Researchers have developed FedCC, a new algorithm designed to improve federated learning by addressing label distribution skews. In distillation-based federated learning, clients share predictions on public data, but differing local datasets can lead to biased models. FedCC allows clients to tag ambiguous samples as 'unknown,' which, combined with calibrated pseudo-labels, helps balance confidence and uncertainty. Experiments show FedCC significantly outperforms existing methods, particularly in scenarios with severe label skew, achieving 67.3% accuracy in an extreme case where baselines performed near-randomly. AI
IMPACT Improves robustness of federated learning models in scenarios with imbalanced data distributions.
RANK_REASON The cluster contains an academic paper detailing a new algorithm for federated learning.
Read on Hugging Face Daily Papers →
AI-generated summary · Google Gemini · from 2 sources. How we write summaries →
