I have the following question: during the local training phase, the model parameters on the client side are not updated with the parameters from the server side, they are only stored in self.W and not loaded into self.model.

[nanqiaobei]

Hello, we have observed that in the code within the client's download_from_server function, the server parameters are only saved in self.W, and the self.model parameters are not updated. This means that during the local training phase, the model's parameters only use the locally trained parameters and do not incorporate the server parameters. Consequently, during the testing phase, the model only uses the parameters from the last local training iteration and does not utilize the exchanged parameters for evaluation.

I'm not sure if I might have missed something, so I'd like to ask for advice.

[LDer66]

Hello, we have observed that in the code within the client's download_from_server function, the server parameters are only saved in self.W, and the self.model parameters are not updated. This means that during the local training phase, the model's parameters only use the locally trained parameters and do not incorporate the server parameters. Consequently, during the testing phase, the model only uses the parameters from the last local training iteration and does not utilize the exchanged parameters for evaluation.

I'm not sure if I might have missed something, so I'd like to ask for advice. 这个代码时期是在服务器端更新了编码器，分类器是没有聚合的，你可以看看模型定义那里，分类器相当于每个客户端是个性化的