Explore G-Drift MIA, a novel white-box method for detecting memorized training data in LLMs by analyzing gradient-induced feature drift and internal represen...
Level: advanced
By Ravi Ranjan
Category: discussion