Tracing Privacy Leakage of Language Models to Training Data via Adjusted Influence Functions

Open in new window