Federated Learning Framework via Distributed Mutual Learning

--Federated Learning has gained a significant amount of attraction in recent years. Federated learning enables clients to learn about their private data and then share their learnings with the central server to create a generalized global model and further share the generalized model with all clients. This aggregation of knowledge is based on the aggregation of model weights which has many associated issues. The model weights are more susceptible to model inversion attacks and would use a significant amount of bandwidth to share. In this work, we propose a loss-based federated learning framework using deep mutual learning between all clients using knowledge distillation.