Authors: Yuxin Shi, Han Yu

Published on: January 05, 2024

Impact Score: 8.22

Arxiv code: Arxiv:2401.02740

Summary

• What is new: The introduction of a Fairness-aware Federated Job Scheduling (FairFedJS) approach in Federated Learning, focusing on fair allocation of FL client datasets among multiple FL servers.
• Why this is important: Existing FL research does not sufficiently address scenarios with multiple FL servers competing for clients, leading to unfair allocation and prolonged waiting times.
• What the research proposes: The proposed FairFedJS uses Lyapunov optimization to ensure fair allocation of client datasets to various FL jobs, considering both demand and job payment bids.
• Results: FairFedJS outperforms four state-of-the-art approaches in scheduling fairness by 31.9% and in convergence time by 1.0%, with comparable test accuracy.

Technical Details

Technological frameworks used: Lyapunov optimization

Models used: Federated Learning (FL)

Data used: Two datasets

Potential Impact

Data-intensive sectors such as healthcare, finance, and personalized advertising could benefit from FairFedJS by improving efficiency and fairness in data usage.

Want to implement this idea in a business?

We have generated a startup concept here: EquiLearn.