arxiv: 2604.22678 · v1 · submitted 2026-04-24 · 💻 cs.CL

Recognition: unknown

BERAG: Bayesian Ensemble Retrieval-Augmented Generation for Knowledge-based Visual Question Answering

Jinghong Chen , Jingbiao Mei , Guangyu Yang , Bill Byrne

Authors on Pith no claims yet

Pith reviewed 2026-05-08 11:31 UTC · model grok-4.3

classification 💻 cs.CL

keywords retrieval-augmented generationBayesian ensemblevisual question answeringdocument attributionlost-in-the-middleprobabilistic re-rankingknowledge-based VQA

0 comments

The pith

BERAG conditions language models on individual documents and updates their posterior probabilities token by token with Bayes' rule.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper tries to establish that standard retrieval-augmented generation suffers when all documents are concatenated into one context because this hides individual contributions and triggers the lost-in-the-middle effect. It proposes to instead run the model on each document separately while using Bayes' rule to update the probability that each document remains relevant after every new token. These evolving posteriors act as ensemble weights to combine the separate generations. A sympathetic reader would care because the change promises clearer attribution of answers to sources, parallel memory handling, and better scaling when retrieval lists grow long or contain visual data. The authors demonstrate the method on knowledge-based visual question answering tasks with gains over concatenation and reduced lost-in-the-middle problems.

Core claim

BERAG treats document posterior probabilities as ensemble weights and updates them token by token using Bayes' rule during generation while conditioning the language model on one document at a time rather than a combined context.

What carries the argument

Token-by-token Bayesian update of document posterior probabilities that serve as dynamic ensemble weights for generations from single-document conditioned models.

If this is right

Probabilistic re-ranking of documents happens automatically as each token is generated.
Memory usage can proceed in parallel across documents instead of building one growing sequence.
Each generated segment can be traced to the documents whose posteriors supported it.
Low-posterior documents can be pruned to speed decoding beyond standard RAG.
The final document posterior signals insufficient grounding and can trigger answer deflection.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The same separate-conditioning and Bayesian tracking could extend to text-only tasks with very large noisy corpora to control context costs.
Monitoring how posteriors shift across tokens might expose when the model begins to favor one document over another in conflicting cases.
Direct tests on purely textual retrieval lists would clarify whether the reported gains depend on the visual component of the evaluated tasks.

Load-bearing premise

The premise that conditioning on each document separately and updating posteriors via Bayes' rule will capture the necessary information without missing joint effects across documents that a concatenated context would provide.

What would settle it

Measure whether answer accuracy and source attribution remain stable when a key relevant document is moved to the middle of a long retrieval list under BERAG versus standard concatenation on the same benchmark.

Figures

Figures reproduced from arXiv: 2604.22678 by Bill Byrne, Guangyu Yang, Jingbiao Mei, Jinghong Chen.

**Figure 1.** Figure 1: VQA scores with different ground-truth docu view at source ↗

**Figure 2.** Figure 2: The inference procedure for Bayesian Ensemble Retrieval Augmented Generation (BERAG). Given view at source ↗

**Figure 3.** Figure 3: An example image panel from the MMNeedle benchmark for view at source ↗

read the original abstract

A common approach to question answering with retrieval-augmented generation (RAG) is to concatenate documents into a single context and pass it to a language model to generate an answer. While simple, this strategy can obscure the contribution of individual documents, making attribution difficult and contributing to the ``lost-in-the-middle'' effect, where relevant information in long contexts is overlooked. Concatenation also scales poorly: computational cost grows quadratically with context length, a problem that becomes especially severe when the context includes visual data, as in visual question answering. Attempts to mitigate these issues by limiting context length can further restrict performance by preventing models from benefiting from the improved recall offered by deeper retrieval. We propose Bayesian Ensemble Retrieval-Augmented Generation (BERAG), along with Bayesian Ensemble Fine-Tuning (BEFT), as a RAG framework in which language models are conditioned on individual retrieved documents rather than a single combined context. BERAG treats document posterior probabilities as ensemble weights and updates them token by token using Bayes' rule during generation. This approach enables probabilistic re-ranking, parallel memory usage, and clear attribution of document contribution, making it well-suited for large document collections. We evaluate BERAG and BEFT primarily on knowledge-based visual question answering tasks, where models must reason over long, imperfect retrieval lists. The results show substantial improvements over standard RAG, including strong gains on Document Visual Question Answering and multimodal needle-in-a-haystack benchmarks. We also demonstrate that BERAG mitigates the ``lost-in-the-middle'' effect. The document posterior can be used to detect insufficient grounding and trigger deflection, while document pruning enables faster decoding than standard RAG.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

BERAG's token-by-token Bayesian reweighting on single-document conditionings targets attribution and scaling in visual RAG, but the abstract's performance claims lack any supporting numbers or controls.

read the letter

The main new piece is conditioning the LM on individual documents rather than one concatenated context, then updating document posterior weights at each generated token via Bayes' rule. This setup is paired with BEFT for fine-tuning and is positioned for knowledge-based VQA with long, noisy retrieval lists. It directly tackles lost-in-the-middle effects, attribution opacity, and quadratic cost growth when visuals are involved, while allowing parallel memory use and pruning for speed. Those are practical pain points in multimodal RAG, and the framing around probabilistic re-ranking and deflection on weak grounding is straightforward. The paper does a decent job explaining why standard concatenation falls short on imperfect retrievals. The soft spot is the complete absence of quantitative results, baselines, ablations, or error analysis in the abstract; claims of substantial gains on DocVQA and needle-in-a-haystack tasks cannot be evaluated without the actual data. The stress-test point on LM miscalibration is relevant here—if the per-token likelihoods are unreliable, the weight updates could concentrate too fast or fluctuate for reasons unrelated to true document utility, which would weaken both the attribution guarantee and any efficiency story. Without calibration checks or controls on the update rule itself, it is hard to attribute improvements to the Bayesian ensemble rather than the parallel conditioning alone. This is for people building reliable multimodal retrieval systems who need better interpretability on long contexts. A reader focused on VQA or scalable RAG would find the framework description useful even if the experiments need more detail. It deserves a serious referee because the mechanism is coherent and the targeted problems are real, though the write-up will need the full results and ablations to hold up.

Referee Report

2 major / 1 minor

Summary. The paper proposes Bayesian Ensemble Retrieval-Augmented Generation (BERAG) and Bayesian Ensemble Fine-Tuning (BEFT) as an alternative RAG framework for knowledge-based visual question answering. Rather than concatenating retrieved documents into one context, the LM is conditioned on individual documents; document posterior probabilities are maintained as ensemble weights and updated token-by-token via Bayes' rule during generation. The approach is claimed to enable probabilistic re-ranking, parallel memory access, explicit attribution of document contributions, mitigation of the lost-in-the-middle effect, document pruning for faster decoding, and deflection on insufficient grounding. Experiments are reported to show substantial gains over standard RAG on Document VQA and multimodal needle-in-a-haystack tasks.

Significance. If the performance gains and attribution properties are robustly demonstrated, BERAG would provide a principled, scalable alternative to concatenation-based RAG that is particularly relevant for long, imperfect retrieval lists and multimodal inputs. The probabilistic weighting and pruning mechanisms address practical deployment issues around context length and interpretability.

major comments (2)

[Abstract] Abstract: The abstract asserts 'substantial improvements over standard RAG, including strong gains on Document Visual Question Answering and multimodal needle-in-a-haystack benchmarks' yet supplies no numerical results, baseline scores, statistical tests, or error analysis. Because the central claim is empirical superiority, the absence of quantitative support in the summary section prevents direct evaluation of the magnitude and reliability of the reported benefits.
[Methods] Methods (posterior update rule): The sequential update w_d ∝ w_d · p(t_k | q, d, t_<k) treats per-document LM likelihoods as reliable re-weighting factors. Decoder-only models are known to produce miscalibrated probabilities on long multimodal sequences; without calibration diagnostics, ablation of the update rule itself, or controls for probability artifacts, it is unclear whether observed gains and attribution properties arise from the Bayesian ensemble or from parallel conditioning alone.

minor comments (1)

[Abstract] The description of BEFT is limited to a single sentence; a brief comparison to standard fine-tuning objectives and any additional hyperparameters would improve clarity.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive and insightful comments. We address each major comment below and describe the revisions we will incorporate to strengthen the manuscript.

read point-by-point responses

Referee: [Abstract] Abstract: The abstract asserts 'substantial improvements over standard RAG, including strong gains on Document Visual Question Answering and multimodal needle-in-a-haystack benchmarks' yet supplies no numerical results, baseline scores, statistical tests, or error analysis. Because the central claim is empirical superiority, the absence of quantitative support in the summary section prevents direct evaluation of the magnitude and reliability of the reported benefits.

Authors: We agree that the abstract would be strengthened by including specific quantitative results. In the revised manuscript we will update the abstract to report the key performance gains (e.g., relative improvements on Document VQA and the multimodal needle-in-a-haystack tasks versus standard RAG), while remaining within abstract length constraints. revision: yes
Referee: [Methods] Methods (posterior update rule): The sequential update w_d ∝ w_d · p(t_k | q, d, t_<k) treats per-document LM likelihoods as reliable re-weighting factors. Decoder-only models are known to produce miscalibrated probabilities on long multimodal sequences; without calibration diagnostics, ablation of the update rule itself, or controls for probability artifacts, it is unclear whether observed gains and attribution properties arise from the Bayesian ensemble or from parallel conditioning alone.

Authors: This is a fair concern about probability calibration in decoder-only models. Our experiments demonstrate consistent gains from the full Bayesian ensemble over standard RAG, including better mitigation of the lost-in-the-middle effect and more accurate document attribution. To isolate the contribution of the posterior updates, we will add an ablation comparing the Bayesian weighting rule against uniform (non-updated) parallel conditioning. We will also include a short discussion of calibration considerations for long multimodal sequences. revision: yes

Circularity Check

0 steps flagged

No significant circularity in BERAG derivation chain

full rationale

The paper's central mechanism applies standard Bayes' rule to update document posterior weights token-by-token as w_d ∝ w_d · p(t_k | q, d, t_<k), using LM likelihoods conditioned on single documents. This is a direct, non-reductive use of an external probabilistic rule rather than any self-definition, fitted parameter renamed as prediction, or load-bearing self-citation. No equations equate outputs to inputs by construction, no uniqueness theorems or ansatzes are smuggled via prior work, and the framework is presented as an independent proposal evaluated against standard RAG concatenation on external benchmarks. The derivation remains self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review based solely on abstract; no explicit free parameters or invented entities are described. Relies on standard application of Bayes' rule to document posteriors in a generation setting.

axioms (1)

domain assumption Bayes' rule applies directly to updating document posterior probabilities token-by-token during autoregressive generation
This is the core mechanism stated for BERAG.

pith-pipeline@v0.9.0 · 5605 in / 1370 out tokens · 65705 ms · 2026-05-08T11:31:56.266527+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

34 extracted references · 25 canonical work pages · 2 internal anchors

[1]

online" 'onlinestring :=

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list INTEGERS output.state before.all mid.sentence after.sentence after.block STRING...
[2]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 global.max substring 't := if while FUNCTION word.in bbl.in capitalize " " * FUNCT...
[3]

Cyril Allauzen and Michael Riley. 2011. Bayesian language model interpolation for mobile speech input. In Interspeech, pages 1429--1432

2011
[4]

Davide Caffagni, Sara Sarto, Marcella Cornia, Lorenzo Baraldi, and Rita Cucchiara. 2025. Recurrence-Enhanced Vision-and-Language Transformers for Robust Multimodal Document Retrieval . In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition

2025
[5]

Guo Chen, Zhiqi Li, Shihao Wang, Jindong Jiang, Yicheng Liu, Lidong Lu, De - An Huang, Wonmin Byeon, Matthieu Le, Tuomas Rintamaki, Tyler Poon, Max Ehrlich, Tong Lu, Limin Wang, Bryan Catanzaro, Jan Kautz, Andrew Tao, Zhiding Yu, and Guilin Liu. 2025. https://doi.org/10.48550/ARXIV.2504.15271 Eagle 2.5: Boosting long-context post-training for frontier vis...

work page doi:10.48550/arxiv.2504.15271 2025
[6]

Yang Chen, Hexiang Hu, Yi Luan, Haitian Sun, Soravit Changpinyo, Alan Ritter, and Ming - Wei Chang. 2023. https://doi.org/10.18653/V1/2023.EMNLP-MAIN.925 Can pre-trained vision and language models answer visual information-seeking questions? In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, EMNLP 2023, Singapore, D...

work page doi:10.18653/v1/2023.emnlp-main.925 2023
[7]

Federico Cocchi, Nicholas Moratelli, Marcella Cornia, Lorenzo Baraldi, and Rita Cucchiara. 2025. https://doi.org/10.1109/CVPR52734.2025.00859 Augmenting multimodal llms with self-reflective tokens for knowledge-based visual question answering . In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-15, 20...

work page doi:10.1109/cvpr52734.2025.00859 2025
[8]

Lianghao Deng, Yuchong Sun, Shizhe Chen, Ning Yang, Yunfeng Wang, and Ruihua Song. 2025. https://aclanthology.org/2025.coling-main.647/ M u KA : Multimodal knowledge augmented visual information-seeking . In Proceedings of the 31st International Conference on Computational Linguistics, pages 9675--9686, Abu Dhabi, UAE. Association for Computational Linguistics

2025
[9]

Kelvin Guu, Kenton Lee, Zora Tung, Panupong Pasupat, and Mingwei Chang. 2020. https://proceedings.mlr.press/v119/guu20a.html Retrieval augmented language model pre-training . In Proceedings of the 37th International Conference on Machine Learning, volume 119 of Proceedings of Machine Learning Research, pages 3929--3938. PMLR

2020
[10]

Jongha Kim, Byungoh Ko, Jeehye Na, Jinsung Yoon, and Hyunwoo J. Kim. 2026. https://doi.org/10.48550/ARXIV.2602.06050 Relevance-aware multi-context contrastive decoding for retrieval-augmented visual question answering . CoRR, abs/2602.06050

work page doi:10.48550/arxiv.2602.06050 2026
[11]

u ttler, Mike Lewis, Wen - tau Yih, Tim Rockt \

Patrick Lewis, Ethan Perez, Aleksandra Piktus, Fabio Petroni, Vladimir Karpukhin, Naman Goyal, Heinrich K \" u ttler, Mike Lewis, Wen - tau Yih, Tim Rockt \" a schel, Sebastian Riedel, and Douwe Kiela. 2020. https://proceedings.neurips.cc/paper/2020/hash/6b493230205f780e1bc26945df7481e5-Abstract.html Retrieval-augmented generation for knowledge-intensive ...

2020
[12]

Weitao Li, Junkai Li, Weizhi Ma, and Yang Liu. 2024. https://doi.org/10.18653/v1/2024.acl-long.79 Citation-enhanced generation for LLM -based chatbots . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1451--1466, Bangkok, Thailand. Association for Computational Linguistics

work page doi:10.18653/v1/2024.acl-long.79 2024
[13]

Zongmin Li, Yachuan Li, Lei Kang, Dimosthenis Karatzas, and Wenkang Ma. 2026. https://arxiv.org/abs/2601.11976 Avir: Adaptive visual in-document retrieval for efficient multi-page document question answering . Preprint, arXiv:2601.11976

work page arXiv 2026
[14]

Weizhe Lin and Bill Byrne. 2022. https://doi.org/10.18653/v1/2022.emnlp-main.772 Retrieval augmented visual question answering with outside knowledge . In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 11238--11254, Abu Dhabi, United Arab Emirates. Association for Computational Linguistics

work page doi:10.18653/v1/2022.emnlp-main.772 2022
[15]

Weizhe Lin, Jinghong Chen, Jingbiao Mei, Alexandru Coca, and Bill Byrne. 2023. http://papers.nips.cc/paper\_files/paper/2023/hash/47393e8594c82ce8fd83adc672cf9872-Abstract-Conference.html Fine-grained late-interaction multi-modal retrieval for retrieval augmented visual question answering . In Advances in Neural Information Processing Systems 36: Annual C...

2023
[16]

Weizhe Lin, Jingbiao Mei, Jinghong Chen, and Bill Byrne. 2024. https://doi.org/10.18653/V1/2024.ACL-LONG.289 Preflmr: Scaling up fine-grained late-interaction multi-modal retrievers . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2024, Bangkok, Thailand, August 11-16, 2024 , pages 5...

work page doi:10.18653/v1/2024.acl-long.289 2024
[17]

Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang

Nelson F. Liu, Kevin Lin, John Hewitt, Ashwin Paranjape, Michele Bevilacqua, Fabio Petroni, and Percy Liang. 2024. https://doi.org/10.1162/tacl_a_00638 Lost in the middle: How language models use long contexts . Transactions of the Association for Computational Linguistics, 12:157--173

work page doi:10.1162/tacl_a_00638 2024
[18]

Xinwei Long, Zhiyuan Ma, Ermo Hua, Kaiyan Zhang, Biqing Qi, and Bowen Zhou. 2025. https://doi.org/10.1609/AAAI.V39I23.34653 Retrieval-augmented visual question answering via built-in autoregressive search engines . In Thirty-Ninth AAAI Conference on Artificial Intelligence, Thirty-Seventh Conference on Innovative Applications of Artificial Intelligence, F...

work page doi:10.1609/aaai.v39i23.34653 2025
[19]

Xueguang Ma, Sheng - Chieh Lin, Minghan Li, Wenhu Chen, and Jimmy Lin. 2024. https://doi.org/10.18653/V1/2024.EMNLP-MAIN.373 Unifying multimodal retrieval via document screenshot embedding . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, EMNLP 2024, Miami, FL, USA, November 12-16, 2024 , pages 6492--6505. Associ...

work page doi:10.18653/v1/2024.emnlp-main.373 2024
[20]

Thomas Mensink, Jasper R. R. Uijlings, Llu \' s Castrej \' o n, Arushi Goel, Felipe Cadar, Howard Zhou, Fei Sha, Andr \' e Ara \' u jo, and Vittorio Ferrari. 2023. https://doi.org/10.1109/ICCV51070.2023.00289 Encyclopedic VQA: visual questions about detailed properties of fine-grained categories . In IEEE/CVF International Conference on Computer Vision, I...

work page doi:10.1109/iccv51070.2023.00289 2023
[21]

Jirui Qi, Gabriele Sarti, Raquel Fern \'a ndez, and Arianna Bisazza. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.347 Model internals-based answer attribution for trustworthy retrieval-augmented generation . In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 6037--6053, Miami, Florida, USA. Association fo...

work page doi:10.18653/v1/2024.emnlp-main.347 2024
[22]

Zexuan Qiu, Zijing Ou, Bin Wu, Jingjing Li, Aiwei Liu, and Irwin King. 2025. https://doi.org/10.18653/v1/2025.naacl-long.236 Entropy-based decoding for retrieval-augmented large language models . In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volum...

work page doi:10.18653/v1/2025.naacl-long.236 2025
[23]

Danielle Saunders, Felix Stahlberg, Adri \`a de Gispert, and Bill Byrne. 2019. https://doi.org/10.18653/v1/P19-1022 Domain adaptive inference for neural machine translation . In Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 222--228, Florence, Italy. Association for Computational Linguistics

work page doi:10.18653/v1/p19-1022 2019
[24]

Weijia Shi, Xiaochuang Han, Mike Lewis, Yulia Tsvetkov, Luke Zettlemoyer, and Wen - tau Yih. 2024. https://doi.org/10.18653/V1/2024.NAACL-SHORT.69 Trusting your evidence: Hallucinate less with context-aware decoding . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Techno...

work page doi:10.18653/v1/2024.naacl-short.69 2024
[25]

Quan Sun, Yuxin Fang, Ledell Wu, Xinlong Wang, and Yue Cao. 2023. https://doi.org/10.48550/ARXIV.2303.15389 EVA-CLIP: improved training techniques for CLIP at scale . CoRR, abs/2303.15389

work page internal anchor Pith review doi:10.48550/arxiv.2303.15389 2023
[26]

Ryota Tanaka, Taichi Iki, Taku Hasegawa, Kyosuke Nishida, Kuniko Saito, and Jun Suzuki. 2025. https://doi.org/10.1109/CVPR52734.2025.02312 Vdocrag: Retrieval-augmented generation over visually-rich documents . In IEEE/CVF Conference on Computer Vision and Pattern Recognition, CVPR 2025, Nashville, TN, USA, June 11-15, 2025 , pages 24827--24837. Computer V...

work page doi:10.1109/cvpr52734.2025.02312 2025
[27]

Ryota Tanaka, Kyosuke Nishida, Kosuke Nishida, Taku Hasegawa, Itsumi Saito, and Kuniko Saito. 2023. https://doi.org/10.1609/AAAI.V37I11.26598 Slidevqa: A dataset for document visual question answering on multiple images . In Thirty-Seventh AAAI Conference on Artificial Intelligence, AAAI 2023, Thirty-Fifth Conference on Innovative Applications of Artifici...

work page doi:10.1609/aaai.v37i11.26598 2023
[28]

Hengyi Wang, Haizhou Shi, Shiwei Tan, Weiyi Qin, Wenyuan Wang, Tunyu Zhang, Akshay Nambi, Tanuja Ganu, and Hao Wang. 2025. https://doi.org/10.18653/V1/2025.NAACL-LONG.166 Multimodal needle in a haystack: Benchmarking long-context capability of multimodal large language models . In Proceedings of the 2025 Conference of the Nations of the Americas Chapter o...

work page doi:10.18653/v1/2025.naacl-long.166 2025
[29]

Cong Wei, Yang Chen, Haonan Chen, Hexiang Hu, Ge Zhang, Jie Fu, Alan Ritter, and Wenhu Chen. 2024. https://doi.org/10.1007/978-3-031-73021-4\_23 Uniir: Training and benchmarking universal multimodal information retrievers . In Computer Vision - ECCV 2024 - 18th European Conference, Milan, Italy, September 29-October 4, 2024, Proceedings, Part LXXXVII , Le...

work page doi:10.1007/978-3-031-73021-4 2024
[30]

Thomas Wolf, Lysandre Debut, Victor Sanh, Julien Chaumond, Clement Delangue, Anthony Moi, Perric Cistac, Clara Ma, Yacine Jernite, Julien Plu, Canwen Xu, Teven Le Scao, Sylvain Gugger, Mariama Drame, Quentin Lhoest, and Alexander M. Rush. 2020. https://www.aclweb.org/anthology/2020.emnlp-demos.6 Transformers: State-of-the-art natural language processing ....

2020
[31]

Yibin Yan and Weidi Xie. 2024. https://doi.org/10.18653/v1/2024.findings-emnlp.83 E cho S ight: Advancing visual-language models with W iki knowledge . In Findings of the Association for Computational Linguistics: EMNLP 2024, pages 1538--1551, Miami, Florida, USA. Association for Computational Linguistics

work page doi:10.18653/v1/2024.findings-emnlp.83 2024
[32]

Guangyu Yang, Jinghong Chen, Weizhe Lin, and Bill Byrne. 2024. https://doi.org/10.18653/v1/2024.naacl-short.34 Direct preference optimization for neural machine translation with minimum B ayes risk decoding . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (V...

work page doi:10.18653/v1/2024.naacl-short.34 2024
[33]

Yizhe Zhang, Siqi Sun, Xiang Gao, Yuwei Fang, Chris Brockett, Michel Galley, Jianfeng Gao, and Bill Dolan. 2022. https://doi.org/10.1609/AAAI.V36I10.21429 Retgen: A joint framework for retrieval and grounded text generation modeling . In Thirty-Sixth AAAI Conference on Artificial Intelligence, AAAI 2022, Thirty-Fourth Conference on Innovative Applications...

work page doi:10.1609/aaai.v36i10.21429 2022
[34]

Yaowei Zheng, Richong Zhang, Junhao Zhang, Yanhan Ye, Zheyan Luo, Zhangchi Feng, and Yongqiang Ma. 2024. http://arxiv.org/abs/2403.13372 Llamafactory: Unified efficient fine-tuning of 100+ language models . In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 3: System Demonstrations), Bangkok, Thailand. Assoc...

work page internal anchor Pith review arXiv 2024