Research map / Large language models / Language, safety and interpretability
Downgrade to Upgrade: Optimizer Simplification Enhances Robustness in LLM Unlearning
Yicheng Lang, Yihua Zhang, Chongyu Fan, Changsheng Wang, Jinghan Jia, Sijia Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace Adam with zeroth-order AdaZO optimizer (ablation) | NPO unlearning · MUSE-Books, 4-bit quantization | VerbMem on forget set after quantization: 48.5 -> 16.2 | Improved |
| replace Adam with zeroth-order AdaZO optimizer (ablation) | NPO unlearning · MUSE-Books, no quantization or attack | KnowMem on retain set: 53.65 -> 36.6 | Hurt |
| replace Adam with zeroth-order AdaZO optimizer (ablation) | NPO unlearning · MUSE-Books, before quantization | VerbMem on forget set: 0 -> 20.7 | Hurt |
| replace Adam with signSGD or signAdam gradient compression (ablation) | NPO unlearning · MUSE-Books, 4-bit quantization | VerbMem on forget set after quantization: 48.5 -> 15.6 (signSGD), 30.7 (signAdam) | Improved |
| replace Adam with signSGD or signAdam (ablation) | unlearning · MUSE-Books, relearning | robustness to relearning | No clear effect |
| replace Adam with first-order randomized smoothing (RS) objective (ablation) | NPO unlearning · MUSE-Books, 4-bit quantization | VerbMem on forget set after quantization: 48.5 -> 34.5 | Improved |
| add Hybrid FO-ZO alternation instead of Adam only | NPO unlearning · MUSE-Books, 4-bit quantization | VerbMem on forget set after quantization: 48.5 -> 0 | Improved |
| add Hybrid FO-ZO alternation instead of Adam only | NPO unlearning · MUSE-Books, relearning attack | KnowMem on forget set after relearning: 34.8 -> 0 | Improved |
| replace SAM with Hybrid FO-ZO | NPO unlearning · MUSE-Books, 4-bit quantization | VerbMem on forget set after quantization: 30.0 -> 0 | Improved |
| vary hybrid switch step N between FO and ZO (5, 20, 100) (ablation) | Hybrid optimizer · MUSE-Books, NPO | robustness to quantization and relearning | No clear effect |
| assign more FO than ZO steps in each hybrid round (FO=20, ZO=10) (ablation) | Hybrid optimizer · MUSE-Books, NPO | unlearning robustness | Hurt |
| assign more ZO than FO steps in each hybrid round (FO=10, ZO=20) (ablation) | Hybrid optimizer · MUSE-Books, NPO | unlearning effectiveness | Hurt |
Closest papers by meaning
- Simplicity Prevails: Rethinking Negative Preference Optimization for LLM Unlearning (NeurIPS 2025 · similarity 0.93)
- Invariance Makes LLM Unlearning Resilient Even to Unanticipated Downstream Fine-Tuning (ICML 2025 · similarity 0.92)
- LLM Unlearning via Loss Adjustment with Only Forget Data (ICLR 2025 · similarity 0.91)
- Rethinking LLM Unlearning Objectives: A Gradient Perspective and Go Beyond (ICLR 2025 · similarity 0.91)
- Towards Effective Evaluations and Comparisons for LLM Unlearning Methods (ICLR 2025 · similarity 0.91)
- WAGLE: Strategic Weight Attribution for Effective and Modular Unlearning in Large Language Models (NeurIPS 2024 · similarity 0.91)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
