Reproduce “Combating Data Imbalances in Federated Semi-supervised Learning with Dual Regulators” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Sikai Bai, Shuaicheng Li, Weiming Zhuang, Jie Zhang, Song Guo, Kunlin Yang, Jun Hou, Shuai Zhang and 2 more
A framework called FedDure was developed to manage data imbalances in Federated Semi-supervised Learning, addressing heterogeneity both across clients and within a client's labeled and unlabeled datasets. This approach, utilizing dual regulators, significantly improves model performance and robustness in real-world decentralized settings.
Key insights
- Addressing both external (across clients) and internal (labeled versus unlabeled within client) data imbalances simultaneously is crucial for practical Federated Semi-supervised Learning.
- An adaptive, bi-level optimization approach with separate coarse-grained and fine-grained regulation is effective in guiding local learning to mitigate bias from skewed data distributions.
- The synergistic combination of instance-level weighting and overall learning effect quantification by dual regulators yields robust and theoretically convergent FSSL training.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.