Repository for code used in the paper "Real-Time Decentralized knowledge Transfer at the Edge"

In this repo you can find the experiments discussed in the paper, in addtion to methods compared with.

The "compared_with" folder contains the 3 methods used as a comparison in the paper: Knowledge Distillaiton, Federated Learning, and Gossip Algorithms. The folders contining code for Federated Learning and Gossip Algorithms retains the original readme and copyright as was cloned from their origin repositories. The experiments we ran using these works can be found in the jupyter notebook in each of these folders.

The different network compositions constructed in our experiments correspond to the 3 notebooks: cifar10_multi_agent.ipynb, cifar10_multi_agent_full_mesh.ipynb, cifar10_multi_agent-transitive.ipynb.