MUC-FL: Block-Wise Marginal Utility Contribution for Communication-Efficient Federated Learning
TL;DR
This work proposes Block-Wise Marginal Utility Contribution (MUC), a framework that selectively transmits only the most impactful data blocks based on their contribution to model performance, and applies it to a multi-modal dataset integrated from multiple MIMIC clinical datasets.
Abstract
Federated Learning (FL) enables distributed model training without centralizing data but suffers from high communication overhead. To address this, we propose Block-Wise Marginal Utility Contribution (MUC), a framework that selectively transmits only the most impactful data blocks based on their contribution to model performance. To evaluate our framework, we apply it to a multi-modal dataset integrated from multiple MIMIC clinical datasets and show that only 24 out of $\mathbf{1, 1 3 5}$ candidate blocks $(\mathbf{1. 7 6 \%})$ carry meaningful improvement signals, enabling a potential communication reduction of $\mathbf{4 5}-\mathbf{5 0} \boldsymbol{\%}$ while maintaining or improving model quality. Our deduplication-based block selection achieves a macro F1 score of 0.8566 compared to 0.8155 for standard federated optimization, demonstrating that selective transmission can improve performance, particularly in underrepresented classes.