Elamur Research⌕
☰
home / papers / 2026 / Test-Time Training for Zero-Resource Dense Retrieval Reranking

Test-Time Training for Zero-Resource Dense Retrieval Reranking

Paper · 2026 · Shiyan Liu and Yichen Li

Summary

The paper addresses reranking for dense retrieval when the target domain has no relevance labels. Cross-encoders need supervised training and add hundreds of milliseconds per query. BM25 is often unavailable in vector-database-only deployments. The authors propose DART (Dense Adaptive Reranking at Test-time), a test-time training method that tunes the scoring function itself and leaves query and document embeddings unchanged.

For each query, cosine scoring is generalized to a bilinear form whose matrix W starts at the identity. The dense retriever's own ranked list supplies noisy supervision: the highest-ranked candidates act as pseudo-positives and the lowest-ranked as pseudo-negatives. A few gradient steps then minimize a margin loss with three refinements:

  • confidence weights on the pseudo-labels,
  • a margin that grows with query difficulty,
  • regularization that keeps W near the identity.

A cross-query momentum mechanism (MetaInit and EMA) carries adaptation from one query to the next. The optimizer (SGD or Lion) is picked per dataset.

Across six BEIR datasets, DART raises NDCG@10 by about 2.1% relative to the dense baseline, adding under 10ms per query. This makes it a cheap, label-free reranking option.

Key points

  • Evaluation covers six BEIR benchmarks (NFCorpus, SCIDOCS, FiQA, ArguAna, TREC-COVID, SciFact). DART beats the dense baseline on five of them, with a mean relative NDCG@10 gain of +2.1%.
  • In the ablation, EMA is the only component that helps on all four tested datasets, lifting FiQA from −5.7% to −1.8%. Adding Lion last gives +4.1% on SCIDOCS.
  • The updates to W are tiny and low-rank:
    • With 384-dimensional embeddings, the median ‖ΔW‖_F is 0.095, roughly 0.5% of ‖I‖_F.
    • The leading singular value explains 19.4% of the variation, and the top three explain 28.4%.
  • Harder queries (lower top-1 similarity) receive larger updates, which supports the difficulty-adaptive margin.
  • The authors note two limitations:
    • Choosing the optimizer requires a warm-up of 50–100 queries; SGD is the safer default.
    • The full d×d matrix scales quadratically with embedding size, so a low-rank form of W is proposed for encoders with d ≥ 768.

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

🏢ByteDanceBeijing · China41 works · 4,180 citations🎓Huazhong University of Science and TechnologyWuhan · China45 works · 6,345 citations

 Fields, topics and concepts

📚 Shelves 6The 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 worksComputer Vision and Pattern Recognition · 2,869 worksInformation Systems · 1,437 works

🧭 Topics 1Research subjects and languages the work is about.Information retrieval & search · 473 works

Cite

Cite as Shiyan Liu and Yichen Li. 2026. Test-Time Training for Zero-Resource Dense Retrieval Reranking. Proceedings of the 4th Workshop on Towards Knowledgeable Foundation Models (KnowFM 2026).

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2026.knowfm-1.8

@inproceedings{liu2026test,
  title = {{Test-Time Training for Zero-Resource Dense Retrieval Reranking}},
  author = {Shiyan Liu and Yichen Li},
  booktitle = {Proceedings of the 4th Workshop on Towards Knowledgeable Foundation Models (KnowFM 2026)},
  year = {2026},
  doi = {10.18653/v1/2026.knowfm-1.8},
  url = {https://doi.org/10.18653/v1/2026.knowfm-1.8}
}

References

39 references, 17 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abdelrahman Abdallah, Jamshid Mozafari, Bhawna Piryani, and Adam Jatowt. 2025. Asrank: Zero-shot re-ranking with answer scent for document retrieval. In Findings of the Association for Computational Linguistics: NAACL 2025, pages 2950–2970. find
  2. Ekin Akyürek, Mehul Damani, Adam Zweiger, Linlu Qiu, Han Guo, Jyothish Pari, Yoon Kim, and Jacob Andreas. 2024. The surprising effectiveness of test-time training for few-shot learning. arXiv preprint arXiv:2411.07279. find
  3. Shijie Chen, Bernal Jiménez Gutiérrez, and Yu Su. 2024. Attention in large language models yields efficient zero-shot re-rankers. arXiv preprint arXiv:2410.02642. — [arXiv:2410.02642](https://arxiv.org/abs/2410.02642) find
  4. Xiangning Chen, Chen Liang, Da Huang, Esteban Real, Kaiyuan Wang, Hieu Pham, Xuanyi Dong, Thang Luong, Cho-Jui Hsieh, Yifeng Lu, and 1 others. 2023. Symbolic discovery of optimization algorithms. Advances in neural information processing systems, 36:49205–49233. find
  5. DeepSeek-AI, Daya Guo, Dejian Yang, Haowei Zhang, Jun-Mei Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiaoling Bi, Xiaokang Zhang, Xingkai Yu, Yu Wu, Z. F. Wu, Zhibin Gou, Zhihong Shao, Zhuoshu Li, Ziyi Gao, and 179 others. 2025. Deepseek-r1 incentivizes reasoning in llms throug… find
  6. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies,…
  7. Jiafeng Guo, Yixing Fan, Liang Pang, Liu Yang, Qingyao Ai, Hamed Zamani, Chen Wu, W Bruce Croft, and Xueqi Cheng. 2020. A deep look into neural ranking models for information retrieval. Information Processing & Management, 57(6):102067. find
  8. Sebastian Hofstätter, Sheng-Chieh Lin, Jheng-Hong Yang, Jimmy Lin, and Allan Hanbury. 2021. Efficiently teaching an effective dense retriever with balanced topic aware sampling. In Proceedings of the 44th International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR…
  9. Chao-Wei Huang and Yun-Nung Chen. 2024. Instupr: Instruction-based unsupervised passage reranking with large language models. arXiv preprint arXiv:2403.16435. — [arXiv:2403.16435](https://arxiv.org/abs/2403.16435) find
  10. Gautier Izacard, Mathilde Caron, Lucas Hosseini, Sebastian Riedel, Piotr Bojanowski, Armand Joulin, and Edouard Grave. 2021. Unsupervised dense information retrieval with contrastive learning.
  11. Trans. Mach. Learn. Res., 2022. find
  12. Jeff Johnson, Matthijs Douze, and Hervé Jégou. 2019. Billion-scale similarity search with gpus. IEEE transactions on big data, 7(3):535–547.
  13. Vladimir Karpukhin, Barlas Oguz, Sewon Min, Patrick Lewis, Ledell Wu, Sergey Edunov, Danqi Chen, and Wen-tau Yih. 2020. Dense passage retrieval for open-domain question answering. In Proceedings of the 2020 conference on empirical methods in natural language processing (EMNLP), pages 6769–6781.
  14. Omar Khattab and Matei Zaharia. 2020. Colbert: Efficient and effective passage search via contextualized late interaction over bert. In Proceedings of the 43rd International ACM SIGIR conference on research and development in Information Retrieval, pages 39– 48.
  15. Victor Lavrenko and W Bruce Croft. 2017. Relevance-based language models. In ACM SIGIR Forum, volume 51, pages 260–267. ACM New York, NY, USA. find
  16. Hang Li, Ahmed Mourad, Shengyao Zhuang, Bevan Koopman, and Guido Zuccon. 2023. Pseudo relevance feedback with deep language models and dense retrievers: Successes and pitfalls. ACM Transactions on Information Systems, 41(3):1–40. find
  17. Hang Li, Xiao Wang, Bevan Koopman, and Guido Zuccon. 2025. Pseudo relevance feedback is enough to close the gap between small and large dense retrieval models. arXiv preprint arXiv:2503.14887. find
  18. Jimmy Lin, Rodrigo Nogueira, and Andrew Yates. 2022. Pretrained transformers for text ranking: Bert and beyond. Springer Nature. find
  19. Yuejiang Liu, Parth Kothari, Bastien Van Delft, Baptiste Bellot-Gurlet, Taylor Mordan, and Alexandre Alahi. 2021. Ttt++: When does self-supervised test-time training fail or thrive? Advances in Neural Information Processing Systems, 34:21808–21820. find
  20. Rui Meng, Ye Liu, Semih Yavuz, Divyansh Agarwal, Lifu Tu, Ning Yu, Jianguo Zhang, Meghana Bhat, and Yingbo Zhou. 2022. Unsupervised dense retrieval deserves better positive pairs: Scalable augmentation with query extraction and generation. arXiv preprint arXiv:2212.08841. — [arXiv:2212.08841](https://arxiv.org/abs/2212.08841) find
  21. Rodrigo Nogueira and Kyunghyun Cho. 2019. Passage re-ranking with bert. arXiv preprint arXiv:1901.04085.
  22. Rodrigo Nogueira, Zhiying Jiang, Ronak Pradeep, and Jimmy Lin. 2020. Document ranking with a pretrained sequence-to-sequence model. In Findings of the association for computational linguistics: EMNLP 2020, pages 708–718.
  23. Liang Wang, Nan Yang, Xiaolong Huang, Binxing Jiao, Linjun Yang, Daxin Jiang, Rangan Majumder, and Furu Wei. 2022b. Text embeddings by weakly-supervised contrastive pre-training. arXiv preprint arXiv:2212.03533.
  24. Ronak Pradeep, Sahel Sharifymoghaddam, and Jimmy Lin. 2023. Rankvicuna: Zero-shot listwise document reranking with open-source large language models. arXiv preprint arXiv:2309.15088. find
  25. Xiao Wang, Craig Macdonald, Nicola Tonellotto, and Iadh Ounis. 2023. Colbert-prf: Semantic pseudo-relevance feedback for dense passage and document retrieval. ACM Transactions on the Web, 17(1):1–39. find
  26. Nils Reimers and Iryna Gurevych. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. In Proceedings of the 2019 conference on empirical methods in natural language processing and the 9th international joint conference on natural language processing (EMNLP-IJCNLP), pages 3982–3992.
  27. Orion Weller, Kathryn Ricci, Eugene Yang, Andrew Yates, Dawn Lawrie, and Benjamin Van Durme. 2025. Rank1: Test-time compute for reranking in information retrieval. arXiv preprint arXiv:2502.18418. — [arXiv:2502.18418](https://arxiv.org/abs/2502.18418) find
  28. Stephen Robertson and Hugo Zaragoza. 2009. The probabilistic relevance framework: BM25 and beyond, volume 4. Now Publishers Inc. find
  29. Jon Saad-Falcon, Omar Khattab, Keshav Santhanam, Radu Florian, Martin Franz, Salim Roukos, Avirup Sil, Md Sultan, and Christopher Potts. 2023. Udapdr: unsupervised domain adaptation via llm prompting and distillation of rerankers. In Proceedings of the 2023 conference on empirical methods in natural… find
  30. Shitao Xiao, Zheng Liu, Peitian Zhang, Niklas Muennighoff, Defu Lian, and Jian-Yun Nie. 2024. C-pack: Packed resources for general chinese embeddings. In Proceedings of the 47th international ACM SIGIR conference on research and development in information retrieval, pages 641–649.
  31. Ji Xin, Chenyan Xiong, Ashwin Srinivasan, Ankita Sharma, Damien Jose, and Paul Bennett. 2022. Zero-shot dense retrieval with momentum adversarial domain invariant representations. In Findings of the Association for Computational Linguistics: ACL 2022, pages 4008–4020. find
  32. Keshav Santhanam, Omar Khattab, Jon Saad-Falcon, Christopher Potts, and Matei Zaharia. 2022. Colbertv2: Effective and efficient retrieval via lightweight late interaction. In Proceedings of the 2022 Conference of the North American Chapter of the Association for Computational Linguistics: Human Lang…
  33. Weiwei Sun, Lingyong Yan, Xinyu Ma, Shuaiqiang Wang, Pengjie Ren, Zhumin Chen, Dawei Yin, and Zhaochun Ren. 2023. Is chatgpt good at search? investigating large language models as re-ranking agents. In Proceedings of the 2023 conference on empirical methods in natural language processing, pages 1491…
  34. Yu Sun, Xiaolong Wang, Zhuang Liu, John Miller, Alexei Efros, and Moritz Hardt. 2020. Test-time training with self-supervision for generalization under distribution shifts. In International conference on machine learning, pages 9229–9248. PMLR. find
  35. Lee Xiong, Chenyan Xiong, Ye Li, Kwok-Fung Tang, Jialin Liu, Paul Bennett, Junaid Ahmed, and Arnold Overwijk. 2020. Approximate nearest neighbor negative contrastive learning for dense text retrieval. arXiv preprint arXiv:2007.00808.
  36. Longhui Yu, Weisen Jiang, Han Shi, Jincheng Yu, Zhengying Liu, Yu Zhang, James T Kwok, Zhenguo Li, Adrian Weller, and Weiyang Liu. 2023. Metamath: Bootstrap your own mathematical questions for large language models. arXiv preprint arXiv:2309.12284.
  37. Shengyao Zhuang, Linjun Shou, and Guido Zuccon. 2023. Augmenting passage representations with query generation for enhanced cross-lingual dense retrieval. In Proceedings of the 46th International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 1827–1832. find
  38. Nandan Thakur, Nils Reimers, Andreas Rücklé, Abhishek Srivastava, and Iryna Gurevych. 2021. Beir: A heterogenous benchmark for zero-shot evaluation of information retrieval models. arXiv preprint arXiv:2104.08663.
  39. Kexin Wang, Nandan Thakur, Nils Reimers, and Iryna Gurevych. 2022a. Gpl: Generative pseudo labeling for unsupervised domain adaptation of dense retrieval. In Proceedings of the 2022 conference of the North American chapter of the association for computational linguistics: human language technologies…

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/liu-2026-test-time-training-zero-resource/