Thank you for your article and code, which gave me a lot of help. I encountered difficulties during the experiment. I would like to ask if you can share with me the code to reproduce the previous defense method in the paper experiment, because my code I have limited abilities and can't do it myself, if you can help me, I will be very grateful!!
In the main.py file, lines 167–194, what is the purpose of the phrase "logger.inflogger.info("end test the returned updates").
What I found in this code is that their local model doesn't change because each agent_name_key only contains one element.
Should we remove these lines of code, or is this my misunderstanding?
Anw, I thought FedAvg aggregation models would be based on one number of samples trained in each client, but in your code, the weight is equal for all clients, = self.params["lr_para"] / self.params["no_models"]. Could you explain this?