Withdrawn paper analyzes differential privacy in wireless federated learning

This paper, titled "When Differential Privacy Meets Wireless Federated Learning: An Improved Analysis for Privacy and Convergence," was withdrawn by its author, Hao Liang. The research aimed to address limitations in existing differentially private wireless federated learning (DPWFL) frameworks, particularly concerning the characterization of privacy loss and convergence analyses under restrictive assumptions. The authors proposed a comprehensive analysis for DPWFL with non-convex loss objectives, incorporating device selection and mini-batch sampling, and aimed to demonstrate that privacy loss could converge to a constant. The work also intended to establish convergence guarantees with gradient clipping and derive an explicit privacy-utility trade-off, with numerical results validating the theoretical findings. AI

IMPACT This research, though withdrawn, explored methods to improve privacy and convergence in wireless federated learning, a key area for secure AI model training on distributed data.

RANK_REASON The item is a withdrawn academic paper discussing technical aspects of differential privacy and federated learning. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.LG →

• arXiv
• differential privacy
• DPWFL
• Hao Liang
• Wireless Federated Learning

• paper
• safety

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