Federal learning method, device and system

The invention provides a federated learning method, device and system, which are used for retraining a received model by a server in a federated learning process to realize depersonalization processing to a certain degree so as to obtain a model with higher output precision. The method comprises the steps that firstly, a first server receives information of at least one first model sent by at least one downstream device, and the at least one downstream device can comprise other servers or clients connected to the first server; the first server trains the at least one first model to obtain at least one trained first model; and then, the first server aggregates the trained at least one first model, and updates a locally stored second model by using an aggregation result to obtain an updated second model.