Abstract
Federated learning (FL) alleviates the need for data sharing, easing site participation in collaborative studies, and enabling improved model generalization by increased data access. However, the training process can be arduous, and its convergence can be delayed by sites with substantially large datasets and long local training times. To address this, we introduce FL Optimized Convergence via Uncertainty-guided Sampling (FOCUS), as a framework to accelerate local training without sacrificing global model performance. Its principle is to reduce unnecessary FL rounds through entropy-based predictive uncertainty guiding local case sampling - to both temporarily exclude easy-to-learn cases and focus learning on more informative samples. A dynamic memory bank records the learning state of each case at each FL round and tracks cases that remain consistently easy, thereby enabling adaptive local sampling that evolves with the global model training. Catastrophic forgetting is mitigated by a periodic replay mechanism that reintroduces excluded cases for training at regular intervals. We evaluate FOCUS incorporated with FedAvg on the Federated brain Tumor Segmentation (FeTS) 2024 challenge data and compare it with the baseline approach of training on all local cases per round. Our results demonstrate a substantial reduction in overall training time over the baseline, while maintaining segmentation performance. This suggests FOCUS as an effective strategy for addressing sample size imbalance in FL, improving efficiency and scalability.
Links to Paper and Supplementary Materials
Main Paper (Open Access Version): https://papers.miccai.org/miccai-2026-sat/paper/DeCaF_003.pdf
SharedIt Link: Not yet available
SpringerLink (DOI): Not yet available
Supplementary Material: Not Submitted
Link to Open Review
Open Review Page: https://openreview.net/forum?id=5JCiwrEpj7
BibTex
@InProceedings{KouKon_FOCUS_MICCAISAT2026,
author = { Koukoutegos, Konstantinos AND Linardos, Akis AND Tang, Wenyi AND Bakas, Spyridon},
title = { { FOCUS: Federated Learning Optimized Convergence via Uncertainty-guided Sampling } },
booktitle = {Medical Image Computing and Computer Assisted Intervention -- MICCAI 2026 Workshops and Challenges},
year = {2026},
publisher = {Springer Nature Switzerland},
volume = {LNCS 17252},
month = {pending},
page = {pending}
}