Federated Learning Workbench

2023, September 26th:The workbench is out. This represents the beginning, and it is difficult to say that this is federated learning as the protocol is entirely cut down to ensure release over perfection.

As is aparent I decided to make this using TypeScript. Why not a Python notebook to explorer these techniques? Well, I might very well add that onto it (Well, probably Livebook as I am more into the Elixir ecosystem, but time will tell). First and foremost this is an investigation into practical federated learning. I want to ensure that I handle resource constraints of browsers, complexities of sending weights and models over the network. Error cases, etc.

Currently the most grave error is that all clients train on the same dataset. For the next update, I am going to split the training set by label, so each client act i accordance with the description. Hopefully, this will uncover a worse performaning model for which I have to implement techniques to evaluate and develop the model.