Modifying federated learning

[Nicholas-B1]

Hello, I would like to know how I would go about cutting down datasets or features or CNN layers to make faster running federated learning. I want to be able to test the speed of these examples like the cifar10 example but make it run faster even if it means sacrificing some accuracy.

[YuanTingHsieh]

This is a general research question, @holgerroth @ZiyueXu77 can you help sharing your thoughts on this discussion, thanks!

[holgerroth]

Hi @Nicholas-B1 , you can reduce the number of layers or use the SimpleCNN in the network definitions. Also, changing the number of training samples being loaded in the data loader should speed things up. The easiest would be to break the loop after a certain number of iterations in training and validation.