TRACE: A trust-aware incentive mechanism for federated learning in IoMT
Abstract Federated Learning (FL) has emerged as a promising privacy-preserving framework for the Internet of Medical Things (IoMT), yet issues like client selfishness and data heterogeneity impair model generalization, resulting in suboptimal and potentially unsafe medical diagnostics. Existing ince...
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Springer
2025-08-01
|
| Series: | Journal of King Saud University: Computer and Information Sciences |
| Subjects: | |
| Online Access: | https://doi.org/10.1007/s44443-025-00172-6 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Abstract Federated Learning (FL) has emerged as a promising privacy-preserving framework for the Internet of Medical Things (IoMT), yet issues like client selfishness and data heterogeneity impair model generalization, resulting in suboptimal and potentially unsafe medical diagnostics. Existing incentive mechanisms focus on motivating participation but ignore user honesty, leading to free-riding behaviors that degrade the model quality. To overcome these challenges, we propose TRACE, a truthful incentive mechanism designed for hierarchical FL in IoMT. TRACE not only encourages client participation but also guarantees truthful behavior by dynamically rewarding verifiable contributions. Unlike conventional approaches that rely on passive user selection or self-reported bids, TRACE proactively counteracts selfish behavior and data heterogeneity through a synergistic combination of coalition formation and contract theory. Specifically, TRACE models the hierarchical FL framework as a Stackelberg game: the upper layer organizes stable user coalitions leveraging social affinity, while the lower layer designs type-specific contracts to elicit high-quality data contributions. By jointly addressing strategic dishonesty and data heterogeneity, this dual-layer strategy ensures reliable model performance. Comprehensive experiments on both synthetic and real datasets demonstrate that TRACE consistently outperforms seven state-of-the-art baselines, achieving a 20% improvement in server utility, 60% reduction in training loss, while maintaining superior test accuracy even under adversarial conditions with dishonest users. |
|---|---|
| ISSN: | 1319-1578 2213-1248 |