New XCal-FL algorithm enhances explainability in differentially private federated learning

Researchers have developed XCal-FL, a novel federated learning algorithm that dynamically calibrates differential privacy noise to enhance explainability. This closed-loop system uses prediction logit variations, counterfactual margins, and saliency concentration to adjust noise levels during training, improving both model accuracy and the trustworthiness of explanations. Experiments show XCal-FL significantly outperforms static-noise federated learning and other adaptive DP methods, particularly in decision-critical applications like medical diagnosis, by offering better privacy-budget efficiency and a distinct dimension of explainability beyond mere predictive performance. AI

IMPACT Enhances trustworthiness of AI models in sensitive applications by improving explainability alongside privacy.

RANK_REASON The cluster contains an academic paper detailing a new algorithm and its experimental results. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• adaptive DP
• differential privacy
• DP noise
• federated learning
• Privacy Budget Allocation Technique Based on Variable Length Window for Traffic Data Publishing with Differential Privacy in Road Networks
• XCal-FL

• paper
• safety
• other

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