JensUn applies Jensen-Shannon Divergence to both forget and retain losses in LLM unlearning, and a new worst-case paraphrased evaluation shows it preserves utility better than baselines while resisting benign relearning.
Extended discussions and proofs A D ATASET AND PARAPHRASING DETAILS In this section, we explain in detail the LKF generation process and the paraphrasing details
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Unlearning That Lasts: Utility-Preserving, Robust, and Almost Irreversible Forgetting in LLMs
JensUn applies Jensen-Shannon Divergence to both forget and retain losses in LLM unlearning, and a new worst-case paraphrased evaluation shows it preserves utility better than baselines while resisting benign relearning.