---
title: "Maximizing Local Entropy Where It Matters: Prefix-Aware"
description: "The authors propose PALU (Prefix-Aware Localized Unlearning), which applies entropy maximization… 54 references, cited by 2 papers here."
authors: ["Naixin Zhai", "Pengyang Shao", "Binbin Zheng", "Yonghui Yang", "Fei Shen", "Long Bai", "Xun Yang"]
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.893/"
doi: "10.18653/v1/2026.acl-long.893"
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/zhai-2026-maximizing-local-entropy-where-it/"
---
[Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/zhai-2026-maximizing-local-entropy-where-it/)

[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/8/) / Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning

# Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning

**Authors:** Naixin Zhai, Pengyang Shao, [Binbin Zheng](https://md.research.elamur.ai/free/en/authors/binbin-zheng/), [Yonghui Yang](https://md.research.elamur.ai/free/en/authors/yonghui-yang/), Fei Shen, [Long Bai](https://md.research.elamur.ai/free/en/authors/long-bai/) and [Xun Yang](https://md.research.elamur.ai/free/en/authors/xun-yang/) · **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.893/) · [PDF](https://aclanthology.org/2026.acl-long.893.pdf) · [DOI 10.18653/v1/2026.acl-long.893](https://doi.org/10.18653/v1/2026.acl-long.893) · [OpenAlex](https://openalex.org/W7166820582)

**Citations:** 0 citations (OpenAlex, 2026-10-14) · [2](https://md.research.elamur.ai/free/en/papers/zhai-2026-maximizing-local-entropy-where-it/cited-by/) in our library, of the 149,439 papers whose bibliographies we read

## Summary
Unlearning methods for LLMs aim to erase private or copyrighted knowledge without retraining. Most of them, such as GA and NPO, push down likelihoods over every token of a response and across the whole vocabulary. This wastes compute and harms general ability. The authors propose PALU (Prefix-Aware Localized Unlearning), which applies entropy maximization only where it is needed.

PALU rests on two observations:
- **Temporal locality:** a short prefix of initiating tokens is what triggers the sensitive content, so disrupting that prefix is enough to derail generation. The remaining, content-agnostic tokens are kept under preservation constraints.
- **Vocabulary locality:** decoding is dominated by a few high-probability candidates, so PALU flattens only the top-K logits, pulling them toward the global mean of the model's logits rather than a uniform distribution.

This reduces cost to O(TK). On the TOFU benchmark with Llama-2-7B and Llama-3.1-8B, PALU is compared with GA, GD, DPO, NPO, SimNPO, PDU and TPO. It reports the best balance between forget quality and model utility, and it converges faster. This matters for practical, regulation-driven ("right to be forgotten") deletion of data from deployed models.

## Key points
- On Llama-2-7B (TOFU, 5% forget split), PALU's adaptive prefix reaches FQ 0.7126 and MU 0.6238. Fixed truncation ratios do worse: 25% gives FQ 0.1420, 50% gives FQ 0.6872 with MU 0.5951, and 100% gives FQ 0.6284 with MU 0.6086.
- As the target for the top-K logits, the global mean of the current model's logits worked better than uniform, Mean K or Mean ref targets.
- Training speed: PALU saturates FQ by epoch 5 and recovers MU within 2 epochs, while NPO needs roughly twice as many iterations.
- Membership-inference metrics (LOSS, ZLib, MinK, MinK++) on Llama-3.1-8B:
  - LOSS directional AUC falls to 0.1434, below the Retain model's 0.3620.
  - MinK++ AUC is 38.3% lower than TPO's.
- Limitations: the method has been tested only on text LLMs, not multimodal models. Code is available at github.com/nxZhai/PALU.

*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

- [National University of Singapore](https://md.research.elamur.ai/free/en/universities/national-university-of-singapore/) — University, Singapore, Singapore · 207 works · 39,576 citations
- [University of Science and Technology of China](https://md.research.elamur.ai/free/en/universities/university-of-science-and-technology-of-china/) — University, Hefei, China · 145 works · 114,409 citations

📚 Shelves (4): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works

🧭 Topics (1): [🧠 Language models & pretraining](https://md.research.elamur.ai/free/en/topics/language-models/) · 3,894 works

## Cite

Cite as: Naixin Zhai, Pengyang Shao, Binbin Zheng, Yonghui Yang, Fei Shen, Long Bai and Xun Yang. 2026. Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning. 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{zhai2026maximizing,
  title = {{Maximizing Local Entropy Where It Matters: Prefix-Aware Localized LLM Unlearning}},
  author = {Naixin Zhai, Pengyang Shao, Binbin Zheng, Yonghui Yang, Fei Shen, Long Bai and Xun Yang},
  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.893},
  url = {https://doi.org/10.18653/v1/2026.acl-long.893}
}
```

- [← previous MALicious INTent Dataset and Inoculating LLMs for Enhanced Disinformation Detection](https://md.research.elamur.ai/free/en/papers/modzelewski-2026-malicious-intent-dataset-inoculating-llms/)
- [next → Mechanisms of Prompt-Induced Hallucination in Vision–Language Models](https://md.research.elamur.ai/free/en/papers/rudman-2026-mechanisms-prompt-induced-hallucination-vision/)
- [↑ back to papers of 2026](https://md.research.elamur.ai/free/en/papers/2026/page/8/)

## References

54 references, 21 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Adam Berger, Stephen A Della Pietra, and Vincent J Della Pietra. 1996. [A maximum entropy approach to natural language processing](https://md.research.elamur.ai/free/en/papers/berger-1996-maximum-entropy-approach-natural-language/). Computational linguistics, 22(1):39–71.
2. Nicholas Carlini, Florian Tramer, Eric Wallace, Matthew Jagielski, Ariel Herbert-Voss, Katherine Lee, Adam Roberts, Tom Brown, Dawn Song, Ulfar Erlingsson, and 1 others. 2021. [Extracting training data from large language models](https://md.research.elamur.ai/free/en/papers/carlini-2020-extracting-training-data-large-language/). In 30th USENIX security symposium (USENIX Security 21), pages 2633–2650…
3. Chen-Hao Chao, Chien Feng, Wei-Fang Sun, Cheng-Kuang Lee, Simon See, and Chun-Yi Lee. 2024. Maximum entropy reinforcement learning via energy-based normalizing flow. Advances in Neural Information Processing Systems, 37:56136–56165. [find](https://scholar.google.com/scholar?q=Chen%2DHao+Chao%2C+Chien+Feng%2C+Wei%2DFang+Sun%2C+Cheng%2DKuang+Lee%2C+Simon+See%2C+and+Chun%2DYi+Lee%2E+2024%2E+Maximum+entropy+reinforcement+learning+via+energy%2Dbased+normalizing+flow%2E+Advances+in+Neural+Information+Pro)
4. Jiaao Chen and Diyi Yang. 2023. [Unlearn what you want to forget: Efficient unlearning for llms](https://md.research.elamur.ai/free/en/papers/chen-2023-unlearn-what-you-want-forget/). In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages 12041– 12052.
5. Daixuan Cheng, Shaohan Huang, Xuekai Zhu, Bo Dai, Xin Zhao, Zhenliang Zhang, and Furu Wei. 2026. [Reasoning with exploration: An entropy perspective](https://md.research.elamur.ai/free/en/papers/cheng-2025-reasoning-exploration-entropy-perspective/). In Proceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 30377–30385.
6. Vineeth Dorna, Anmol Mekala, Wenlong Zhao, Andrew McCallum, Zachary C Lipton, J Zico Kolter, and Pratyush Maini. 2025. OpenUnlearning: Accelerating LLM unlearning via unified benchmarking of methods and metrics. arXiv preprint arXiv:2506.12618. — [arXiv:2506.12618](https://arxiv.org/abs/2506.12618) [find](https://scholar.google.com/scholar?q=Vineeth+Dorna%2C+Anmol+Mekala%2C+Wenlong+Zhao%2C+Andrew+McCallum%2C+Zachary+C+Lipton%2C+J+Zico+Kolter%2C+and+Pratyush+Maini%2E+2025%2E+OpenUnlearning%3A+Accelerating+LLM+unlearning+via+unified+benchmarking+of+methods+a)
7. Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al-Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Amy Yang, Angela Fan, and 1 others. 2024. [The llama 3 herd of models](https://md.research.elamur.ai/free/en/papers/grattafiori-2024-llama-3-herd-models/). arXiv e-prints, pages arXiv–2407.
8. Taha Entesari, Arman Hatami, Rinat Khaziev, Anil Ramakrishna, and Mahyar Fazlyab. 2025. Constrained entropic unlearning: A primal-dual framework for large language models. In The Thirty-ninth Annual Conference on Neural Information Processing Systems. [find](https://scholar.google.com/scholar?q=Taha+Entesari%2C+Arman+Hatami%2C+Rinat+Khaziev%2C+Anil+Ramakrishna%2C+and+Mahyar+Fazlyab%2E+2025%2E+Constrained+entropic+unlearning%3A+A+primal%2Ddual+framework+for+large+language+models%2E+In+The+Thirty%2Dninth+Annual+C)
9. Chongyu Fan, Jiancheng Liu, Licong Lin, Jinghan Jia, Ruiqi Zhang, Song Mei, and Sijia Liu. 2025. Simplicity prevails: Rethinking negative preference optimization for llm unlearning. In Neurips Safe Generative AI Workshop 2024. [find](https://scholar.google.com/scholar?q=Chongyu+Fan%2C+Jiancheng+Liu%2C+Licong+Lin%2C+Jinghan+Jia%2C+Ruiqi+Zhang%2C+Song+Mei%2C+and+Sijia+Liu%2E+2025%2E+Simplicity+prevails%3A+Rethinking+negative+preference+optimization+for+llm+unlearning%2E+In+Neurips+Safe+Ge)
10. Qingkai Fang, Yan Zhou, Shoutao Guo, Shaolei Zhang, and Yang Feng. 2025. Llama-omni2: Llm-based real-time spoken chatbot with autoregressive streaming speech synthesis. arXiv preprint arXiv:2505.02625. — [arXiv:2505.02625](https://arxiv.org/abs/2505.02625) [find](https://scholar.google.com/scholar?q=Qingkai+Fang%2C+Yan+Zhou%2C+Shoutao+Guo%2C+Shaolei+Zhang%2C+and+Yang+Feng%2E+2025%2E+Llama%2Domni2%3A+Llm%2Dbased+real%2Dtime+spoken+chatbot+with+autoregressive+streaming+speech+synthesis%2E+arXiv+preprint+arXiv%3A2505%2E02625)
11. Xiaoliang Fu, Jiaye Lin, Yangyi Fang, Binbin Zheng, Chaowen Hu, Zekai Shao, Cong Qin, Lu Pan, Ke Zeng, and Xunliang Cai. 2026. [Maspo: Unifying gradient utilization, probability mass, and signal reliability for robust and sample-efficient llm reasoning](https://md.research.elamur.ai/free/en/papers/fu-2026-maspo-unifying-gradient-utilization-probability/). arXiv preprint arXiv:2602.17550.
12. Jun Gao, Di He, Xu Tan, Tao Qin, Liwei Wang, and Tie-Yan Liu. 2019. [Representation degeneration problem in training natural language generation models](https://md.research.elamur.ai/free/en/papers/gao-2019-representation-degeneration-problem-training-natural/). arXiv preprint arXiv:1907.12009.
13. Anmol Goel, Alan Ritter, and Iryna Gurevych. 2026. Auditing language model unlearning via information decomposition. arXiv preprint arXiv:2601.15111. [find](https://scholar.google.com/scholar?q=Anmol+Goel%2C+Alan+Ritter%2C+and+Iryna+Gurevych%2E+2026%2E+Auditing+language+model+unlearning+via+information+decomposition%2E+arXiv+preprint+arXiv%3A2601%2E15111%2E)
14. Zhiyuan Han, Beier Zhu, Yanlong Xu, Peipei Song, and Xun Yang. 2025. Benchmarking and bridging emotion conflicts for multimodal emotion reasoning. In Proceedings of the 33rd ACM International Conference on Multimedia, pages 5528–5537. [find](https://scholar.google.com/scholar?q=Zhiyuan+Han%2C+Beier+Zhu%2C+Yanlong+Xu%2C+Peipei+Song%2C+and+Xun+Yang%2E+2025%2E+Benchmarking+and+bridging+emotion+conflicts+for+multimodal+emotion+reasoning%2E+In+Proceedings+of+the+33rd+ACM+International+Conferen)
15. Ari Holtzman, Jan Buys, Li Du, Maxwell Forbes, and Yejin Choi. 2019. [The curious case of neural text degeneration](https://md.research.elamur.ai/free/en/papers/holtzman-2019-curious-case-neural-text-degeneration/). arXiv preprint arXiv:1904.09751.
16. Jinpeng Hu, Tengteng Dong, Gang Luo, Hui Ma, Peng Zou, Xiao Sun, Dan Guo, Xun Yang, and Meng Wang. 2024. Psycollm: Enhancing llm for psychological understanding and evaluation. IEEE Transactions on Computational Social Systems, 12(2):539–551. [find](https://scholar.google.com/scholar?q=Jinpeng+Hu%2C+Tengteng+Dong%2C+Gang+Luo%2C+Hui+Ma%2C+Peng+Zou%2C+Xiao+Sun%2C+Dan+Guo%2C+Xun+Yang%2C+and+Meng+Wang%2E+2024%2E+Psycollm%3A+Enhancing+llm+for+psychological+understanding+and+evaluation%2E+IEEE+Transactions+on+Co)
17. Edwin T Jaynes. 1957. Information theory and statistical mechanics. Physical review, 106(4):620. [find](https://scholar.google.com/scholar?q=Edwin+T+Jaynes%2E+1957%2E+Information+theory+and+statistical+mechanics%2E+Physical+review%2C+106%284%29%3A620%2E)
18. 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)
19. Jinghan Jia, Yihua Zhang, Yimeng Zhang, Jiancheng Liu, Bharat Runwal, James Diffenderfer, Bhavya Kailkhura, and Sijia Liu. 2024. [Soul: Unlocking the power of second-order optimization for llm unlearning](https://md.research.elamur.ai/free/en/papers/jia-2024-soul-unlocking-power-second-order/). In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages…
20. Yuxian Jiang, Yafu Li, Guanxu Chen, Dongrui Liu, Yu Cheng, and Jing Shao. 2025. Rethinking entropy regularization in large reasoning models. arXiv preprint arXiv:2509.25133. — [arXiv:2509.25133](https://arxiv.org/abs/2509.25133) [find](https://scholar.google.com/scholar?q=Yuxian+Jiang%2C+Yafu+Li%2C+Guanxu+Chen%2C+Dongrui+Liu%2C+Yu+Cheng%2C+and+Jing+Shao%2E+2025%2E+Rethinking+entropy+regularization+in+large+reasoning+models%2E+arXiv+preprint+arXiv%3A2509%2E25133%2E+%E2%80%94+%5BarXiv%3A2509%2E25133%5D%28https)
21. Antonia Karamolegkou, Jiaang Li, Li Zhou, and Anders Søgaard. 2023. [Copyright violations and large language models](https://md.research.elamur.ai/free/en/papers/karamolegkou-2023-copyright-violations-large-language-models/). arXiv preprint arXiv:2310.13771.
22. Hang Li, Tianlong Xu, Kaiqi Yang, Yucheng Chu, Yanling Chen, Yichi Song, Qingsong Wen, and Hui Liu. 2025. Ask-before-detection: Identifying and mitigating conformity bias in llm-powered error detector for math word problem solutions. In Proceedings of the 63rd Annual Meeting of the Association for C… [find](https://scholar.google.com/scholar?q=Hang+Li%2C+Tianlong+Xu%2C+Kaiqi+Yang%2C+Yucheng+Chu%2C+Yanling+Chen%2C+Yichi+Song%2C+Qingsong+Wen%2C+and+Hui+Liu%2E+2025%2E+Ask%2Dbefore%2Ddetection%3A+Identifying+and+mitigating+conformity+bias+in+llm%2Dpowered+error+detector)
23. Jilong Liu, Pengyang Shao, Wei Qin, Fei Liu, Yonghui Yang, and Richang Hong. 2026. Debate over mixed-knowledge: A robust multi-agent reasoning framework for incomplete knowledge graph question answering. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 15333– 15341. [find](https://scholar.google.com/scholar?q=Jilong+Liu%2C+Pengyang+Shao%2C+Wei+Qin%2C+Fei+Liu%2C+Yonghui+Yang%2C+and+Richang+Hong%2E+2026%2E+Debate+over+mixed%2Dknowledge%3A+A+robust+multi%2Dagent+reasoning+framework+for+incomplete+knowledge+graph+question+answeri)
24. Ruixuan Liu, Li Xiong, and 1 others. 2025a. Direct token optimization: A self-contained approach to large language model unlearning. arXiv preprint arXiv:2510.00125. [find](https://scholar.google.com/scholar?q=Ruixuan+Liu%2C+Li+Xiong%2C+and+1+others%2E+2025a%2E+Direct+token+optimization%3A+A+self%2Dcontained+approach+to+large+language+model+unlearning%2E+arXiv+preprint+arXiv%3A2510%2E00125%2E)
25. Sijia Liu, Yuanshun Yao, Jinghan Jia, Stephen Casper, Nathalie Baracaldo, Peter Hase, Yuguang Yao, Chris Yuhao Liu, Xiaojun Xu, Hang Li, and 1 others. 2025b. Rethinking machine unlearning for large language models. Nature Machine Intelligence, pages 1–14. [find](https://scholar.google.com/scholar?q=Sijia+Liu%2C+Yuanshun+Yao%2C+Jinghan+Jia%2C+Stephen+Casper%2C+Nathalie+Baracaldo%2C+Peter+Hase%2C+Yuguang+Yao%2C+Chris+Yuhao+Liu%2C+Xiaojun+Xu%2C+Hang+Li%2C+and+1+others%2E+2025b%2E+Rethinking+machine+unlearning+for+large+la)
26. Ziyang Luo, Kaixin Li, Hongzhan Lin, Yuchen Tian, Mohan Kankanhalli, and Jing Ma. 2025. Tree-of-evolution: Tree-structured instruction evolution for code generation in large language models. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Pa… [find](https://scholar.google.com/scholar?q=Ziyang+Luo%2C+Kaixin+Li%2C+Hongzhan+Lin%2C+Yuchen+Tian%2C+Mohan+Kankanhalli%2C+and+Jing+Ma%2E+2025%2E+Tree%2Dof%2Devolution%3A+Tree%2Dstructured+instruction+evolution+for+code+generation+in+large+language+models%2E+In+Procee)
27. Pratyush Maini, Zhili Feng, Avi Schwarzschild, Zachary Chase Lipton, and J Zico Kolter. 2024. [TOFU: A task of fictitious unlearning for LLMs](https://md.research.elamur.ai/free/en/papers/maini-2024-tofu-task-fictitious-unlearning-llms/). In First Conference on Language Modeling.
28. Anmol Reddy Mekala, Vineeth Dorna, Shreya Dubey, Abhishek Lalwani, David Koleczek, Mukund Rungta, Sadid A Hasan, and Elita AA Lobo. 2025. [Alternate preference optimization for unlearning factual knowledge in large language models](https://md.research.elamur.ai/free/en/papers/mekala-2025-alternate-preference-optimization-unlearning-factual/). In Proceedings of the 31st International Conference on Computational…
29. Eric Michaud, Ziming Liu, Uzay Girit, and Max Tegmark. 2023. [The quantization model of neural scaling](https://md.research.elamur.ai/free/en/papers/michaud-2023-quantization-model-neural-scaling/). Advances in Neural Information Processing Systems, 36:28699–28722.
30. Haowen Pan, Yixin Cao, Xiaozhi Wang, Xun Yang, and Meng Wang. 2024. Finding and editing multi-modal neurons in pre-trained transformers. In Findings of the Association for Computational Linguistics: ACL 2024, pages 1012–1037. [find](https://scholar.google.com/scholar?q=Haowen+Pan%2C+Yixin+Cao%2C+Xiaozhi+Wang%2C+Xun+Yang%2C+and+Meng+Wang%2E+2024%2E+Finding+and+editing+multi%2Dmodal+neurons+in+pre%2Dtrained+transformers%2E+In+Findings+of+the+Association+for+Computational+Linguistics%3A+A)
31. Haowen Pan, Xiaozhi Wang, Yixin Cao, Zenglin Shi, Xun Yang, Juanzi Li, and Meng Wang. 2025. Precise localization of memories: A fine-grained neuron-level knowledge editing technique for LLMs. In The Thirteenth International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Haowen+Pan%2C+Xiaozhi+Wang%2C+Yixin+Cao%2C+Zenglin+Shi%2C+Xun+Yang%2C+Juanzi+Li%2C+and+Meng+Wang%2E+2025%2E+Precise+localization+of+memories%3A+A+fine%2Dgrained+neuron%2Dlevel+knowledge+editing+technique+for+LLMs%2E+In+The+T)
32. Rafael Rafailov, Archit Sharma, Eric Mitchell, Christopher D Manning, Stefano Ermon, and Chelsea Finn. 2023. [Direct preference optimization: Your language model is secretly a reward model](https://md.research.elamur.ai/free/en/papers/rafailov-2023-direct-preference-optimization-language-model/). Advances in neural information processing systems, 36:53728–53741.
33. Victor Sanh, Lysandre Debut, Julien Chaumond, and Thomas Wolf. 2019. [Distilbert, a distilled version of bert: smaller, faster, cheaper and lighter](https://md.research.elamur.ai/free/en/papers/sanh-2019-distilbert-distilled-version-bert-smaller/). arXiv preprint arXiv:1910.01108.
34. Pengyang Shao, Naixin Zhai, Lei Chen, Yonghui Yang, Fengbin Zhu, Xun Yang, and Meng Wang. 2026. Baldro: A distributionally robust optimization based framework for large language model unlearning. In Proceedings of the ACM Web Conference 2026, WWW ’26, page 8874–8884. Association for Computing Machin… [find](https://scholar.google.com/scholar?q=Pengyang+Shao%2C+Naixin+Zhai%2C+Lei+Chen%2C+Yonghui+Yang%2C+Fengbin+Zhu%2C+Xun+Yang%2C+and+Meng+Wang%2E+2026%2E+Baldro%3A+A+distributionally+robust+optimization+based+framework+for+large+language+model+unlearning%2E+In+P)
35. Weijia Shi, Anirudh Ajith, Mengzhou Xia, Yangsibo Huang, Daogao Liu, Terra Blevins, Danqi Chen, and Luke Zettlemoyer. 2023. Detecting pretraining data from large language models. arXiv preprint arXiv:2310.16789. [find](https://scholar.google.com/scholar?q=Weijia+Shi%2C+Anirudh+Ajith%2C+Mengzhou+Xia%2C+Yangsibo+Huang%2C+Daogao+Liu%2C+Terra+Blevins%2C+Danqi+Chen%2C+and+Luke+Zettlemoyer%2E+2023%2E+Detecting+pretraining+data+from+large+language+models%2E+arXiv+preprint+arXiv%3A)
36. 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. [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+Smith%2C+and+Chiyuan+Zhang%2E+2025%2E+Muse%3A+Machine+unlearning+six%2Dway+evaluation+for)
37. Nianwen Si, Hao Zhang, Heyu Chang, Wenlin Zhang, Dan Qu, and Weiqiang Zhang. 2023. Knowledge unlearning for llms: Tasks, methods, and challenges. arXiv preprint arXiv:2311.15766. — [arXiv:2311.15766](https://arxiv.org/abs/2311.15766) [find](https://scholar.google.com/scholar?q=Nianwen+Si%2C+Hao+Zhang%2C+Heyu+Chang%2C+Wenlin+Zhang%2C+Dan+Qu%2C+and+Weiqiang+Zhang%2E+2023%2E+Knowledge+unlearning+for+llms%3A+Tasks%2C+methods%2C+and+challenges%2E+arXiv+preprint+arXiv%3A2311%2E15766%2E+%E2%80%94+%5BarXiv%3A2311%2E15766%5D%28)
38. Kushal Tirumala, Aram Markosyan, Luke Zettlemoyer, and Armen Aghajanyan. 2022. [Memorization without overfitting: Analyzing the training dynamics of large language models](https://md.research.elamur.ai/free/en/papers/tirumala-2022-memorization-overfitting-analyzing-training-dynamics/). Advances in Neural Information Processing Systems, 35:38274–38290.
39. Hugo Touvron, Louis Martin, Kevin Stone, Peter Albert, Amjad Almahairi, Yasmine Babaei, Nikolay Bashlykov, Soumya Batra, Prajjwal Bhargava, Shruti Bhosale, and 1 others. 2023. [Llama 2: Open foundation and fine-tuned chat models](https://md.research.elamur.ai/free/en/papers/touvron-2023-llama-2-open-foundation-fine/). arXiv preprint arXiv:2307.09288.
40. Yixin Wan, Anil Ramakrishna, Kai-Wei Chang, Volkan Cevher, and Rahul Gupta. 2025. Not every token needs forgetting: Selective unlearning balancing forgetting and utility in large language models. In Findings of the Association for Computational Linguistics: EMNLP 2025, pages 1827–1835, Suzhou, China… [find](https://scholar.google.com/scholar?q=Yixin+Wan%2C+Anil+Ramakrishna%2C+Kai%2DWei+Chang%2C+Volkan+Cevher%2C+and+Rahul+Gupta%2E+2025%2E+Not+every+token+needs+forgetting%3A+Selective+unlearning+balancing+forgetting+and+utility+in+large+language+models%2E+In+F)
41. Lingzhi Wang, Xingshan Zeng, Jinsong Guo, Kam-Fai Wong, and Georg Gottlob. 2025a. Selective forgetting: Advancing machine unlearning techniques and evaluation in language models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, pages 843–851. [find](https://scholar.google.com/scholar?q=Lingzhi+Wang%2C+Xingshan+Zeng%2C+Jinsong+Guo%2C+Kam%2DFai+Wong%2C+and+Georg+Gottlob%2E+2025a%2E+Selective+forgetting%3A+Advancing+machine+unlearning+techniques+and+evaluation+in+language+models%2E+In+Proceedings+of+the)
42. Xinyi Wang, Xun Yang, Yanlong Xu, Yuchen Wu, Zhen Li, and Na Zhao. 2025b. Affordbot: 3d fine-grained embodied reasoning via multimodal large language models. In The Thirty-ninth Annual Conference on Neural Information Processing Systems. [find](https://scholar.google.com/scholar?q=Xinyi+Wang%2C+Xun+Yang%2C+Yanlong+Xu%2C+Yuchen+Wu%2C+Zhen+Li%2C+and+Na+Zhao%2E+2025b%2E+Affordbot%3A+3d+fine%2Dgrained+embodied+reasoning+via+multimodal+large+language+models%2E+In+The+Thirty%2Dninth+Annual+Conference+on+N)
43. Yangyang Xu, Jinpeng Hu, Zhuoer Zhao, Zhangling Duan, Xiao Sun, and Xun Yang. 2025. Multiagen-tesc: A llm-based multi-agent collaboration framework for emotional support conversation. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 4665–4681. [find](https://scholar.google.com/scholar?q=Yangyang+Xu%2C+Jinpeng+Hu%2C+Zhuoer+Zhao%2C+Zhangling+Duan%2C+Xiao+Sun%2C+and+Xun+Yang%2E+2025%2E+Multiagen%2Dtesc%3A+A+llm%2Dbased+multi%2Dagent+collaboration+framework+for+emotional+support+conversation%2E+In+Proceedings+o)
44. Puning Yang, Qizhou Wang, Zhuo Huang, Tongliang Liu, Chengqi Zhang, and Bo Han. 2025. Exploring criteria of loss reweighting to enhance llm unlearning. In Forty-second International Conference on Machine Learning. [find](https://scholar.google.com/scholar?q=Puning+Yang%2C+Qizhou+Wang%2C+Zhuo+Huang%2C+Tongliang+Liu%2C+Chengqi+Zhang%2C+and+Bo+Han%2E+2025%2E+Exploring+criteria+of+loss+reweighting+to+enhance+llm+unlearning%2E+In+Forty%2Dsecond+International+Conference+on+Mach)
45. 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.
46. Samuel Yeom, Irene Giacomelli, Matt Fredrikson, and Somesh Jha. 2018. [Privacy risk in machine learning: Analyzing the connection to overfitting](https://md.research.elamur.ai/free/en/papers/yeom-2017-privacy-risk-machine-learning-analyzing/). In 2018 IEEE 31st computer security foundations symposium (CSF), pages 268–282. IEEE.
47. Dan Zhang, Min Cai, Jonathan Light, Ziniu Hu, Yisong Yue, and Jie Tang. 2025a. Tdrm: Smooth reward models with temporal difference for llm rl and inference. arXiv preprint arXiv:2509.15110. — [arXiv:2509.15110](https://arxiv.org/abs/2509.15110) [find](https://scholar.google.com/scholar?q=Dan+Zhang%2C+Min+Cai%2C+Jonathan+Light%2C+Ziniu+Hu%2C+Yisong+Yue%2C+and+Jie+Tang%2E+2025a%2E+Tdrm%3A+Smooth+reward+models+with+temporal+difference+for+llm+rl+and+inference%2E+arXiv+preprint+arXiv%3A2509%2E15110%2E+%E2%80%94+%5BarXiv%3A2)
48. Jingyang Zhang, Jingwei Sun, Eric Yeats, Yang Ouyang, Martin Kuo, Jianyi Zhang, Hao Frank Yang, and Hai Li. 2025b. Min-k%++: Improved baseline for pre-training data detection from large language models. In The Thirteenth International Conference on Learning Representations. [find](https://scholar.google.com/scholar?q=Jingyang+Zhang%2C+Jingwei+Sun%2C+Eric+Yeats%2C+Yang+Ouyang%2C+Martin+Kuo%2C+Jianyi+Zhang%2C+Hao+Frank+Yang%2C+and+Hai+Li%2E+2025b%2E+Min%2Dk%25%2B%2B%3A+Improved+baseline+for+pre%2Dtraining+data+detection+from+large+language+model)
49. 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.
50. Mengnan Zhao, Lihe Zhang, Xingyi Yang, Tianhang Zheng, and Baocai Yin. 2024a. Advanchor: Enhancing diffusion model unlearning with adversarial anchors. arXiv preprint arXiv:2501.00054. — [arXiv:2501.00054](https://arxiv.org/abs/2501.00054) [find](https://scholar.google.com/scholar?q=Mengnan+Zhao%2C+Lihe+Zhang%2C+Xingyi+Yang%2C+Tianhang+Zheng%2C+and+Baocai+Yin%2E+2024a%2E+Advanchor%3A+Enhancing+diffusion+model+unlearning+with+adversarial+anchors%2E+arXiv+preprint+arXiv%3A2501%2E00054%2E+%E2%80%94+%5BarXiv%3A2501%2E0)
51. Mengnan Zhao, Lihe Zhang, Tianhang Zheng, Yuqiu Kong, and Baocai Yin. 2024b. Separable multi-concept erasure from diffusion models. arXiv preprint arXiv:2402.05947. — [arXiv:2402.05947](https://arxiv.org/abs/2402.05947) [find](https://scholar.google.com/scholar?q=Mengnan+Zhao%2C+Lihe+Zhang%2C+Tianhang+Zheng%2C+Yuqiu+Kong%2C+and+Baocai+Yin%2E+2024b%2E+Separable+multi%2Dconcept+erasure+from+diffusion+models%2E+arXiv+preprint+arXiv%3A2402%2E05947%2E+%E2%80%94+%5BarXiv%3A2402%2E05947%5D%28https%3A%2F%2Farxiv%2E)
52. Weixiang Zhao, Yulin Hu, Yang Deng, Tongtong Wu, Wenxuan Zhang, Jiahe Guo, An Zhang, Yanyan Zhao, Bing Qin, Tat-Seng Chua, and 1 others. 2025. [Mpo: Multilingual safety alignment via reward gap optimization](https://md.research.elamur.ai/free/en/papers/zhao-2025-mpo-multilingual-safety-alignment-reward/). In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (V…
53. Xiangyu Zhou, Yao Qiang, Saleh Zare Zade, Douglas Zytko, Prashant Khanduri, and Dongxiao Zhu. 2026. Not all tokens are meant to be forgotten. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 38173–38182. [find](https://scholar.google.com/scholar?q=Xiangyu+Zhou%2C+Yao+Qiang%2C+Saleh+Zare+Zade%2C+Douglas+Zytko%2C+Prashant+Khanduri%2C+and+Dongxiao+Zhu%2E+2026%2E+Not+all+tokens+are+meant+to+be+forgotten%2E+In+Proceedings+of+the+AAAI+Conference+on+Artificial+Intell)
54. Xingyu Zhu, Beier Zhu, Shuo Wang, Junfeng Fang, Kesen Zhao, Hanwang Zhang, and Xiangnan He. 2026. Principled steering via null-space projection for jailbreak defense in vision-language models. arXiv preprint arXiv:2603.22094. — [arXiv:2603.22094](https://arxiv.org/abs/2603.22094) [find](https://scholar.google.com/scholar?q=Xingyu+Zhu%2C+Beier+Zhu%2C+Shuo+Wang%2C+Junfeng+Fang%2C+Kesen+Zhao%2C+Hanwang+Zhang%2C+and+Xiangnan+He%2E+2026%2E+Principled+steering+via+null%2Dspace+projection+for+jailbreak+defense+in+vision%2Dlanguage+models%2E+arXiv+p)

*Updated 2 November 2026*