New framework offers private and robust federated multiclass classification

Researchers have developed a new framework for federated multiclass classification that offers both differential privacy and distributional robustness. This approach enables collaborative training of models across multiple sites without centralizing sensitive data, while also protecting individual records and handling label noise. The method utilizes a differentially private federated Wasserstein distributionally robust objective with a Crammer-Singer loss, solved via an ADMM-based algorithm. Experiments demonstrate its effectiveness compared to existing differentially private federated methods, particularly under stringent privacy constraints. AI

IMPACT Enhances privacy and robustness in collaborative AI model training for sensitive data.

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

Read on arXiv cs.LG →

• Andrea Hernandez Guzman
• Crammer-Singer
• Gaussian function
• University of California, Irvine

• paper
• safety
• other

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