Performance of Federated Learning under Realistic Communication Conditions

In recent years, federated learning has emerged as a promising approach for training machine learning models on distributed data sources. However, the performance of federated learning heavily relies on the quality and efficiency of communication between the devices participating in the learning process.

This project aims to investigate the performance of federated learning under realistic communication conditions. By simulating various communication scenarios, including different network latencies, bandwidth limitations, and packet losses, we can assess the impact of these factors on the overall learning performance.

The objectives of this project include:

1. Evaluating the impact of network latency on federated learning performance.
2. Analyzing the effect of limited bandwidth on model convergence and accuracy.
3. Investigating the influence of packet losses on the reliability of federated learning.
4. Developing strategies and techniques to mitigate the negative effects of communication conditions on federated learning.

To achieve these objectives, students will design and implement a federated learning framework that can simulate realistic communication conditions. You will use state-of-the-art machine learning algorithms and datasets to conduct experiments and measure the performance metrics.

Tools to be used:

Flower: https://flower.dev

CORE: https://www.nrl.navy.mil/Our-Work/Areas-of-Research/Information-Technology/NCS/CORE/