Industry model joint training method in federal collaborative learning mode

• Main Authors: GUAN DEKANG, ZHANG JIANNAN, ZHAO JIA, JI HAIPENG, DONG YONGFENG, LIU JING
• Format: Patent
• Language: Chinese; English
• Published: 14.06.2024
• Subjects: CALCULATING; COMPUTER SYSTEMS BASED ON SPECIFIC COMPUTATIONAL MODELS; COMPUTING; COUNTING; DATA PROCESSING SYSTEMS OR METHODS, SPECIALLY ADAPTED FORADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORYOR FORECASTING PURPOSES; ELECTRIC DIGITAL DATA PROCESSING; PHYSICS; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE,COMMERCIAL, FINANCIAL, MANAGERIAL, SUPERVISORY OR FORECASTINGPURPOSES, NOT OTHERWISE PROVIDED FOR

The invention discloses an industry model joint training method in a federal collaborative learning mode. The method comprises the following steps: S1, building a multi-stage federal collaborative architecture; s2, all factory sub-terminals collect production data; s3, the factory sub-terminal generates a local model through the convolutional neural network; s4, part of the factory sub-terminals upload the model parameters to the group terminal; s5, the group end performs selection and weighted aggregation on local model parameters through a group contribution degree adaptive weighted model, and selects a factory sub-end which uploads the local model in the next round; s6, the group end issues a group model to all the factory sub-ends; s7, the group end is combined with the factory sub-end to upload the group model and the group contribution-income information to the industry end; s8, the industry end evaluates and aggregates the group model through an industry competition income game model, and calculates a