HEAL: A New Decentralized Learning Framework Combining FL, Gossip, and Epidemic Learning
A research paper introduces HEAL, a novel decentralized learning framework that integrates Federated Learning, Gossip Learning, and Epidemic Learning. HEAL is the first cross-layer framework to exploit an optimized self-organizing and self-healing P2P overlay, leveraging the Elevator algorithm to dynamically promote nodes as hubs. This approach aims to overcome the limitations of existing decentralized learning methods, such as server vulnerabilities, scalability issues, privacy risks, and slow convergence. The paper is available on arXiv under identifier 2605.27475.
Key facts
- HEAL stands for Resilient and Self-* Hub-based Learning.
- It is a decentralized learning framework.
- It combines Federated Learning, Gossip Learning, and Epidemic Learning.
- It uses an optimized self-organizing and self-healing P2P overlay.
- It leverages the Elevator algorithm to dynamically promote nodes as hubs.
- The paper is published on arXiv with ID 2605.27475.
- The announcement type is cross.
- The framework addresses server vulnerabilities, scalability, privacy, and single point of failure.
Entities
Institutions
- arXiv