Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

FedSM: Robust Semantics-Guided Feature Mixup for Bias Reduction in Federated Learning with Long-Tail Data

Jingrui Zhang, Yimeng Xu, Shujie Li, Feng Liang, Haihan Duan, Yanjie Dong, Victor C. M. Leung, Xiping Hu

TL;DR

FedSM is a novel client-centric framework that mitigates bias through semantics-guided feature mixup and lightweight classifier retraining, and consistently outperforms state-of-the-art methods in accuracy, with high robustness to domain shift and computational efficiency.

Abstract

Federated Learning (FL) enables collaborative model training across decentralized clients without sharing private data. However, FL suffers from biased global models due to non-IID and long-tail data distributions. We propose \textbf{FedSM}, a novel client-centric framework that mitigates this bias through semantics-guided feature mixup and lightweight classifier retraining. FedSM uses a pretrained image-text-aligned model to compute category-level semantic relevance, guiding the category selection of local features to mix-up with global prototypes to generate class-consistent pseudo-features. These features correct classifier bias, especially when data are heavily skewed. To address the concern of potential domain shift between the pretrained model and the data, we propose probabilistic category selection, enhancing feature diversity to effectively mitigate biases. All computations are performed locally, requiring minimal server overhead. Extensive experiments on long-tail datasets with various imbalanced levels demonstrate that FedSM consistently outperforms state-of-the-art methods in accuracy, with high robustness to domain shift and computational efficiency.

FedSM: Robust Semantics-Guided Feature Mixup for Bias Reduction in Federated Learning with Long-Tail Data

TL;DR

FedSM is a novel client-centric framework that mitigates bias through semantics-guided feature mixup and lightweight classifier retraining, and consistently outperforms state-of-the-art methods in accuracy, with high robustness to domain shift and computational efficiency.

Abstract

Federated Learning (FL) enables collaborative model training across decentralized clients without sharing private data. However, FL suffers from biased global models due to non-IID and long-tail data distributions. We propose \textbf{FedSM}, a novel client-centric framework that mitigates this bias through semantics-guided feature mixup and lightweight classifier retraining. FedSM uses a pretrained image-text-aligned model to compute category-level semantic relevance, guiding the category selection of local features to mix-up with global prototypes to generate class-consistent pseudo-features. These features correct classifier bias, especially when data are heavily skewed. To address the concern of potential domain shift between the pretrained model and the data, we propose probabilistic category selection, enhancing feature diversity to effectively mitigate biases. All computations are performed locally, requiring minimal server overhead. Extensive experiments on long-tail datasets with various imbalanced levels demonstrate that FedSM consistently outperforms state-of-the-art methods in accuracy, with high robustness to domain shift and computational efficiency.

Paper Structure

This paper contains 14 sections, 8 equations, 5 figures, 5 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Work
  3. Method
  4. Problem Setting
  5. Overview of Our Framework
  6. Local Training
  7. Image Feature Mixup Guided by Label Relevance
  8. Evaluation
  9. Experimental Setup and Implementation Details
  10. Results
  11. Results on CIFAR-10/100-LT.
  12. Results on ImageNet-LT.
  13. Ablation Study
  14. Conclusion

Figures (5)

  • Figure 1: Illustration of Mixup problems. (a) Random Mixup ignores semantic relevance between categories and may blend unrelated samples, such as squirrel and train, across boundaries, producing synthetic data that misguides boundary refinement. (b) When the mountain category has significantly fewer samples or the pine tree category is absent, random Mixup has a higher chance of generating unrepresentative or even misleading synthetic samples.
  • Figure 2: Overview of the FedSM framework. The client side consists of three key phases: a) local training, b) label relevance-guided feature mixup, and c) classifier retraining.
  • Figure 3: Results on CIFAR-10-LT under different similarity functions for relevance score.
  • Figure 4: Results in various classifier retraining settings on CIFAR-100-LT with IF=10.
  • Figure 5: Impact of varying the number of active clients.