28-07-2023 дата публикации
Номер: CN116502728A
Принадлежит:
The invention discloses a dynamic aggregation method and device based on federated learning, and relates to the technical field of federated learning. The system comprises at least one terminal device, a central server and a connection terminal, the terminal device is used for receiving a global model, preventing over-fitting through regularization training and uploading a local model, the central server is connected with the at least one terminal device, and the connection terminal is connected with the central server. And the server is used for receiving the local models uploaded by the terminals, calculating aggregation weights of the terminals, dynamically aggregating the local models according to the weights to generate a global model of the round, and finally issuing the global model to the connection terminal. The method has the beneficial effects that on one hand, the dynamic aggregation weight of each client is calculated at the central server side, so that the problem of non-independent ...
Подробнее