Implementations

Contents

1. Clone repo into work directory of Docker
2. OCD-FL: A Novel Communication-Efficient Peer Selection-based Decentralized Federated Learning

Clone repo into work directory of Docker

Once done with the base setup from the previous segment, we can now proceed proper. Actually not quite. We first need to get into Jupyter notebook environment on the browser. This will be our main workplace for future implementations.

Upon running the docker container, simply go to your browser and access the Jupyter environment. (The ' is intentional don't worry)

http://<JETSON_USERNAME>.local:8888

You should be able to access the Jupyter interface if your browser does not blocks pop-ups. Once done, clone the repository to the current directory you are in. At the top right corner of the Jupyter landing page, click on the dropdown labelled as "New", then click on "Terminal". You should have a new window that pops out with the terminal. Clone this repository over there.

git clone https://github.com/ngzhankang/sutdDFL.git

Now you are ready to do work! Update the repo using the terminal in the new windows when you're done for the day of work.

OCD-FL: A Novel Communication-Efficient Peer Selection-based Decentralized Federated Learning

This is a cloned implementation of the above repo. Simulators have been stripped off in order to translate the project into actual communications between Jetson nodes. The entire project can be found in SUTD:/sutdDFL/ocdFL.

More info about implementation can be found at the subfolder directory here.