Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Federated Balanced Learning

Jiaze Li, Haoran Xu, Wanyi Wu, Changwei Wang, Shuaiguang Li, Jianzhong Ju, Zhenbo Luo, Jian Luan, Youyang Qu, Longxiang Gao, Xudong Yang, Lumin Xing

TL;DR

Federated Balanced Learning (FBL) addresses non-IID data challenges in federated multimedia learning by balancing client-side samples through knowledge sampling and knowledge filling, rather than relying solely on post-hoc optimization corrections. It introduces a Knowledge Alignment strategy to bridge synthetic and real data and a Knowledge Drop strategy to regularize embeddings, enabling edge-side diffusion-based data augmentation to balance distributions under constrained resources. The framework extends to computation-aware settings and real-world, complex domains, demonstrating state-of-the-art results on CIFAR-10/100, Tiny ImageNet, and EuroSAT with strong robustness and scalability. Overall, FBL offers a data-centric approach to mitigating client drift, with practical on-device generation capabilities and broad potential for extension to NLP and other modalities.

Abstract

Federated learning is a paradigm of joint learning in which clients collaborate by sharing model parameters instead of data. However, in the non-iid setting, the global model experiences client drift, which can seriously affect the final performance of the model. Previous methods tend to correct the global model that has already deviated based on the loss function or gradient, overlooking the impact of the client samples. In this paper, we rethink the role of the client side and propose Federated Balanced Learning, i.e., FBL, to prevent this issue from the beginning through sample balance on the client side. Technically, FBL allows unbalanced data on the client side to achieve sample balance through knowledge filling and knowledge sampling using edge-side generation models, under the limitation of a fixed number of data samples on clients. Furthermore, we design a Knowledge Alignment Strategy to bridge the gap between synthetic and real data, and a Knowledge Drop Strategy to regularize our method. Meanwhile, we scale our method to real and complex scenarios, allowing different clients to adopt various methods, and extend our framework to further improve performance. Numerous experiments show that our method outperforms state-of-the-art baselines. The code is released upon acceptance.

Federated Balanced Learning

TL;DR

Federated Balanced Learning (FBL) addresses non-IID data challenges in federated multimedia learning by balancing client-side samples through knowledge sampling and knowledge filling, rather than relying solely on post-hoc optimization corrections. It introduces a Knowledge Alignment strategy to bridge synthetic and real data and a Knowledge Drop strategy to regularize embeddings, enabling edge-side diffusion-based data augmentation to balance distributions under constrained resources. The framework extends to computation-aware settings and real-world, complex domains, demonstrating state-of-the-art results on CIFAR-10/100, Tiny ImageNet, and EuroSAT with strong robustness and scalability. Overall, FBL offers a data-centric approach to mitigating client drift, with practical on-device generation capabilities and broad potential for extension to NLP and other modalities.

Abstract

Federated learning is a paradigm of joint learning in which clients collaborate by sharing model parameters instead of data. However, in the non-iid setting, the global model experiences client drift, which can seriously affect the final performance of the model. Previous methods tend to correct the global model that has already deviated based on the loss function or gradient, overlooking the impact of the client samples. In this paper, we rethink the role of the client side and propose Federated Balanced Learning, i.e., FBL, to prevent this issue from the beginning through sample balance on the client side. Technically, FBL allows unbalanced data on the client side to achieve sample balance through knowledge filling and knowledge sampling using edge-side generation models, under the limitation of a fixed number of data samples on clients. Furthermore, we design a Knowledge Alignment Strategy to bridge the gap between synthetic and real data, and a Knowledge Drop Strategy to regularize our method. Meanwhile, we scale our method to real and complex scenarios, allowing different clients to adopt various methods, and extend our framework to further improve performance. Numerous experiments show that our method outperforms state-of-the-art baselines. The code is released upon acceptance.
Paper Structure (20 sections, 10 equations, 5 figures, 4 tables, 1 algorithm)

This paper contains 20 sections, 10 equations, 5 figures, 4 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Preliminary
  4. Methodology
  5. Class Classification
  6. Knowledge Filling
  7. Knowledge Sampling
  8. Extended and Unified Framework
  9. Unified Perspective and Generalization of FBL
  10. Experiments
  11. Experiment Setting
  12. Performance Evaluation
  13. Extended and Unified Framework
  14. Scalability and Robustness
  15. Ablation Study
  16. ...and 5 more sections

Figures (5)

  • Figure 1: Illustration of FBL. Our framework mainly includes two parts: Knowledge Filling and Knowledge Sampling. Before these two parts, our framework first needs to determine the distributional balance point and then classify all classes into three categories: data-excessive, data-scarce, and data-missing. For data-excessive classes, we use knowledge sampling, while for data-scarce and data-missing classes, we utilize knowledge Filling.
  • Figure 2: Illustration of the Knowledge Alignment Strategy and Knowledge Drop Strategy.
  • Figure 3: Extended and Unified Framework. Clients can self-decide to use either computation-constrained or computation-unconstrained resources based on the duration of available computing resources.
  • Figure 4: (a) Visualization of Knowledge Alignment Strategy and Knowledge Drop Strategy. It brings the synthetic data's domain closer to the domain of the real data. (b) Visualize of the results of our method using different cycles.
  • Figure 5: Real and synthetic images on the EuroSAT dataset.