Authors: Jialuo He, Wei Chen, Xiaojin Zhang

Published on: February 08, 2024

Impact Score: 8.15

Arxiv code: Arxiv:2402.05541

Summary

• What is new: The introduction of Reinforcement Federated Learning (RFL) which employs deep reinforcement learning to optimize client contributions during data aggregation, aiming to combat adversarial attacks and uneven data distribution.
• Why this is important: Current federated learning methods do not adequately address statistical heterogeneity and are vulnerable to adversarial attacks, affecting model robustness and fairness.
• What the research proposes: RFL optimizes client contributions through Deep Deterministic Policy Gradient-based algorithm, a novel client selection method, and a reward mechanism evaluating validation set performance, to improve model robustness and fairness.
• Results: RFL surpasses existing methods in robustness against malicious attacks while maintaining fairness across participants in tests with non-identically distributed settings.

Technical Details

Technological frameworks used: Reinforcement Federated Learning (RFL)

Models used: Deep Deterministic Policy Gradient (DDPG)

Data used: Data from decentralized devices/systems in non-identically distributed settings

Potential Impact

This research could impact markets involving consumer data privacy, such as healthcare, finance, and personalized advertising, by offering a more robust and fair federated learning framework.

Want to implement this idea in a business?

We have generated a startup concept here: FairFleet.