---
title: "Decoding-Unlearning: Fact Forgetting via Entropy-Guided Inference"
description: "Pu et al. propose SEGUE, which forgets at inference time with no training and no change to weights. 54 references."
authors: ["Jingwen Pu", "Mingjun Shi", "Xinrui Ren", "Yizhe Wang", "Xinyu Zhang", "Zhaokun Wang", "Kun She"]
year: 2026
venue: "Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)"
source: "https://aclanthology.org/2026.acl-long.1850/"
doi: "10.18653/v1/2026.acl-long.1850"
licence: "CC BY 4.0 (ACL Anthology)"
licence_url: "https://creativecommons.org/licenses/by/4.0/"
summary_licence: "Elamur Cosmic License"
canonical: "https://research.elamur.ai/free/en/papers/pu-2026-decoding-unlearning-fact-forgetting-entropy/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/pu-2026-decoding-unlearning-fact-forgetting-entropy/)

[home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2026](https://md.research.elamur.ai/free/en/papers/2026/page/27/) / Decoding-Unlearning: Fact Forgetting via Entropy-Guided Inference

# Decoding-Unlearning: Fact Forgetting via Entropy-Guided Inference

**Authors:** [Jingwen Pu](https://md.research.elamur.ai/free/en/authors/jingwen-pu/), Mingjun Shi, Xinrui Ren, Yizhe Wang, [Xinyu Zhang](https://md.research.elamur.ai/free/en/authors/xinyu-zhang/), [Zhaokun Wang](https://md.research.elamur.ai/free/en/authors/zhaokun-wang/) and Kun She · **Year:** 2026 · **Venue:** [Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)](https://md.research.elamur.ai/free/en/venues/acl/) ([ACL 2026](https://md.research.elamur.ai/free/en/venues/acl/2026/)) · **Publisher:** [Association for Computational Linguistics](https://md.research.elamur.ai/free/en/publishers/association-for-computational-linguistics/) · **Licence:** [CC BY 4.0 (ACL Anthology)](https://creativecommons.org/licenses/by/4.0/)

**Links:** [Publisher's page](https://aclanthology.org/2026.acl-long.1850/) · [PDF](https://aclanthology.org/2026.acl-long.1850.pdf) · [DOI 10.18653/v1/2026.acl-long.1850](https://doi.org/10.18653/v1/2026.acl-long.1850) · [OpenAlex](https://openalex.org/W7166806340)

**Citations:** 0 citations (OpenAlex, 2026-10-14)

## Summary
Large language models memorize copyrighted, private, or hazardous facts. Most unlearning methods remove them by fine-tuning, which is expensive and irreversible and requires forget and retain datasets. Pu et al. propose SEGUE, which forgets at inference time with no training and no change to weights. It has two stages:

- **Single2Dual (S2D) Probe:** a short natural-language suffix is attached to the query. In one forward pass, a yes/no-restricted prediction flags whether the query touches a concept to be forgotten, while the cached query states are kept for generation.
- **Entropy-guided decoding:** a Target Token Entropy (TTE) signal, read from an informative layer and combined with logits, identifies low-entropy "factual units." For flagged queries, these tokens are hard-masked and then smoothly decayed. This pushes the model toward fluent but non-factual answers instead of refusals or gibberish. For other queries, stable paths are reinforced.

On MUSE, RWKU, and WMDP (copyright, entity, and hazardous knowledge), SEGUE strongly suppresses target knowledge while keeping general ability and fluency. It outperforms most inference-time baselines. This matters because unlearning becomes cheap, reversible, and possible without full training data.

## Key points
- On LlaMA-3-8B-Instruct, RWKU forget scores fall from 88.31/74.92/76.09 to 8.93/6.42/5.71. The general score changes only slightly (64.38 → 61.13), and WMDP-Bio accuracy drops from 72.74 to 25.69.
- On Qwen2.5-14B, results are similar: RWKU scores go from about 64.66 to 6.12, while the general score moves from 79.50 to 77.63.
- On RWKU-QA, SEGUE adds 0 parameters and needs about 14 GB of memory and roughly 9 minutes in total (as read from a garbled table). Fine-tuning baselines such as NPO take up to about 115 minutes and add around 20M parameters.
- Computing TTE once (static) beats recomputing it at every step (dynamic): 508 s vs 1673 s, with slightly better forgetting (4.65 vs 6.80).
- Stated limitations: evaluation covers only single-turn, short-form QA, copyright-related data is limited, and the largest model tested has 14B parameters.

*Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.*

## Institutions

- [University of Electronic Science and Technology of China](https://md.research.elamur.ai/free/en/universities/university-of-electronic-science-and-technology-of-china/) — University, Chengdu, China · 31 works · 2,012 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 16,667 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 16,194 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 16,039 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 15,891 works

## Cite

Cite as: Jingwen Pu, Mingjun Shi, Xinrui Ren, Yizhe Wang, Xinyu Zhang, Zhaokun Wang and Kun She. 2026. Decoding-Unlearning: Fact Forgetting via Entropy-Guided Inference. Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers).

The paper itself, not this page. BibTeX from the library's record of the work.

```bibtex
@inproceedings{pu2026decoding,
  title = {{Decoding-Unlearning: Fact Forgetting via Entropy-Guided Inference}},
  author = {Jingwen Pu, Mingjun Shi, Xinrui Ren, Yizhe Wang, Xinyu Zhang, Zhaokun Wang and Kun She},
  booktitle = {Proceedings of the 64th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)},
  year = {2026},
  doi = {10.18653/v1/2026.acl-long.1850},
  url = {https://doi.org/10.18653/v1/2026.acl-long.1850}
}
```

- [← previous Decoding Text Spans for Efficient and Accurate Named-Entity Recognition](https://md.research.elamur.ai/free/en/papers/maracani-2026-decoding-text-spans-efficient-accurate/)
- [next → Decomposed Prompting Does Not Fix Knowledge Gaps, But Helps Models Say “I Don't Know”](https://md.research.elamur.ai/free/en/papers/madhwal-2026-decomposed-prompting-does-not-fix/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/27/)

## References

54 references, 14 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Karuna Bhaila, Minh-Hao Van, and Xintao Wu. 2025. [Soft prompting for unlearning in large language models](https://md.research.elamur.ai/free/en/papers/bhaila-2025-soft-prompting-unlearning-large-language/). In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), pages 4046– 40…
2. Lucas Bourtoule, Varun Chandrasekaran, Christopher A Choquette-Choo, Hengrui Jia, Adelin Travers, Baiwu Zhang, David Lie, and Nicolas Papernot. 2021. Machine unlearning. In 2021 IEEE symposium on security and privacy (SP), pages 141–159. IEEE. [find](https://scholar.google.com/scholar?q=Lucas+Bourtoule%2C+Varun+Chandrasekaran%2C+Christopher+A+Choquette%2DChoo%2C+Hengrui+Jia%2C+Adelin+Travers%2C+Baiwu+Zhang%2C+David+Lie%2C+and+Nicolas+Papernot%2E+2021%2E+Machine+unlearning%2E+In+2021+IEEE+symposium+on+secu)
3. Yinzhi Cao and Junfeng Yang. 2015. Towards making systems forget with machine unlearning. In 2015 IEEE symposium on security and privacy, pages 463– 480. IEEE. [find](https://scholar.google.com/scholar?q=Yinzhi+Cao+and+Junfeng+Yang%2E+2015%2E+Towards+making+systems+forget+with+machine+unlearning%2E+In+2015+IEEE+symposium+on+security+and+privacy%2C+pages+463%E2%80%93+480%2E+IEEE%2E)
4. Shiqi Chen, Miao Xiong, Junteng Liu, Zhengxuan Wu, Teng Xiao, Siyang Gao, and Junxian He. 2024. In-context sharpness as alerts: An inner representation perspective for hallucination mitigation. In Forty-first International Conference on Machine Learning. [find](https://scholar.google.com/scholar?q=Shiqi+Chen%2C+Miao+Xiong%2C+Junteng+Liu%2C+Zhengxuan+Wu%2C+Teng+Xiao%2C+Siyang+Gao%2C+and+Junxian+He%2E+2024%2E+In%2Dcontext+sharpness+as+alerts%3A+An+inner+representation+perspective+for+hallucination+mitigation%2E+In+For)
5. Guy Dar, Mor Geva, Ankit Gupta, and Jonathan Berant. 2023. [Analyzing transformers in embedding space](https://md.research.elamur.ai/free/en/papers/dar-2023-analyzing-transformers-embedding-space/). In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 16124–16170.
6. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, and 1 others. 2024. [A survey on in-context learning](https://md.research.elamur.ai/free/en/papers/dong-2023-survey-context-learning/). In Proceedings of the 2024 conference on empirical methods in natural language processing, pages 1107–1128.
7. Xuefeng Du, Reshmi Ghosh, Robert Sim, Ahmed Salem, Vitor Carvalho, Emily Lawton, Yixuan Li, and Jack W Stokes. 2024. Vlmguard: Defending vlms against malicious prompts via unlabeled data. arXiv preprint arXiv:2410.00296. — [arXiv:2410.00296](https://arxiv.org/abs/2410.00296) [find](https://scholar.google.com/scholar?q=Xuefeng+Du%2C+Reshmi+Ghosh%2C+Robert+Sim%2C+Ahmed+Salem%2C+Vitor+Carvalho%2C+Emily+Lawton%2C+Yixuan+Li%2C+and+Jack+W+Stokes%2E+2024%2E+Vlmguard%3A+Defending+vlms+against+malicious+prompts+via+unlabeled+data%2E+arXiv+prepri)
8. Ronen Eldan and Mark Russinovich. 2023. Who’s harry potter? approximate unlearning in llms. arXiv preprint arXiv:2310.02238. [find](https://scholar.google.com/scholar?q=Ronen+Eldan+and+Mark+Russinovich%2E+2023%2E+Who%E2%80%99s+harry+potter%3F+approximate+unlearning+in+llms%2E+arXiv+preprint+arXiv%3A2310%2E02238%2E)
9. Tingchen Fu, Yupeng Hou, Julian McAuley, and Rui Yan. 2025. Unlocking decoding-time controllability: Gradient-free multi-objective alignment with contrastive prompts. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human… [find](https://scholar.google.com/scholar?q=Tingchen+Fu%2C+Yupeng+Hou%2C+Julian+McAuley%2C+and+Rui+Yan%2E+2025%2E+Unlocking+decoding%2Dtime+controllability%3A+Gradient%2Dfree+multi%2Dobjective+alignment+with+contrastive+prompts%2E+In+Proceedings+of+the+2025+Confer)
10. Jiahui Gao, Renjie Pi, Tianyang Han, Han Wu, Lanqing HONG, Lingpeng Kong, Xin Jiang, and Zhenguo Li. 2024a. CoCA: Regaining safety-awareness of multimodal large language models with constitutional calibration. In First Conference on Language Modeling. [find](https://scholar.google.com/scholar?q=Jiahui+Gao%2C+Renjie+Pi%2C+Tianyang+Han%2C+Han+Wu%2C+Lanqing+HONG%2C+Lingpeng+Kong%2C+Xin+Jiang%2C+and+Zhenguo+Li%2E+2024a%2E+CoCA%3A+Regaining+safety%2Dawareness+of+multimodal+large+language+models+with+constitutional+cal)
11. Songyang Gao, Qiming Ge, Wei Shen, Shihan Dou, Junjie Ye, Xiao Wang, Rui Zheng, Yicheng Zou, Zhi Chen, Hang Yan, and 1 others. 2024b. Linear alignment: A closed-form solution for aligning human preferences without tuning and feedback. arXiv preprint arXiv:2401.11458. — [arXiv:2401.11458](https://arxiv.org/abs/2401.11458) [find](https://scholar.google.com/scholar?q=Songyang+Gao%2C+Qiming+Ge%2C+Wei+Shen%2C+Shihan+Dou%2C+Junjie+Ye%2C+Xiao+Wang%2C+Rui+Zheng%2C+Yicheng+Zou%2C+Zhi+Chen%2C+Hang+Yan%2C+and+1+others%2E+2024b%2E+Linear+alignment%3A+A+closed%2Dform+solution+for+aligning+human+prefer)
12. Mor Geva, Roei Schuster, Jonathan Berant, and Omer Levy. 2021. Transformer feed-forward layers are key-value memories. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, pages 5484–5495. [find](https://scholar.google.com/scholar?q=Mor+Geva%2C+Roei+Schuster%2C+Jonathan+Berant%2C+and+Omer+Levy%2E+2021%2E+Transformer+feed%2Dforward+layers+are+key%2Dvalue+memories%2E+In+Proceedings+of+the+2021+Conference+on+Empirical+Methods+in+Natural+Language+Pr)
13. Dan Hendrycks, Collin Burns, Steven Basart, Andy Zou, Mantas Mazeika, Dawn Song, and Jacob Steinhardt. 2021. [Measuring massive multitask language understanding](https://md.research.elamur.ai/free/en/papers/hendrycks-2020-measuring-massive-multitask-language-understanding/). Proceedings of the International Conference on Learning Representations (ICLR).
14. James Y. Huang, Wenxuan Zhou, Fei Wang, Fred Morstatter, Sheng Zhang, Hoifung Poon, and Muhao Chen. 2025. Offset unlearning for large language models. Transactions on Machine Learning Research. [find](https://scholar.google.com/scholar?q=James+Y%2E+Huang%2C+Wenxuan+Zhou%2C+Fei+Wang%2C+Fred+Morstatter%2C+Sheng+Zhang%2C+Hoifung+Poon%2C+and+Muhao+Chen%2E+2025%2E+Offset+unlearning+for+large+language+models%2E+Transactions+on+Machine+Learning+Research%2E)
15. Joel Jang, Dongkeun Yoon, Sohee Yang, Sungmin Cha, Moontae Lee, Lajanugen Logeswaran, and Minjoon Seo. 2023. [Knowledge unlearning for mitigating privacy risks in language models](https://md.research.elamur.ai/free/en/papers/jang-2022-knowledge-unlearning-mitigating-privacy-risks/). In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages…
16. Jiabao Ji, Yujian Liu, Yang Zhang, Gaowen Liu, Ramana R Kompella, Sijia Liu, and Shiyu Chang. 2024. Reversing the forget-retain objectives: An efficient llm unlearning framework from logit difference. Advances in Neural Information Processing Systems, 37:12581–12611. [find](https://scholar.google.com/scholar?q=Jiabao+Ji%2C+Yujian+Liu%2C+Yang+Zhang%2C+Gaowen+Liu%2C+Ramana+R+Kompella%2C+Sijia+Liu%2C+and+Shiyu+Chang%2E+2024%2E+Reversing+the+forget%2Dretain+objectives%3A+An+efficient+llm+unlearning+framework+from+logit+difference%2E)
17. Zhuoran Jin, Pengfei Cao, Chenhao Wang, Zhitao He, Hongbang Yuan, Jiachun Li, Yubo Chen, Kang Liu, and Jun Zhao. 2024. Rwku: Benchmarking real-world knowledge unlearning for large language models. Advances in Neural Information Processing Systems, 37:98213–98263. [find](https://scholar.google.com/scholar?q=Zhuoran+Jin%2C+Pengfei+Cao%2C+Chenhao+Wang%2C+Zhitao+He%2C+Hongbang+Yuan%2C+Jiachun+Li%2C+Yubo+Chen%2C+Kang+Liu%2C+and+Jun+Zhao%2E+2024%2E+Rwku%3A+Benchmarking+real%2Dworld+knowledge+unlearning+for+large+language+models%2E+Adv)
18. Divyansh Kaushik, Amrith Setlur, Eduard H Hovy, and Zachary Chase Lipton. 2021. Explaining the efficacy of counterfactually augmented data. In International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Divyansh+Kaushik%2C+Amrith+Setlur%2C+Eduard+H+Hovy%2C+and+Zachary+Chase+Lipton%2E+2021%2E+Explaining+the+efficacy+of+counterfactually+augmented+data%2E+In+International+Conference+on+Learning+Representations%2E)
19. Bolian Li, Yifan Wang, Ananth Grama, and Ruqi Zhang. 2024a. Cascade reward sampling for efficient decoding-time alignment. In ICML 2024 Next Generation of AI Safety Workshop. [find](https://scholar.google.com/scholar?q=Bolian+Li%2C+Yifan+Wang%2C+Ananth+Grama%2C+and+Ruqi+Zhang%2E+2024a%2E+Cascade+reward+sampling+for+efficient+decoding%2Dtime+alignment%2E+In+ICML+2024+Next+Generation+of+AI+Safety+Workshop%2E)
20. Nathaniel Li, Alexander Pan, Anjali Gopal, Summer Yue, Daniel Berrios, Alice Gatti, Justin D Li, Ann-Kathrin Dombrowski, Shashwat Goel, Long Phan, and 1 others. 2024b. The wmdp benchmark: Measuring and reducing malicious use with unlearning. arXiv preprint arXiv:2403.03218. [find](https://scholar.google.com/scholar?q=Nathaniel+Li%2C+Alexander+Pan%2C+Anjali+Gopal%2C+Summer+Yue%2C+Daniel+Berrios%2C+Alice+Gatti%2C+Justin+D+Li%2C+Ann%2DKathrin+Dombrowski%2C+Shashwat+Goel%2C+Long+Phan%2C+and+1+others%2E+2024b%2E+The+wmdp+benchmark%3A+Measuring+an)
21. Xiang Lisa Li, Ari Holtzman, Daniel Fried, Percy Liang, Jason Eisner, Tatsunori B Hashimoto, Luke Zettlemoyer, and Mike Lewis. 2023. [Contrastive decoding: Open-ended text generation as optimization](https://md.research.elamur.ai/free/en/papers/li-2022-contrastive-decoding-open-ended-text/). In Proceedings of the 61st annual meeting of the association for computational linguistics (volume 1:…
22. Yuhui Li, Fangyun Wei, Jinjing Zhao, Chao Zhang, and Hongyang Zhang. 2024c. RAIN: Your language models can align themselves without finetuning. In The Twelfth International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Yuhui+Li%2C+Fangyun+Wei%2C+Jinjing+Zhao%2C+Chao+Zhang%2C+and+Hongyang+Zhang%2E+2024c%2E+RAIN%3A+Your+language+models+can+align+themselves+without+finetuning%2E+In+The+Twelfth+International+Conference+on+Learning+Repr)
23. Chin-Yew Lin. 2004. [Rouge: A package for automatic evaluation of summaries](https://md.research.elamur.ai/free/en/papers/lin-2004-rouge-package-automatic-evaluation-summaries/). In Text summarization branches out, pages 74–81.
24. Zicheng Lin, Tian Liang, Jiahao Xu, Qiuzhi Liu, Xing Wang, Ruilin Luo, Chufan Shi, Siheng Li, Yujiu Yang, and Zhaopeng Tu. 2025. Critical tokens matter: Token-level contrastive estimation enhances LLM’s reasoning capability. In Forty-second International Conference on Machine Learning. [find](https://scholar.google.com/scholar?q=Zicheng+Lin%2C+Tian+Liang%2C+Jiahao+Xu%2C+Qiuzhi+Liu%2C+Xing+Wang%2C+Ruilin+Luo%2C+Chufan+Shi%2C+Siheng+Li%2C+Yujiu+Yang%2C+and+Zhaopeng+Tu%2E+2025%2E+Critical+tokens+matter%3A+Token%2Dlevel+contrastive+estimation+enhances+LLM)
25. Alisa Liu, Maarten Sap, Ximing Lu, Swabha Swayamdipta, Chandra Bhagavatula, Noah A Smith, and Yejin Choi. 2021. [Dexperts: Decoding-time controlled text generation with experts and anti-experts](https://md.research.elamur.ai/free/en/papers/liu-2021-dexperts-decoding-time-controlled-text/). In Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th In…
26. Chris Liu, Yaxuan Wang, Jeffrey Flanigan, and Yang Liu. 2024. Large language model unlearning via embedding-corrupted prompts. Advances in Neural Information Processing Systems, 37:118198–118266. [find](https://scholar.google.com/scholar?q=Chris+Liu%2C+Yaxuan+Wang%2C+Jeffrey+Flanigan%2C+and+Yang+Liu%2E+2024%2E+Large+language+model+unlearning+via+embedding%2Dcorrupted+prompts%2E+Advances+in+Neural+Information+Processing+Systems%2C+37%3A118198%E2%80%93118266%2E)
27. Qin Liu, Chao Shang, Ling Liu, Nikolaos Pappas, Jie Ma, Neha Anna John, Srikanth Doss, Lluis Marquez, Miguel Ballesteros, and Yassine Benajiba. 2025. Unraveling and mitigating safety alignment degradation of vision-language models. In Findings of the Association for Computational Linguistics: ACL 20… [find](https://scholar.google.com/scholar?q=Qin+Liu%2C+Chao+Shang%2C+Ling+Liu%2C+Nikolaos+Pappas%2C+Jie+Ma%2C+Neha+Anna+John%2C+Srikanth+Doss%2C+Lluis+Marquez%2C+Miguel+Ballesteros%2C+and+Yassine+Benajiba%2E+2025%2E+Unraveling+and+mitigating+safety+alignment+degrada)
28. Samuel Marks and Max Tegmark. 2023. [The geometry of truth: Emergent linear structure in large language model representations of true/false datasets](https://md.research.elamur.ai/free/en/papers/marks-2023-geometry-truth-emergent-linear-structure/). arXiv preprint arXiv:2310.06824.
29. Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. 2022. Locating and editing factual associations in gpt. Advances in neural information processing systems, 35:17359–17372. [find](https://scholar.google.com/scholar?q=Kevin+Meng%2C+David+Bau%2C+Alex+Andonian%2C+and+Yonatan+Belinkov%2E+2022%2E+Locating+and+editing+factual+associations+in+gpt%2E+Advances+in+neural+information+processing+systems%2C+35%3A17359%E2%80%9317372%2E)
30. Martin Pawelczyk, Seth Neel, and Himabindu Lakkaraju. 2024. In-context unlearning: Language models as few-shot unlearners. In International Conference on Machine Learning, pages 40034–40050. PMLR. [find](https://scholar.google.com/scholar?q=Martin+Pawelczyk%2C+Seth+Neel%2C+and+Himabindu+Lakkaraju%2E+2024%2E+In%2Dcontext+unlearning%3A+Language+models+as+few%2Dshot+unlearners%2E+In+International+Conference+on+Machine+Learning%2C+pages+40034%E2%80%9340050%2E+PMLR%2E)
31. Zixuan Qin, Qingchen Yu, Kunlin Lyu, Zhaoxin Fan, and Yifan Sun. 2025. The achilles’ heel of llms: How altering a handful of neurons can cripple language abilities. arXiv preprint arXiv:2510.10238. — [arXiv:2510.10238](https://arxiv.org/abs/2510.10238) [find](https://scholar.google.com/scholar?q=Zixuan+Qin%2C+Qingchen+Yu%2C+Kunlin+Lyu%2C+Zhaoxin+Fan%2C+and+Yifan+Sun%2E+2025%2E+The+achilles%E2%80%99+heel+of+llms%3A+How+altering+a+handful+of+neurons+can+cripple+language+abilities%2E+arXiv+preprint+arXiv%3A2510%2E10238%2E+%E2%80%94+)
32. Jiahao Qiu, Yifu Lu, Yifan Zeng, Jiacheng Guo, Jiayi Geng, Huazheng Wang, Kaixuan Huang, Yue Wu, and Mengdi Wang. 2024. Treebon: Enhancing inference-time alignment with speculative tree-search and best-of-n sampling. arXiv preprint arXiv:2410.16033. [find](https://scholar.google.com/scholar?q=Jiahao+Qiu%2C+Yifu+Lu%2C+Yifan+Zeng%2C+Jiacheng+Guo%2C+Jiayi+Geng%2C+Huazheng+Wang%2C+Kaixuan+Huang%2C+Yue+Wu%2C+and+Mengdi+Wang%2E+2024%2E+Treebon%3A+Enhancing+inference%2Dtime+alignment+with+speculative+tree%2Dsearch+and+bes)
33. Ruichen Qiu, Jiajun Tan, Jiayue Pu, Honglin Wang, Xiao-Shan Gao, and Fei Sun. 2025. A survey on unlearning in large language models. arXiv preprint arXiv:2510.25117. — [arXiv:2510.25117](https://arxiv.org/abs/2510.25117) [find](https://scholar.google.com/scholar?q=Ruichen+Qiu%2C+Jiajun+Tan%2C+Jiayue+Pu%2C+Honglin+Wang%2C+Xiao%2DShan+Gao%2C+and+Fei+Sun%2E+2025%2E+A+survey+on+unlearning+in+large+language+models%2E+arXiv+preprint+arXiv%3A2510%2E25117%2E+%E2%80%94+%5BarXiv%3A2510%2E25117%5D%28https%3A%2F%2Farxiv)
34. Ori Ram, Liat Bezalel, Adi Zicher, Yonatan Belinkov, Jonathan Berant, and Amir Globerson. 2023. What are you token about? dense retrieval as distributions over the vocabulary. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 24… [find](https://scholar.google.com/scholar?q=Ori+Ram%2C+Liat+Bezalel%2C+Adi+Zicher%2C+Yonatan+Belinkov%2C+Jonathan+Berant%2C+and+Amir+Globerson%2E+2023%2E+What+are+you+token+about%3F+dense+retrieval+as+distributions+over+the+vocabulary%2E+In+Proceedings+of+the+61)
35. Anil Ramakrishna, Yixin Wan, Xiaomeng Jin, Kai-Wei Chang, Zhiqi Bu, Bhanukiran Vinzamuri, Volkan Cevher, Mingyi Hong, and Rahul Gupta. 2025. Lume: Llm unlearning with multitask evaluations. arXiv preprint arXiv:2502.15097. [find](https://scholar.google.com/scholar?q=Anil+Ramakrishna%2C+Yixin+Wan%2C+Xiaomeng+Jin%2C+Kai%2DWei+Chang%2C+Zhiqi+Bu%2C+Bhanukiran+Vinzamuri%2C+Volkan+Cevher%2C+Mingyi+Hong%2C+and+Rahul+Gupta%2E+2025%2E+Lume%3A+Llm+unlearning+with+multitask+evaluations%2E+arXiv+prep)
36. Protection Regulation. 2016. Regulation (eu) 2016/679 of the european parliament and of the council. Regulation (eu), 679(2016):10–13. [find](https://scholar.google.com/scholar?q=Protection+Regulation%2E+2016%2E+Regulation+%28eu%29+2016%2F679+of+the+european+parliament+and+of+the+council%2E+Regulation+%28eu%29%2C+679%282016%29%3A10%E2%80%9313%2E)
37. Alexander Robey, Eric Wong, Hamed Hassani, and George J. Pappas. 2025. [SmoothLLM: Defending large language models against jailbreaking attacks](https://md.research.elamur.ai/free/en/papers/robey-2023-smoothllm-defending-large-language-models/). Transactions on Machine Learning Research.
38. Debdeep Sanyal and Murari Mandal. 2025. Agents are all you need for LLM unlearning. In Second Conference on Language Modeling. [find](https://scholar.google.com/scholar?q=Debdeep+Sanyal+and+Murari+Mandal%2E+2025%2E+Agents+are+all+you+need+for+LLM+unlearning%2E+In+Second+Conference+on+Language+Modeling%2E)
39. Weijia Shi, Xiaochuang Han, Mike Lewis, Yulia Tsvetkov, Luke Zettlemoyer, and Wen-tau Yih. 2024. [Trusting your evidence: Hallucinate less with context-aware decoding](https://md.research.elamur.ai/free/en/papers/shi-2024-trusting-your-evidence-hallucinate-less/). In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language…
40. Weijia Shi, Jaechan Lee, Yangsibo Huang, Sadhika Malladi, Jieyu Zhao, Ari Holtzman, Daogao Liu, Luke Zettlemoyer, Noah A. Smith, and Chiyuan Zhang. 2025. MUSE: Machine unlearning six-way evaluation for language models. In The Thirteenth International Conference on Learning Representations. 2022 IEEE… [find](https://scholar.google.com/scholar?q=Weijia+Shi%2C+Jaechan+Lee%2C+Yangsibo+Huang%2C+Sadhika+Malladi%2C+Jieyu+Zhao%2C+Ari+Holtzman%2C+Daogao+Liu%2C+Luke+Zettlemoyer%2C+Noah+A%2E+Smith%2C+and+Chiyuan+Zhang%2E+2025%2E+MUSE%3A+Machine+unlearning+six%2Dway+evaluation+fo)
41. Bozhong Tian, Xiaozhuan Liang, Siyuan Cheng, Qingbin Liu, Mengru Wang, Dianbo Sui, Xi Chen, Huajun Chen, and Ningyu Zhang. 2024. To forget or not? towards practical knowledge unlearning for large language models. In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 1524–15… [find](https://scholar.google.com/scholar?q=Bozhong+Tian%2C+Xiaozhuan+Liang%2C+Siyuan+Cheng%2C+Qingbin+Liu%2C+Mengru+Wang%2C+Dianbo+Sui%2C+Xi+Chen%2C+Huajun+Chen%2C+and+Ningyu+Zhang%2E+2024%2E+To+forget+or+not%3F+towards+practical+knowledge+unlearning+for+large+lang)
42. Paul Voigt and Axel Von dem Bussche. 2017. The eu general data protection regulation (gdpr). A practical guide, 1st ed., Cham: Springer International Publishing, 10(3152676):10–5555. [find](https://scholar.google.com/scholar?q=Paul+Voigt+and+Axel+Von+dem+Bussche%2E+2017%2E+The+eu+general+data+protection+regulation+%28gdpr%29%2E+A+practical+guide%2C+1st+ed%2E%2C+Cham%3A+Springer+International+Publishing%2C+10%283152676%29%3A10%E2%80%935555%2E)
43. Tianlong Wang, Junzhe Chen, Xueting Han, and Jing Bai. 2024. Cpl: Critical plan step learning boosts llm generalization in reasoning tasks. arXiv preprint arXiv:2409.08642. — [arXiv:2409.08642](https://arxiv.org/abs/2409.08642) [find](https://scholar.google.com/scholar?q=Tianlong+Wang%2C+Junzhe+Chen%2C+Xueting+Han%2C+and+Jing+Bai%2E+2024%2E+Cpl%3A+Critical+plan+step+learning+boosts+llm+generalization+in+reasoning+tasks%2E+arXiv+preprint+arXiv%3A2409%2E08642%2E+%E2%80%94+%5BarXiv%3A2409%2E08642%5D%28https%3A)
44. Xiaoyu Xu, Minxin Du, Qingqing Ye, and Haibo Hu. 2025a. Obliviate: Robust and practical machine unlearning for large language models. arXiv preprint arXiv:2505.04416. [find](https://scholar.google.com/scholar?q=Xiaoyu+Xu%2C+Minxin+Du%2C+Qingqing+Ye%2C+and+Haibo+Hu%2E+2025a%2E+Obliviate%3A+Robust+and+practical+machine+unlearning+for+large+language+models%2E+arXiv+preprint+arXiv%3A2505%2E04416%2E)
45. Yuancheng Xu, Udari Madhushani Sehwag, Alec Koppel, Sicheng Zhu, Bang An, Furong Huang, and Sumitra Ganesh. 2025b. GenARM: Reward guided generation with autoregressive reward model for test-time alignment. In The Thirteenth International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Yuancheng+Xu%2C+Udari+Madhushani+Sehwag%2C+Alec+Koppel%2C+Sicheng+Zhu%2C+Bang+An%2C+Furong+Huang%2C+and+Sumitra+Ganesh%2E+2025b%2E+GenARM%3A+Reward+guided+generation+with+autoregressive+reward+model+for+test%2Dtime+align)
46. Zhangchen Xu, Fengqing Jiang, Luyao Niu, Jinyuan Jia, Bill Yuchen Lin, and Radha Poovendran. 2024. [Safedecoding: Defending against jailbreak attacks via safety-aware decoding](https://md.research.elamur.ai/free/en/papers/xu-2024-safedecoding-defending-against-jailbreak-attacks/). In ICLR 2024 Workshop on Secure and Trustworthy Large Language Models.
47. Yuanshun Yao, Xiaojun Xu, and Yang Liu. 2024. [Large language model unlearning](https://md.research.elamur.ai/free/en/papers/yao-2023-large-language-model-unlearning/). Advances in Neural Information Processing Systems, 37:105425–105475.
48. Feifan Song, Yuxuan Fan, Xin Zhang, Peiyi Wang, and Houfeng Wang. 2025. Instantly learning preference alignment via in-context dpo. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Lo… [find](https://scholar.google.com/scholar?q=Feifan+Song%2C+Yuxuan+Fan%2C+Xin+Zhang%2C+Peiyi+Wang%2C+and+Houfeng+Wang%2E+2025%2E+Instantly+learning+preference+alignment+via+in%2Dcontext+dpo%2E+In+Proceedings+of+the+2025+Conference+of+the+Nations+of+the+Americas)
49. Qingchen Yu, Zifan Zheng, Ding Chen, Simin Niu, Bo Tang, Feiyu Xiong, and Zhiyu Li. 2025. Gues-sarena: Guess who i am? a self-adaptive framework for evaluating llms in domain-specific knowledge and reasoning. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics… [find](https://scholar.google.com/scholar?q=Qingchen+Yu%2C+Zifan+Zheng%2C+Ding+Chen%2C+Simin+Niu%2C+Bo+Tang%2C+Feiyu+Xiong%2C+and+Zhiyu+Li%2E+2025%2E+Gues%2Dsarena%3A+Guess+who+i+am%3F+a+self%2Dadaptive+framework+for+evaluating+llms+in+domain%2Dspecific+knowledge+and+re)
50. Shota Takashiro, Takeshi Kojima, Andrew Gambardella, Qi Cao, Yusuke Iwasawa, and Yutaka Matsuo. 2025. Answer when needed, forget when not: Language models pretend to forget via in-context knowledge unlearning. In Findings of the Association for Computational Linguistics: ACL 2025, pages 24872–24885. [find](https://scholar.google.com/scholar?q=Shota+Takashiro%2C+Takeshi+Kojima%2C+Andrew+Gambardella%2C+Qi+Cao%2C+Yusuke+Iwasawa%2C+and+Yutaka+Matsuo%2E+2025%2E+Answer+when+needed%2C+forget+when+not%3A+Language+models+pretend+to+forget+via+in%2Dcontext+knowledge+un)
51. Kechi Zhang, Ge Li, Jia Li, Yihong Dong, and Zhi Jin. 2025. Focused-dpo: Enhancing code generation through focused preference optimization on error-prone points. arXiv preprint arXiv:2502.11475. [find](https://scholar.google.com/scholar?q=Kechi+Zhang%2C+Ge+Li%2C+Jia+Li%2C+Yihong+Dong%2C+and+Zhi+Jin%2E+2025%2E+Focused%2Ddpo%3A+Enhancing+code+generation+through+focused+preference+optimization+on+error%2Dprone+points%2E+arXiv+preprint+arXiv%3A2502%2E11475%2E)
52. Pratiksha Thaker, Yash Maurya, Shengyuan Hu, Zhiwei Steven Wu, and Virginia Smith. 2024. Guardrail baselines for unlearning in llms. arXiv preprint arXiv:2403.03329. — [arXiv:2403.03329](https://arxiv.org/abs/2403.03329) [find](https://scholar.google.com/scholar?q=Pratiksha+Thaker%2C+Yash+Maurya%2C+Shengyuan+Hu%2C+Zhiwei+Steven+Wu%2C+and+Virginia+Smith%2E+2024%2E+Guardrail+baselines+for+unlearning+in+llms%2E+arXiv+preprint+arXiv%3A2403%2E03329%2E+%E2%80%94+%5BarXiv%3A2403%2E03329%5D%28https%3A%2F%2Farxiv)
53. Ruiqi Zhang, Licong Lin, Yu Bai, and Song Mei. 2024. [Negative preference optimization: From catastrophic collapse to effective unlearning](https://md.research.elamur.ai/free/en/papers/zhang-2024-negative-preference-optimization-catastrophic-collapse/). In First Conference on Language Modeling.
54. Anvith Thudi, Gabriel Deza, Varun Chandrasekaran, and Nicolas Papernot. 2022. Unrolling sgd: Understanding factors influencing machine unlearning. In [find](https://scholar.google.com/scholar?q=Anvith+Thudi%2C+Gabriel+Deza%2C+Varun+Chandrasekaran%2C+and+Nicolas+Papernot%2E+2022%2E+Unrolling+sgd%3A+Understanding+factors+influencing+machine+unlearning%2E+In)

*Updated 1 November 2026*