US 11,947,680 B2
Model parameter training method, terminal, and system based on federation learning, and medium
Yang Liu, Shenzhen (CN); Yan Kang, Shenzhen (CN); Tianjian Chen, Shenzhen (CN); Qiang Yang, Shenzhen (CN); and Tao Fan, Shenzhen (CN)
Assigned to WEBANK CO., LTD, Shenzhen (CN)
Filed by WEBANK CO., LTD, Shenzhen (CN)
Filed on Apr. 25, 2021, as Appl. No. 17/239,623.
Application 17/239,623 is a continuation of application No. PCT/CN2019/119226, filed on Nov. 18, 2019.
Claims priority of application No. 201811620130.X (CN), filed on Dec. 28, 2018.
Prior Publication US 2021/0248244 A1, Aug. 12, 2021
Int. Cl. G06F 21/60 (2013.01); G06F 21/62 (2013.01); G06N 5/04 (2023.01); G06N 20/00 (2019.01); G06N 20/20 (2019.01)
CPC G06F 21/602 (2013.01) [G06F 21/6218 (2013.01); G06N 5/04 (2013.01); G06N 20/00 (2019.01); G06N 20/20 (2019.01)] 11 Claims
OG exemplary drawing
 
1. A model parameter training method based on federation learning, applied to a first terminal, comprising the following operations:
determining a feature intersection of a first sample of the first terminal and a second sample of a second terminal, training the first sample based on the feature intersection to obtain a first mapping model, encrypting and sending the first mapping model to the second terminal, for the second terminal to predict a missing feature of the second sample to obtain a second encryption supplementary sample;
receiving a second encryption mapping model sent by the second terminal, predicting a missing feature of the first sample of the first terminal according to the second encryption mapping model to obtain a first encryption supplementary sample, wherein the second encryption mapping model is obtained by the second terminal training the second sample based on the feature intersection;
receiving a first encryption federation learning model parameter sent by a third terminal, training a federation learning model to be trained according to the first encryption federation learning model parameter, the first sample and the first encryption supplementary sample, and calculating a first encryption loss value;
sending the first encryption loss value to the third terminal, for the third terminal to calculate a loss sum according to the first encryption loss value and a second encryption loss value, determining whether the federation learning model to be trained is in a convergent state according to the loss sum, wherein the second encryption loss value is calculated by the second terminal according to the second sample, the second encryption supplementary sample, and the first encryption federation learning model parameter sent by the third terminal; and
using the first encryption federation learning model parameter as a final parameter of the federation learning model to be trained after receiving a stop training instruction sent by the third terminal, wherein the stop training instruction is sent by the third terminal after determining that the federation learning model to be trained is in the convergent state.