Föderiertes Lernsystem

Details

Veröffentlichungs-Nr.

EP4760602

Anmeldetag

12. Dezember 2024

Veröffentlichung

17. Juni 2026

Rechtsraum

EP

IPC

G06N3/098

Abstract

The invention relates to a federated learning system (1) comprising client devices (10) and a server (20), wherein each client device (10) comprises a training unit configured to train a machine learning model received from the at least one server (20) on a loaded local dataset during an iteration to generate trained model weights and a managing unit configured to control the training unit to train the received model for a predetermined number of iterations, each server (20) comprises a managing unit configured to identify each client device (10), to determine the number of training iterations realized by all client devices (10) and to send a machine learning model to each client device (10), and an aggregator configured to aggregate the training model weights generated after a predetermined number of iterations of all the client devices (10) to generate an aggregated model.

Anmelder

Firma

BULL SAS

Land

🇫🇷 Frankreich

Fachgebiete