Pysyft: Federated learning with plenty of clients?

[MetaT1an]

Is PySyft suitable for large-scaled federated learning scenario where hundreds or thounds of clients are participated?

In the tutorial document, I only see clients such as Alice and Bob, which seems to be far away form the actual cross-device learning tasks.

Many thanks.

[NeuZhangQiang]

@MetaT1an Yes, PySyft can be used for large-scaled federated learning scenario. There is a tutorial that is helpful for your question. However, for some unclear reason, the pysyft 0.2.6 do not support the above tutorial.

In addition, I can not find the similar tutorial for keras or tensorflow.

[MetaT1an]

@NeuZhangQiang Thanks for your kind reply. Suppose I need 100 clients for a certain experiment, is it a good idea to create these clients with consecutive numbers(1~100) being their id property , and to put them into a list for future training?

[NeuZhangQiang]

Yes, you can. You only need to make sure the id is unique.