From Dormant to Deleted: Tamper-Resistant Unlearning Through Weight-Space Regularization

Eleni Triantafillou (Google DeepMind) · David Krueger (University of Cambridge) · Gintare Karolina Dziugaite (Google Research, Brain Team) · Adrian Weller (Cambridge, Alan Turing Institute) · Shoaib Ahmed Siddiqui (University of Cambridge) · Michael Mozer (Google DeepMind)
example-level unlearningfine-tuningforget-set accuracyknowledge recoveryl2-distancelinear mode connectivitymodel retrainingmodel robustnessrelearning attacksretain setstate-of-the-art methodsunlearning methodsunlearning resistancevision classifiersweight-space properties

Recent unlearning methods for LLMs are vulnerable to relearning attacks: knowledge believed-to-be-unlearned re-emerges by fine-tuning on a small set of (even seemingly-unrelated) examples. We study this phenomenon in a controlled setting for example-level unlearning in vision classifiers. We make the surprising discovery that forget-set accuracy can recover from around 50\% post-unlearning to nearly 100\% with fine-tuning on just the *retain* set