New optimizer tackles class imbalance in federated learning

Researchers have developed FedCGNM, a novel client-side optimizer designed to address class imbalance in federated learning. This method partitions classes into groups, maintains and normalizes momentum for each group, and uses the sum of normalized group momentums for updates. This approach aims to equalize gradient magnitudes across majority and minority classes and reduce noise from rare-class gradients. Additionally, FedHOO, an X-armed-bandit algorithm, is introduced to efficiently optimize hyperparameter rates in smaller federations by exploiting federated parallelism. AI

IMPACT Introduces novel methods to improve the performance of federated learning models on imbalanced datasets, potentially enhancing applications in sensitive areas like healthcare or finance.

RANK_REASON The cluster contains a research paper detailing a new algorithm for federated learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• FedCGNM
• Federated Learning
• FedHOO

• paper
• other

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →