← Back to home

Blog

Paper Accepted at COLM 2026! LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning
Pinned

Paper Accepted at COLM 2026! LACUNA: A Testbed for Evaluating Localization Precision for LLM Unlearning

Measuring whether unlearning methods actually target the parameters that store the memorized data

July 8, 2026
UnlearningLocalizationPrivacyLLMsCOLM 2026
Paper Accepted at ICLR 2026! Optimizing Canaries for Privacy Auditing with Metagradient Descent
Pinned

Paper Accepted at ICLR 2026! Optimizing Canaries for Privacy Auditing with Metagradient Descent

Improving privacy lower bounds for DP-SGD using metagradient optimization

January 26, 2026
Privacy AuditingDP-SGDMetagradientsICLR 2026