VARS-FL: Validation-Aligned Client Selection for Non-IID Federated Learning in IoT
A new federated learning framework, VARS-FL (Validation-Aligned Reputation Scoring for Federated Learning), addresses the challenge of client selection under non-IID data distributions in IoT and IIoT environments. Traditional stateless selection methods ignore accumulated evidence of client contribution quality, leading to slow convergence and unstable training. VARS-FL quantifies each client's contribution by measuring the reduction in server-side validation loss induced by its update, then aggregates these per-round signals into a Reputation score using a sliding-window mechanism. This approach aligns selection with the global optimization objective, improving efficiency and stability. The paper is published on arXiv with ID 2605.05896.
Key facts
- VARS-FL stands for Validation-Aligned Reputation Scoring for Federated Learning.
- It addresses client selection in non-IID federated learning for IoT systems.
- Traditional stateless selection treats each round independently.
- VARS-FL uses server-side validation loss reduction to measure client contribution.
- Reputation scores are aggregated via a sliding-window mechanism.
- The approach targets slow convergence and unstable training.
- IoT and IIoT environments have highly heterogeneous data.
- The paper is available on arXiv under ID 2605.05896.
Entities
Institutions
- arXiv