Highlights
- Pro
Stars
Exponential Family Discriminant Analysis (EFDA) with Lean 4 verified proofs.
Inference-time privacy edits for GPT-Neo style causal language models to suppress personally identifiable information (PII) leakage without retraining.
The adoption of large language models (LLMs) has raised concerns about the extraction of sensitive personal identifiable information (PII) from their training data. We propose a novel query-based d…