Layerwise Cosine Aggregation Schema

This repository includes the source code for the paper "Improving alpha-f Byzantine Resilience in Federated Learning via layerwise aggregation and cosine distance".

Find preprint here.

Installation

This project uses Python dependencies. To install them using uv, run:

uv sync

Now, you can execute the code with

uv run layerwise_krum.py

Run the following command for getting more information about how to configure the experiments:

uv run layerwise_krum.py --help

Note: The code requires you to have a CUDA device. Feel free to change the assert at the top of the file (after imports).