CKU freezes neurons identified as important for useful knowledge during gradient-based unlearning, reporting lower jailbreak success rates with minimal utility loss.
Masahiro Kaneko, Danushka Bollegala, and Naoaki Okazaki
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Safety Alignment via Constrained Knowledge Unlearning
CKU freezes neurons identified as important for useful knowledge during gradient-based unlearning, reporting lower jailbreak success rates with minimal utility loss.