US 12,437,104 B2
Federated learning method using synonym
Chih-Fan Hsu, Taipei (TW); Wei-Chao Chen, Taipei (TW); and Ming-Ching Chang, Taipei (TW)
Assigned to Inventec (Pudong) Technology Corporation, Shanghai (CN); and INVENTEC CORPORATION, Taipei (TW)
Filed by Inventec (Pudong) Technology Corporation, Shanghai (CN); and INVENTEC CORPORATION, Taipei (TW)
Filed on Mar. 21, 2022, as Appl. No. 17/699,263.
Claims priority of application No. 202210225728.9 (CN), filed on Mar. 9, 2022.
Prior Publication US 2023/0289468 A1, Sep. 14, 2023
Int. Cl. G06N 20/00 (2019.01); G06F 18/214 (2023.01); G06F 21/62 (2013.01); G06F 40/247 (2020.01)
CPC G06F 21/6245 (2013.01) [G06F 18/2148 (2023.01); G06F 40/247 (2020.01)] 10 Claims
OG exemplary drawing
 
1. A federated learning method using synonym, comprising:
sending a general model to each of a plurality of client devices by a moderator;
performing a training procedure by each of the plurality of client devices, wherein the training procedure comprises:
removing a private portion from original private data to obtain processed private data, and encoding processed private data into a digest by an encoder;
training a client model according to the processed private data, the digest and the general model; and
sending the digest and a client parameter of the client model to the moderator, wherein the client parameter is associated with a weight of the client model;
determining an absent client device among the plurality of client devices by the moderator;
generating a synonym of the digest corresponding to the absent client device by a synonym generator;
training a replacement model according to the synonym and the digest corresponding to the absent client device by the moderator; and
performing an aggregation to generate an updated parameter to update the general model by the moderator according to a replacement parameter of the replacement model and the client parameter of each of the plurality of client devices except the absent client device.