一作ACL 2026Poster· Natural Language Processing
Exploring Cross-Client Memorization of Training Data in Large Language Models for Federated Learning
Tinnakit Udsa · Can Udomcharoenchaikit · Patomporn Payoungkhamdee · Sarana Nutanong · Norrathep Rattanavipanon
摘要
Federated learning (FL) enables collaborative training without raw data sharing, but still risks training data memorization. Existing FL memorization detection techniques focus on one sample at a time, underestimating more subtle risks of cross-sample memorization. In contrast, recent work on cen...