r A secure aggregation method for multiple servers colluding. Background Technology

[0002] In secure aggregation problems, most work focuses on scenarios where users already know the target statistics of the simulation center and can share information with each other. However, in application scenarios where the target statistics change over time or the simulation center does not want users to know the target statistics, it is meaningful to study situations where users do not know the target statistics of the simulation center and cannot share information with each other. Furthermore, when considering the addition of a set of caching servers, it is also meaningful to study the situation where a user exits some servers. This invention studies a secure aggregation problem with a caching layer and proposes a secure aggregation method with a caching layer for situations where a user exits some servers. Summary of the Invention

[0003] Technical Problem: The technical problem to be solved by this invention is to provide a method for K users, N cache layer servers, where each user has any number of cache layers not exceeding N. r One server falls out, and no more than NN servers can be added. r For each server, the simulation center falls behind and any number of servers not exceeding T (T < N) r A secure aggregation method for multiple server collusion is proposed. This method utilizes the Vandermonde matrix to encode target statistics to obtain user-uploaded messages, including messages sent by the server to the simulation center and the decoding steps taken by the simulation center to obtain the statistics. Furthermore, to address the issue of users exiting some servers, a caching layer is designed to handle server-side noise random variables and shared messages between servers to recover user-uploaded messages not received by the servers. This results in a secure aggregation method with a caching layer for users exiting some servers.

[0004] Technical solution: The technical solution adopted in this invention is as follows:

[0005] (1) A trusted third party distributes the server-side noise random variable Z to the server n∈[N] in the cache layer. n .

[0006] (2) Each user k∈[K] has a private message W k Each message contains There are L symbols in the array, and each user k sends a message D to each server n∈[N]. k,n .

[0007] (3) Servers that have not fallen behind i∈N A Successfully received from user k∈K i The message requests information about user k∈[K]\K from other servers, specifically requests information about user k∈[K]\K. i The news.

[0008] (4) For a request from server i for a message about user k, if server n has received a message from user k, i.e., k∈K n Server n sends a shared message to server i in By Z n,i Divide the material into K equal parts and take the kth part to obtain the result, i.e.

[0009] (5) Server n recovers the user upload messages D that it did not receive from the received shared messages. k,n (k∈[K]\K i ).

[0010] (6) When the simulation center (SC) receives a statistical request and sends the coefficient vector f = (f1,...f... K When shared with all servers, server n sends a message to the simulation center based on the coefficient vector of the required statistical task.

[0011] (7) The simulation center, based on the received message Decode the target statistic

[0012] Wherein, the noise random variable Z at server n n The encoding method is as follows:

[0013]

[0014]

[0015] in S represents n The i-th row; Q n,i (n∈[N],i∈[N r -1]) is a trusted third party in The N(N) generated above r -1) row vectors of 1×KL encoding with iid, where each element in the vector is independent, identically distributed, and uniformly distributed. Additionally, the encoding matrix G... nG' n and S n Known by all entities in the system The above are different produce:

[0016]

[0017]

[0018] Upload message D k,n The encoding steps are as follows:

[0019] User k independently Generate a 1×LT / (N) r The row vector F of -T) k Each element in the vector is independent and identically distributed, and follows a uniform distribution. Then user k sends the private message W. k Evenly divided into N r The -T part represents the random variable F on the user side. k Divide the material evenly into T parts, that is:

[0020]

[0021] For each user k, the message D sent to server n k,n The encoding method is as follows:

[0022]

[0023] Where, α i , i∈[N r ] is from a finite field Different element values ​​and noisy random variable Z n The encoding uses consistency.

[0024] Recover unreceived user upload messages D k,n The decoding method is as follows:

[0025] When server n receives N r Decoding and calculation begin when there is a message about user k. This N r The message comes from the server set The messages received at this time are organized into a message matrix. Right now:

[0026]

[0027] The encoding matrix S n The corresponding in Part, namely the Rows are extracted to form a new matrix calculate And extract the first line of the result to get the user's uploaded message to be restored, that is:

[0028]

[0029] The decoding steps are as follows:

[0030] The simulation center received N r Decoding and calculation begin when a message is received:

[0031]

[0032] Where, N C ={n1,n2,…,n Nr} represents the received user set at this time, and the target statistic is obtained by merging and decoding the results.

[0033]

[0034] Beneficial effects: This invention proposes a general method for the security aggregation problem of users exiting some servers, applicable to K users, N caching layer servers, and each user exiting any number of servers not exceeding N. r One server falls out, and no more than NN servers can be added. r For each server, the simulation center falls behind and any number of servers not exceeding T (T < N) r This invention addresses the scenario of collusion among multiple servers. It ensures the correctness of the calculation results and the security of user privacy data, while minimizing the communication rate between the first and second hops. Compared to existing secure aggregation methods, this invention incorporates a cache layer consisting of a group of servers into the secure aggregation problem model. The new model still guarantees the correctness of the target statistic calculation and the security of user privacy data even when the target statistic changes over time or the simulation center does not want users to know the target statistic. Under the new model, the allocation of random noise variables on the cache layer server side and the communication messages between servers are rationally designed using the properties of the Vandermonde matrix to ensure that for each user, for any number of no more than N... r The accuracy of calculation results and the security of user privacy data in the event that a server fails to load. Attached Figure Description

[0035] Figure 1 This is a system model with K users and N cache layer servers.

[0036] Figure 2 This is a system model with 2 users, 4 cache layer servers, and N1 = {1, 2, 4} and N2 = {1, 2, 3}. Detailed Implementation

[0037] The technical solution of the present invention will be described in detail below, but the scope of protection of the present invention is not limited to the embodiments described.

[0038] This invention proposes a secure aggregation method with a caching layer for users exiting certain servers. For example... Figure 1 As shown, the system model of this invention includes K users and N cache layer servers. The system model considers K users and N servers, where each user successfully uploads messages to at least N servers. r There are N servers, and at least N servers are involved in the entire communication process. r One server remains responsive, and at most T servers will collude; each user k has a private message W. k and random noise F used for encoding k It uploads message D to server n. k,n Each server n has random noise Z used for encoding. n Servers can send messages to each other; the simulation center (SC) sends a coefficient vector f of statistical requests to the servers, and server n sends a message to the simulation center based on f. Used to decode the target statistic W.

[0039] The following is an example:

[0040] like Figure 2 As shown, this embodiment considers K = 2 users and N = 4 servers, where each user successfully uploads a message to at least N. r = 3 servers, at least N during the entire communication process r = 3 servers maintain responses, and no server colludes (T=1). Assume the first-hop communication is: N1={1,2,4}, N2={1,2,3}, the user message length is 2, the target statistic is W1+W2, and the domain is finite. The size q = 8, and the specific plan is as follows:

[0041] (1) A trusted third party distributes the server-side noise random variable Z to the server n∈[N] in the cache layer. n :

[0042]

[0043]

[0044]

[0045] Among them, Q n,i (n∈[4], i∈[2]) is a trusted third party in The generated 8 row vectors of length 2 for each iid.

[0046] (2) Each user k∈[K] has a private message W k Each message contains There are L symbols in the array, and each user k sends a message D to each server n∈[N]. k,n :

[0047]

[0048] Among them, F1 and F2 are user-independent. The generated row vector of length 1, W k (1), W k (2) is from W k It was obtained by dividing it evenly into two parts.

[0049] (3) For the first hop transmission case N1={1,2,4}, N2={1,2,3}, server 3 requests information about user 1 from other servers, and server 4 requests information about user 2 from other servers.

[0050] (4) Server i∈{1,2,4} sends a shared message to server 3. Server j∈{1,2,3} sends a shared message to server 4 Among them (Z) n,i (1),Z n,i (2))=Z n,i .

[0051] (5) Server 3 from Recovering the D that was not received 1,3 Server 4 from Recovering the D that was not received 2,4 :

[0052]

[0053] (6) When the simulation center (SC) receives a statistical request and shares the coefficient vector f = (1,1) with all servers, server n sends a message to the simulation center according to the coefficient vector of the required statistical task.

[0054] (7) The simulation center, based on the received message The target statistic W = W1 + W2 is decoded. The simulation center receives N. r Decoding and calculation begin when 3 messages have been received. If the received messages are from servers 1, 2, and 3, then the calculation proceeds as follows:

[0055]

[0056] The result is W = W1 + W2. Other cases follow a similar decoding method.

Claims

1. A secure aggregation method with a caching layer for user logouts of partial servers, characterized in that, Includes the following steps: (1) A trusted third party distributes the server-side noise random variable Z to the server n∈[N] in the cache layer. n ; (2) Each user k∈[K] has a private message W k Each message contains a finite field There are L symbols in the array, and each user k sends a message D to each server n∈[N]. k,n ; (3) Servers that have not fallen behind i∈N A Successfully received from user k∈N i The message requests information about user k∈[K]\K from other servers, specifically requests information about user k∈[K]\K. i The message; (4) For a request from server i for a message about user k, if server n has received a message from user k, i.e., k∈K n Server n sends a shared message to server i in By Z n,i Divide the material into K equal parts and take the kth part to obtain the result, i.e. (5) Server n recovers the user upload messages D that it did not receive from the received shared messages. k,n (k∈[K]\K i ); (6) When the simulation center SC receives a statistical request and sends the coefficient vector f = (f1, ... f2) to the center, the simulation center SC will send the coefficient vector to the center. K When shared with all servers, server n sends a message to the simulation center based on the coefficient vector of the required statistical task. (7) The simulation center, based on the received message N B This represents the set of servers that successfully sent messages to the simulation center, and the target statistics are decoded.

2. The secure aggregation method with a caching layer for user exits of partial servers as described in claim 1, characterized in that, In step (1), the noise random variable Z at server n n The encoding method is as follows: in Represents the encoding matrix S n The i-th row; Q n,i (n∈[N],i∈[N r -1]) is a trusted third party in The N(N) generated above r -1) row vectors of 1×KL with iid, N r This represents the number of servers where each user successfully uploaded a message. Each element in the vector is independent, identically distributed, and follows a uniform distribution; the encoding matrix G... n G' n and S n Known by all entities in the system The above are different Generate, α n It comes from Different element values:

3. The secure aggregation method with a caching layer for user exits of partial servers as described in claim 1, characterized in that, In step (2), message D is uploaded. k,n The encoding steps are as follows: User k independently Generate a 1×LT / (N) r The row vector F of -T) k Each element in the vector is independent and identically distributed and follows a uniform distribution; then user k sends the private message W k Evenly divided into N r The -T part represents the random variable F on the user side. k Divide the material evenly into T parts, that is: For each user k, the message D sent to server n k,n The encoding method is as follows: Where, α i , i∈[N r ] is from a finite field Different element values ​​and noisy random variable Z n The encoding uses consistency.

4. The secure aggregation method with a caching layer for user exits of partial servers as described in claim 1, characterized in that, In step (5), the unreceived user upload message D is recovered. k,n The decoding method is as follows: When server n receives N r Decoding and computation begin when there are N messages about user k; r The message comes from the server set The messages received at this time are organized into a message matrix. Right now: The encoding matrix S n The corresponding in Part, namely the Rows are extracted to form a new matrix calculate And extract the first line of the result to get the user's uploaded message to be restored, that is:

5. The secure aggregation method with a caching layer for user exits of partial servers as described in claim 1, characterized in that, In step (7) The decoding steps are as follows: The simulation center received N r Decoding and calculation begin when a message is received: Where, N C ={n1,n2,…,n Nr } represents the received user set at this time, and the target statistic is obtained by merging and decoding the results.

Images