DISTRIBUTED LEARNING MACHINE LEARNING MODELS FOR PERSONALIZATION

摘要

FIELD: data processing.;SUBSTANCE: invention relates to distributed training of machine learning (ML) model of artificial intelligence (AI). Method comprises steps of: initialising one or more ML models on a server, distributing one or more ML models among one or more user devices (UE) connected to a server, storing user-generated data, transmitting training data from the server to one or more UEs, training the ML model based on the collected data and training data, acquiring trained ML models on the server, ML model is updated on the server by averaging the trained ML models obtained from one or more user devices, and transmitting the updated ML models to one or more UEs. System realizes the described method using a machine-readable medium.;EFFECT: technical result consists in improvement of quality of training personalized ML models with prevention of their retraining and reduced costs for data transmission over network connections.;13 cl, 3 dwg