Elamur Research⌕
☰
home / papers / 2026 / From Fake to Real: Mitigating Out-of-Distribution Bias in In-Context Learning via Feedback Supervision from Large Language Models

From Fake to Real: Mitigating Out-of-Distribution Bias in In-Context Learning via Feedback Supervision from Large Language Models

Paper · 2026 · Rui Song, Yingji Li, Jian Li et al.

Summary

In-context learning (ICL) relies on labeled demonstrations, usually taken from a source dataset. When test data comes from a very different distribution, these demonstrations can mislead the model. In such out-of-distribution (OOD) settings, similarity-based example selection can even do worse than random choice. The authors propose LPL, which avoids source demonstrations altogether and instead uses unlabeled target-domain samples that the LLM labels itself.

LPL works in three stages:

  • Initial pool: a seed set of pseudo-labeled target samples is chosen with CPP, a filter that combines model confidence with perplexity under perturbation.
  • Iterative relabeling: the pool grows over several rounds of semantic retrieval and relabeling.
  • Majority voting: predictions from all rounds are combined by vote.

A theoretical argument based on error bounds supports the approach. It suggests that demonstrations from the right distribution with mild label noise do less harm than clean labels from a strongly shifted distribution.

LPL was tested on four task types with open models and API models. It generally beats the retrieval baselines, which suggests that low-noise self-labeling is a cheap way to make ICL more robust to domain shift.

Key points

  • Setup and models: the source context is Amazon reviews. The four tasks and their target datasets are:

    • sentiment analysis (SA): dynasent, semeval, sst
    • toxicity detection (TD): adv, implicit, toxigen
    • NLI: anli, contr, wanli
    • NER: conll, ener, wnut

    Models tested are Qwen2.5-3B, LLaMA3.2-3B, GPT-3.5-turbo and GPT-4o-mini.

  • Results on GPT-3.5-turbo: compared with the KNN baseline, LPL raises average scores on every task:

    TaskKNNLPL
    SA71.4074.60
    TD74.2075.30
    NLI50.5354.80
    NER44.6748.87

    The other API model, GPT-4o-mini, also improves on every task.

  • Ablation: dropping either the confidence or the perplexity component of CPP lowers performance. Removing confidence causes the larger drop.

  • Thresholds and voting: both filtering thresholds (τ, γ) are fixed at 0.8, which was near-optimal in almost all settings. Majority voting stabilizes results and sometimes beats every single iteration.

  • Limitations and code: the iterations add compute cost, closed-source models need proxy labelers, and gains on adversarial data are small. Code is available at github.com/songruiecho/LPL.

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

🏢TencentShenzhen · China134 works · 19,743 citations🎓University of TrentoTrento · Italy85 works · 15,524 citations🎓Jilin UniversityChangchun · China25 works · 5,252 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.Language models & pretraining · 3,320 works

Cite

Cite as Rui Song, Yingji Li, Jian Li, Fausto Giunchiglia and Hao Xu. 2026. From Fake to Real: Mitigating Out-of-Distribution Bias in In-Context Learning via Feedback Supervision from Large Language Models. Findings of the Association for Computational Linguistics: ACL 2026.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.findings-acl.747

@inproceedings{song2026fake,
  title = {{From Fake to Real: Mitigating Out-of-Distribution Bias in In-Context Learning via Feedback Supervision from Large Language Models}},
  author = {Rui Song, Yingji Li, Jian Li, Fausto Giunchiglia and Hao Xu},
  booktitle = {Findings of the Association for Computational Linguistics: ACL 2026},
  year = {2026},
  doi = {10.18653/v1/2026.findings-acl.747},
  url = {https://doi.org/10.18653/v1/2026.findings-acl.747}
}

References

55 references, 13 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. 2024. Qwen2 technical report. find
  2. AI@Meta. 2024. Llama 3 model card. find
  3. Eric Arazo, Diego Ortego, Paul Albert, Noel E. O’Connor, and Kevin McGuinness. 2020. Pseudo-labeling and confirmation bias in deep semi-supervised learning. In 2020 International Joint Conference on Neural Networks, IJCNN 2020, Glasgow, United Kingdom, July 19-24, 2020, pages 1–8. find
  4. Udit Arora, William Huang, and He He. 2021. Types of out-of-distribution texts and how to detect them. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, EMNLP 2021, Virtual Event / Punta Cana, Dominican Republic, 7-11 November, 2021, pages 10687–10701. find
  5. Hadi Askari, Shivanshu Gupta, Terry Tong, Fei Wang, Anshuman Chhabra, and Muhao Chen. 2025. Unraveling indirect in-context learning using influence functions. CoRR, abs/2501.01473. — [arXiv:2501.01473](https://arxiv.org/abs/2501.01473) find
  6. Shai Ben-David, John Blitzer, Koby Crammer, Alex Kulesza, Fernando Pereira, and Jennifer Wortman Vaughan. 2010. A theory of learning from different domains. Mach. Learn., 79(1-2):151–175. find
  7. Paola Cascante-Bonilla, Fuwen Tan, Yanjun Qi, and Vicente Ordonez. 2021. Curriculum labeling: Revisiting pseudo-labeling for semi-supervised learning. In Thirty-Fifth AAAI Conference on Artificial Intelligence, 2021, pages 6912–6920. find
  8. Yen-Chi Chen. 2017. A tutorial on kernel density estimation and recent advances. Biostatistics & Epidemiology, 1(1):161–187. find
  9. Yu Fei, Yifan Hou, Zeming Chen, and Antoine Bosselut. 2023. Mitigating label biases for in-context learning. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2023, Toronto, Canada, July 9-14, 2023, pages 14014–14031.
  10. Hongfu Gao, Feipeng Zhang, Wenyu Jiang, Jun Shu, Feng Zheng, and Hongxin Wei. 2024. On the noise robustness of in-context learning for text generation. In Advances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, B… find
  11. Tianyu Gao, Xingcheng Yao, and Danqi Chen. 2021. Simcse: Simple contrastive learning of sentence embeddings. In Proceedings of the 2021 Conference on Empirical Methods in Natural Language Processing, EMNLP 2021, Virtual Event / Punta Cana, Dominican Republic, 7-11 November, 2021, pages 6894– 6910.
  12. Hila Gonen, Srini Iyer, Terra Blevins, Noah A. Smith, and Luke Zettlemoyer. 2023. Demystifying prompts in language models via perplexity estimation. In Findings of the Association for Computational Linguistics: EMNLP 2023, Singapore, December 6-10, 2023, pages 10136–10148. find
  13. Shuoran Jiang, Qingcai Chen, Yang Xiang, Youcheng Pan, and Yukang Lin. 2024. Linguistic rule induction improves adversarial and OOD robustness in large language models. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COL… find
  14. Damai Dai, Yutao Sun, Li Dong, Yaru Hao, Shuming Ma, Zhifang Sui, and Furu Wei. 2023. Why can gpt learn in-context? language models secretly perform gradient descent as meta-optimizers. In Findings of the Association for Computational Linguistics: ACL 2023, pages 4005–4019.
  15. Janak Kapuriya, Manit Kaushik, Debasis Ganguly, and Sumit Bhatia. 2025. Exploring the role of diversity in example selection for in-context learning. In Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR 2025, Padua, Italy, July 13-… find
  16. Yuyang Ding, Dan Qiao, Juntao Li, Jiajie Xu, Pingfu Chao, Xiaofang Zhou, and Min Zhang. 2025. Towards ds-ner: Unveiling and addressing latent noise in distant annotations. IEEE Transactions on Knowledge and Data Engineering. find
  17. Dong-Hyun Lee and 1 others. 2013. Pseudo-label: The simple and efficient semi-supervised learning method for deep neural networks. In Workshop on challenges in representation learning, ICML, volume 3, page 896. Atlanta. find
  18. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Jingyuan Ma, Rui Li, Heming Xia, Jingjing Xu, Zhiyong Wu, Baobao Chang, Xu Sun, and Zhifang Sui. 2024. A survey on in-context learning. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024, Miami, FL, USA, No…
  19. Ming Li, Qingli Li, and Yan Wang. 2023. Class balanced adaptive pseudo labeling for federated semi-supervised learning. In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2023, Vancouver, BC, Canada, June 17-24, 2023, pages 16292–16301. find
  20. Qingxiu Dong, Lei Li, Damai Dai, Ce Zheng, Zhiyong Wu, Baobao Chang, Xu Sun, Jingjing Xu, and Zhifang Sui. 2022. A survey on in-context learning. arXiv preprint arXiv:2301.00234.
  21. Xiaonan Li and Xipeng Qiu. 2023. Finding support examples for in-context learning. In Findings of the Association for Computational Linguistics: EMNLP 2023, Singapore, December 6-10, 2023, pages 6219– 6235.
  22. Jiachang Liu, Dinghan Shen, Yizhe Zhang, Bill Dolan, Lawrence Carin, and Weizhu Chen. 2022. What makes good in-context examples for gpt-3? In Proceedings of Deep Learning Inside Out: The 3rd Workshop on Knowledge Extraction and Integration for Deep Learning Architectures, DeeLIO@ACL 2022, pages 100–…
  23. Man Luo, Xin Xu, Zhuyun Dai, Panupong Pasupat, Seyed Mehran Kazemi, Chitta Baral, Vaiva Imbrasaite, and Vincent Y. Zhao. 2023. Dr.icl: Demonstration-retrieved in-context learning. CoRR, abs/2305.14128. — [arXiv:2305.14128](https://arxiv.org/abs/2305.14128) find
  24. Manas Madine. 2024. Bridging distribution gap via semantic rewriting with llms to enhance OOD robustness. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics, ACL 2024 - Student Research Workshop, Bangkok, Thailand, August 11-16, 2024, pages 458–468. find
  25. Tai Nguyen and Eric Wong. 2023. In-context example selection with influences. CoRR, abs/2302.11042. — [arXiv:2302.11042](https://arxiv.org/abs/2302.11042) find
  26. Jianmo Ni, Chen Qu, Jing Lu, Zhuyun Dai, Gustavo Hernández Ábrego, Ji Ma, Vincent Y. Zhao, Yi Luan, Keith B. Hall, Ming-Wei Chang, and Yinfei Yang. 2022. Large dual encoders are generalizable retrievers. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP…
  27. Kyle O’Brien, Nathan Ng, Isha Puri, Jorge Mendez, Hamid Palangi, Yoon Kim, Marzyeh Ghassemi, and Thomas Hartvigsen. 2024. Improving black-box robustness with in-context rewriting. CoRR, abs/2402.08225. find
  28. Keqin Peng, Liang Ding, Yancheng Yuan, Xuebo Liu, Min Zhang, Yuanxin Ouyang, and Dacheng Tao. 2024a. Revisiting demonstration selection strategies in in-context learning. CoRR, abs/2401.12087. find
  29. Keqin Peng, Liang Ding, Yancheng Yuan, Xuebo Liu, Min Zhang, Yuanxin Ouyang, and Dacheng Tao. 2024b. Revisiting demonstration selection strategies in in-context learning. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2024, Ban… find
  30. Chengwei Qin, Aston Zhang, Anirudh Dagar, and Wenming Ye. 2023. In-context learning with iterative demonstration selection. CoRR, abs/2310.09881. find
  31. Alec Radford, Jeffrey Wu, Rewon Child, David Luan, Dario Amodei, Ilya Sutskever, and 1 others. 2019. Language models are unsupervised multitask learners. OpenAI blog, 1(8):9. find
  32. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J. Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. J. Mach. Learn. Res., 21:140:1–140:67.
  33. Mamshad Nayeem Rizve, Kevin Duarte, Yogesh S. Rawat, and Mubarak Shah. 2021. In defense of pseudo-labeling: An uncertainty-aware pseudo-label selection framework for semi-supervised learning. In 9th International Conference on Learning Representations, ICLR 2021, Virtual Event, Austria, May 3-7, 202… find
  34. Vinay M. S., Minh-Hao Van, and Xintao Wu. 2024. In-context learning demonstration selection via influence analysis. CoRR, abs/2402.11750. — [arXiv:2402.11750](https://arxiv.org/abs/2402.11750) find
  35. Shai Shalev-Shwartz and Shai Ben-David. 2014. Understanding machine learning: From theory to algorithms. Cambridge university press. find
  36. Weiwei Shi, Yihong Gong, Chris Ding, Zhiheng Ma, Xiaoyu Tao, and Nanning Zheng. 2018. Transductive semi-supervised deep learning using min-max features. In Computer Vision - ECCV 2018 - 15th European Conference, Munich, Germany, September 8-14, 2018, Proceedings, Part V, volume 11209, pages 311–327. find
  37. Rui Song, Fausto Giunchiglia, Yingji Li, Mingjie Tian, and Hao Xu. 2024a. TACIT: A target-agnostic feature disentanglement framework for cross-domain text classification. In Thirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024, Thirty-Sixth Conference on Innovative Applications of Art… find
  38. Rui Song, Yingji Li, Lida Shi, Fausto Giunchiglia, and Hao Xu. 2024b. Shortcut learning in in-context learning: A survey. arXiv preprint arXiv:2411.02018. — [arXiv:2411.02018](https://arxiv.org/abs/2411.02018) find
  39. Hongjin Su, Jungo Kasai, Chen Henry Wu, Weijia Shi, Tianlu Wang, Jiayi Xin, Rui Zhang, Mari Ostendorf, Luke Zettlemoyer, Noah A. Smith, and Tao Yu. 2023. Selective annotation makes language models better few-shot learners. In The Eleventh International Conference on Learning Representations, ICLR 20… find
  40. Tianxiang Sun, Yunfan Shao, Hong Qian, Xuanjing Huang, and Xipeng Qiu. 2022. Black-box tuning for language-model-as-a-service. In International Conference on Machine Learning, ICML 2022, 17-23 July 2022, Baltimore, Maryland, USA, volume 162 of Proceedings of Machine Learning Research, pages 20841–20… find
  41. Ruixiang Tang, Dehan Kong, Longtao Huang, and Hui Xue. 2023. Large language models can be lazy learners: Analyze shortcuts in in-context learning. In Findings of the Association for Computational Linguistics: ACL 2023, Toronto, Canada, July 9-14, 2023, pages 4645–4657. find
  42. Guo-Hua Wang and Jianxin Wu. 2020. Repetitive repre-diction deep decipher for semi-supervised learning. In The Thirty-Fourth AAAI Conference on Artificial Intelligence, AAAI 2020, The Thirty-Second Innovative Applications of Artificial Intelligence Conference, IAAI 2020, The Tenth AAAI Symposium on… find
  43. Jindong Wang, Xixu Hu, Wenxin Hou, Hao Chen, Runkai Zheng, Yidong Wang, Linyi Yang, Wei Ye, Haojun Huang, Xiubo Geng, Binxing Jiao, Yue Zhang, and Xing Xie. 2024. On the robustness of chatgpt: An adversarial and out-of-distribution perspective. IEEE Data Eng. Bull., 48(1):48–62. find
  44. Hui Wu and Xiaodong Shi. 2022. Adversarial soft prompt tuning for cross-domain sentiment analysis. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2022, Dublin, Ireland, May 22-27, 2022, pages 2438–2447. find
  45. Zhiyong Wu, Yaoxiang Wang, Jiacheng Ye, and Lingpeng Kong. 2023. Self-adaptive in-context learning: An information compression perspective for in-context example selection and ordering. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers)… find
  46. Ming-Kun Xie, Jiahao Xiao, Hao-Zhe Liu, Gang Niu, Masashi Sugiyama, and Sheng-Jun Huang. 2024. Class-distribution-aware pseudo-labeling for semi-supervised multi-label learning. Advances in Neural Information Processing Systems, 36. find
  47. Tingyu Xie, Qi Li, Jian Zhang, Yan Zhang, Zuozhu Liu, and Hongwei Wang. 2023. Empirical study of zero-shot NER with chatgpt. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023, Singapore, December 6-10, 2023, pages 7935–7956. find
  48. Fangzhi Xu, Qika Lin, Jiawei Han, Tianzhe Zhao, Jun Liu, and Erik Cambria. 2025. Are large language models really good logical reasoners? a comprehensive evaluation and beyond. IEEE Transactions on Knowledge and Data Engineering.
  49. Shangqing Xu and Chao Zhang. 2024. Misconfidence-based demonstration selection for LLM in-context learning. CoRR, abs/2401.06301. — [arXiv:2401.06301](https://arxiv.org/abs/2401.06301) find
  50. David Yarowsky. 1995. Unsupervised word sense disambiguation rivaling supervised methods. In 33rd Annual Meeting of the Association for Computational Linguistics, 26-30 June 1995, MIT, Cambridge, Massachusetts, USA, Proceedings, pages 189–196.
  51. Honolulu, Hawaii, USA, volume 202 of Proceedings of Machine Learning Research, pages 39818–39833. PMLR. find
  52. Lang Yu, Qin Chen, Jiaju Lin, and Liang He. 2023. Black-box prompt tuning for vision-language model as a service. In Proceedings of the Thirty-Second International Joint Conference on Artificial Intelligence, IJCAI 2023, 19th-25th August 2023, Macao, SAR, China, pages 1686–1694. ijcai.org. find
  53. Lifan Yuan, Yangyi Chen, Ganqu Cui, Hongcheng Gao, Fangyuan Zou, Xingyi Cheng, Heng Ji, Zhiyuan Liu, and Maosong Sun. 2023. Revisiting out-of-distribution robustness in NLP: benchmarks, analysis, and llms evaluations. In Advances in Neural Information Processing Systems 36: Annual Conference on Neur… find
  54. Yiming Zhang, Shi Feng, and Chenhao Tan. 2022. Active example selection for in-context learning. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, EMNLP 2022, Abu Dhabi, United Arab Emirates, December 7-11, 2022, pages 9134–9148.
  55. Zihao Zhao, Eric Wallace, Shi Feng, Dan Klein, and Sameer Singh. 2021. Calibrate before use: Improving few-shot performance of language models. In Proceedings of the 38th International Conference on Machine Learning, ICML 2021, 18-24 July 2021, Virtual Event, volume 139, pages 12697–12706. PMLR.

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/song-2026-fake-real-mitigating-out-distribution/