Beyond Fixed Rounds: Data-Free Early Stopping for Practical Federated Learning
This work addresses practical deployment challenges in Federated Learning by reducing reliance on fixed rounds or validation data, though it is incremental as it builds on existing FL methods.
The paper tackles the problem of high computational costs and privacy risks in Federated Learning by proposing a data-free early stopping framework that monitors server-side parameters to determine the optimal stopping point, achieving comparable performance to validation-based methods with an average of 45/12 additional rounds for 12.3%/8.9% higher performance on skin lesion and blood cell classification tasks.
Federated Learning (FL) facilitates decentralized collaborative learning without transmitting raw data. However, reliance on fixed global rounds or validation data for hyperparameter tuning hinders practical deployment by incurring high computational costs and privacy risks. To address this, we propose a data-free early stopping framework that determines the optimal stopping point by monitoring the task vector's growth rate using solely server-side parameters. The numerical results on skin lesion/blood cell classification demonstrate that our approach is comparable to validation-based early stopping across various state-of-the-art FL methods. In particular, the proposed framework requires an average of 45/12 (skin lesion/blood cell) additional rounds to achieve over 12.3%/8.9% higher performance than early stopping based on validation data. To the best of our knowledge, this is the first work to propose an data-free early stopping framework for FL methods.