Federated learning improves minority class results with efficient data augmentation
Fast Convergence through Distributed Augmentation for Class-Imbalanced Federated Learning
Machine Learning
Summary
Machine learning systems that learn from data spread across many devices often struggle when some categories have very little data. The authors found that adding fake data to balance category counts can help, but adding too much slows the learning process. They studied how the amount of added data affects training speed and accuracy. Then, they designed a method called DAFL that figures out the smallest amount of data each device needs to add, improving accuracy on rare categories while saving training time.
What this means in practice
- •For mobile app developers: Improve model accuracy on rare categories in federated learning apps without greatly increasing training time.
- •For healthcare data teams: Enhance federated analysis of medical data with unbalanced patient categories by efficiently augmenting minority-class data.
Authors
Arathi Nair M, J. Harshan, Anwitaman Datta
Abstract
In federated learning, mitigating class imbalance is essential to improve minority-class performance. A common approach to address this problem is to augment minority-class samples to achieve local class balance. Existing approaches treat augmentation as a heuristic and do not establish how the amount of augmentation influences the convergence of federated learning, leading to excessive augmentation and increased training time. To address this limitation, we first establish the relationship between augmentation and the convergence behavior of federated learning. Leveraging this insight, we propose DAFL, a distributed augmentation framework that determines the minimum augmentation required for each client-class pair by jointly minimizing augmentation and training time while constraining global class imbalance, thereby improving minority-class F1-score. Experimental results demonstrate that DAFL consistently improves minority-class F1-score while substantially reducing training time, particularly under severe global class imbalance and high label proportion imbalance.