Federated learning is a privacy-preserving, distributed approach to machine learning. Currently, federated learning research is mostly done in simulated environments. As a result, the effects of real-world network congestion and instabilities are not realized during testing. The first goal of our project is to create a federated learning infrastructure that can be deployed to a real network for experimentation purposes. Next, we would like to help researchers address the following problem: can we improve the convergence rate of a federated learning model through in-network aggregation? Integrating the Chameleon Cloud network with our infrastructure, our team created an interactive web app for users to run their own federated learning experiments and explore various routing algorithms that leverage intermediate aggregation to optimize the speed of training in a network. These algorithms should help to mitigate the effects of congestion to decrease training time.

Team Members

Michael Chuang
Jeffrey Liu
Jeesoo Min
Sidharth Nair
Adeel Rehman