Privacy-Preserving Federated Learning Framework with General Aggregation and Multiparty Entity Matching
Algorithm 1
MEMP.
Input: a central server CS, a set of participants , and a trusted party .
Output: common entity IDs.
1: CS sets the parameters for model training .
2: generates a public-private key for homomorphic encryption, a public-private key for RSA encryption for CS, also generates average participants’ public-private keys and subshares of the public key and based on the identity of the participants, i.e., . Here, get subshares for and subshares for .
3: Each participant chooses a random number , computes , and operates .
4: CS chooses a random number , uses the private key for signature , gets each , and returns them to after disturbing the order of .
5: fordo
6: fordo
7: CS computes for the entities: .
8: CS sends to corresponding participants .
9: fordo
10: Each participant eliminates the blind factor and for , obtains , and computes their hash values .
11: fordo
12: Each participant generates its own -dimensional matrix by determining whether belongs to its .
13: ifthen
14: .
15: else
16: , where is a random number and .
17: Each participant chooses a set of random number and encrypts its matrix with , operating .
18: CS aggregates matrix values by computing and obtains by decrypting.
