Debunking Optimization Myths in Federated Learning for Medical Image Classification

2025-07-26

AI Summary

Federated learning (FL) for medical image classification often overlooks the decisive impact of local optimization configurations—such as optimizer choice, learning rate, and number of local training epochs—on model performance and deployment robustness, especially on resource-constrained edge devices. Method: We systematically evaluate canonical FL frameworks on colorectal histopathology and blood cell classification tasks, conducting extensive benchmarking across multiple optimizer–learning rate combinations and varying local epoch counts. Contribution/Results: Our analysis reveals that local hyperparameter configuration exerts a stronger influence on convergence and accuracy than algorithmic complexity alone. Crucially, the number of local epochs exhibits a dual effect: moderate increases improve convergence, whereas excessive values degrade global model performance due to client drift and overfitting. By quantifying these effects, we establish the critical role of edge-side optimization tuning and provide a reproducible, empirically grounded guideline—including principled hyperparameter selection protocols—for robust FL deployment in low-resource clinical settings.

Technology Category

• Machine Learning: Learning on the Edge & Model Compression
• Search and Optimization: Local Search
• Computer Vision: Learning & Optimization for CV

Application Category

• User Modeling, Personalization and Recommendation: Federated recommendation systems and personalization
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web
• Systems and Infrastructure for Web, Mobile and WoT: Federated Web and WoT systems, including distributed, federated and edge-based data processing

Abstract

Federated Learning (FL) is a collaborative learning method that enables decentralized model training while preserving data privacy. Despite its promise in medical imaging, recent FL methods are often sensitive to local factors such as optimizers and learning rates, limiting their robustness in practical deployments. In this work, we revisit vanilla FL to clarify the impact of edge device configurations, benchmarking recent FL methods on colorectal pathology and blood cell classification task. We numerically show that the choice of local optimizer and learning rate has a greater effect on performance than the specific FL method. Moreover, we find that increasing local training epochs can either enhance or impair convergence, depending on the FL method. These findings indicate that appropriate edge-specific configuration is more crucial than algorithmic complexity for achieving effective FL.

Problem

• Impact of local optimizers and learning rates on FL performance
• Effect of local training epochs on FL convergence variability
• Edge-specific configurations outweigh algorithmic complexity in FL effectiveness

Innovation

• Revisiting vanilla FL for edge device impact
• Benchmarking FL methods on medical tasks
• Optimizing local configurations over complex algorithms