CPC H04L 9/0819 (2013.01) [G06N 20/20 (2019.01); H04L 9/008 (2013.01); H04L 9/0869 (2013.01); H04L 9/14 (2013.01); H04L 9/30 (2013.01)] | 20 Claims |
1. A method for vertical federated learning, applied to an upper-layer participation node in multiple participation nodes deployed in a multi-way tree topology, the upper-layer participation node corresponding to k lower-layer participation nodes, a submodel in a federated model being locally deployed in each participation node, k being an integer greater than 1, the method comprising:
distributing a first public key corresponding to the upper-layer participation node to the k lower-layer participation nodes, and acquiring k second public keys corresponding to the k lower-layer participation nodes respectively;
performing, by the upper-layer participation node, secure two-party joint computation with the k lower-layer participation nodes respectively with the first public key and the k second public keys as encryption parameters, to obtain k two-party joint outputs of the federated model, the secure two-party joint computation comprising forward computation for the submodel performed, in a manner of homomorphic encryption, by the upper-layer participation node and a lower-layer participation node jointly using respective data of the upper-layer participation node and the lower-layer participation node; and
aggregating the k two-party joint outputs to obtain a multi-party joint output corresponding to the upper-layer participation node and the k lower-layer participation nodes;
wherein the upper-layer participation node is a lower-layer participation node of a higher-layer participation node in the multi-way tree topology, and the method further comprises:
acquiring a third public key corresponding to the higher-layer participation node; and
performing, by the upper-layer participation node as a lower-layer participation node, secure two-party joint computation with the higher-layer participation node with the first public key and the third public key as encryption parameters, to obtain a two-party joint output of the federated model.
|