Pith. sign in

REVIEW 6 major objections 6 minor 38 references

Enhancing Audiovisual Speech Recognition through Bifocal Preference Optimization

T0 review · 6 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read Preference optimization beats 131k-hour audiovisual training

desk verdict New preference-optimization recipe for AV-ASR with real but modest gains; the SOTA headline is mostly the OWSM backbone, not the method. read the letter →

arxiv 2412.19005 v1 pith:37YYULFF submitted 2024-12-26 eess.AS cs.AI

Yihan Wu , Yichen Lu , Yifan Peng , Xihua Wang , Ruihua Song , Shinji Watanabe This is my paper · ORCID
classification eess.AScs.AI
keywords audiovisualspeechrecognitionpreferenceoptimizationdirectmultimodalASRhardnegativeconstructionreal-worldvideoworderrorratevisualfeatures
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper argues that audiovisual speech recognition (AV-ASR) in unconstrained real-world videos can be improved by framing the task as a preference optimization problem rather than as a purely supervised fine-tuning problem. The authors construct preference pairs that simulate the errors an AV-ASR system actually makes — masked or noisy audio, flipped video frames, homophone substitutions, and spontaneous-speech rewrites — and then train the model to prefer the correct transcript over each corrupted variant. On the model side, they propose BPO-AVASR, a 'bifocal' DPO-style objective that combines an input-side preference loss (rewarding correct transcripts when the input is clean rather than corrupted) with an output-side preference loss (rewarding correct transcripts over error-rewritten ones). The authors claim that this approach reaches state-of-the-art word error rates on How2, VisSpeech, and Ego4D, outperforming prior AV-ASR models even when those models were trained on hundreds of times more audiovisual data.

What carries the argument

The central object is the bifocal preference dataset and its associated loss. Input-side preference pairs $(t_w | a_w, v_w) \succ (t_w | a_l, v_w)$ and $(t_w | a_w, v_w) \succ (t_w | a_w, v_l)$ hold the transcript fixed and degrade the audio or video input; output-side pairs $(t_w | a_w, v_w) \succ (t_l | a_w, v_w)$ hold the input fixed and degrade the transcript through homophone, spontaneous-speech, or vision-omission rewrites. The training objective is $L_{BPO} = L_{input} + L_{output}$, where each term is a logistic contrast of the policy's log-probability ratio for chosen versus rejected pairs, anchored against the reference model $\pi_{\text{ref}}$. The construction of hard negatives — choosing input corruptions that genuinely remove information (flipped vision rather than random cropping) and LLM-generated transcript rewrites rather than simple dictionary swaps — is what makes the preference signal informative.

What would settle it

Run BPO-AVASR on a test set where horizontally flipping the frame removes no information relevant to the transcript (e.g., symmetric scenes). If WER still improves relative to the SFT baseline, the input-side loss is acting as a generic regularizer, not as a preference over visual informativeness.

Watch

Extended reading notes

Core claim

The central claim is that BPO-AVASR, trained with the proposed bifocal preference optimization, achieves superior performance compared to previous state-of-the-art models across three datasets. Specifically, BPO-AVASR small and medium reach 9.3% and 9.2% WER on How2, 15.6% and 14.5% on VisSpeech, and 59.2% and 56.5% on Ego4D, while the prior AVFormer baseline obtains 13.6%, 16.4%, and 64.6% respectively — and AVFormer used 131,000 hours of audiovisual fine-tuning data, whereas BPO-AVASR uses only 300 hours. The authors attribute the improvement to the bifocal preference construction, which teaches the model to avoid the specific error patterns that dominate unconstrained video speech: homophone confusions, spontaneous-speech disfluencies, and reliance on missing or misleading visual cues.

Load-bearing premise

The input-side preference loss assumes that the DPO derivation remains valid when chosen and rejected pairs share the same transcript but have different inputs; in the standard DPO objective the partition function cancels only when the input is fixed, so this loss may not be a true preference objective.

Editorial extensions

If this is right

  • AV-ASR no longer needs hundred-thousand-hour audiovisual corpora: preference optimization on 300 hours of How2 matches or beats models trained on 131k hours.
  • The same preference-construction recipe transfers to other domains: building preference pairs from Ego4D's training set improves Ego4D WER from 52.3% to 50.0% without changing the method.
  • The method suggests that ASR can be treated as a preference-alignment problem, opening the door to using human or model feedback beyond correct/wrong transcript labels.
  • Because the construction targets homophone and spontaneous-speech errors, the gains are likely to be largest in noisy, spontaneous, egocentric video — precisely the regime where conventional AV-ASR fails.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the input-side loss is not a valid DPO objective because the partition function does not cancel, the empirical gains might be replicable with a simpler contrastive regularizer; a head-to-head test with such a regularizer would clarify whether the preference framing is essential.
  • The reliance on LLM-based transcript rewrites for preference construction makes the method's performance partly a function of those generators' quality; automated, self-supervised preference generation would make it more portable to low-resource languages.
  • The reported relative reductions are computed against AVFormer, which was trained on different data; a controlled comparison with the same backbone and data would isolate the effect of the preference objective.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

6 major / 6 minor

Summary. The paper proposes BPO-AVASR, an audiovisual ASR method that starts from the audio-only OWSM v3.1 model, adds a CLIP-based visual encoder, performs supervised fine-tuning (SFT) on How2, and then applies a 'bifocal preference optimization' stage. Preference pairs are constructed by masking audio frames, flipping video frames, and using ChatGPT to rewrite transcripts (homophone, spontaneous-speech, and vision-related errors). Experiments on How2, VisSpeech, and Ego4D report WER reductions relative to an SFT-only baseline, and the authors claim state-of-the-art results relative to AVFormer and other prior systems. Ablations compare eight preference construction strategies.

Significance. The paper's main positive contribution is a practical recipe for adapting a strong audio-only ASR model to audiovisual input with modest additional data: the preference data construction pipeline is concrete, and the same-backbone control (OWSM-visual small) in Table 2 is a commendable experimental design. If the small WER gains are reproducible across seeds, the method could be useful for AV-ASR practitioners. However, the headline claim of outperforming previous SOTA is weakened by two facts: (i) the SFT baseline already surpasses AVFormer on all three datasets, so the gains attributed to preference optimization are marginal (1.3%-11.4% relative), and (ii) on How2, AVATAR† reports a lower WER (9.1) than either BPO model. The theoretical derivation of the input-side DPO loss is also questionable. The pledge to release code and data is positive for reproducibility.

major comments (6)
  1. [Section 1 and Section 5.3 (Table 2)] The central claim that BPO-AVASR 'outperforms previous state-of-the-art models across three datasets' is not supported by the paper's own control. Table 2 shows that OWSM-visual small, i.e., the same OWSM v3.1 backbone and CLIP encoder trained with SFT on How2 and no preference optimization, already achieves 10.5/15.8/59.9 WER on How2/VisSpeech/Ego4D, beating AVFormer (13.6/16.4/64.6) on every dataset. BPO-AVASR small improves these to 9.3/15.6/59.2, i.e., absolute gains of 1.2/0.2/0.7 WER. The abstract and Section 1 should attribute the SOTA-level performance primarily to the OWSM backbone and the SFT stage, and state that preference optimization provides a further smaller improvement.
  2. [Section 5.1 (Table 1)] The abstract's claim of superior performance over previous SOTA is also factually contradicted on How2: AVATAR† achieves 9.1 WER, while BPO-AVASR small and medium achieve 9.3 and 9.2, respectively. The paper acknowledges this in Section 5.1 ('AVATAR† achieves the best performance on How2'), so the blanket statement 'outperforming previous state-of-the-art models' should be qualified to refer to AVFormer specifically, or the authors should argue why AVATAR† is not the relevant SOTA baseline.
  3. [Section 3.3 (Eq. 4)] The input-side preference loss L_input is not a valid DPO objective. In the standard DPO derivation (Eqs. 1-2), the input x is fixed across the chosen and rejected responses, so the partition function Z(x) cancels in the Bradley-Terry ratio. In L_input, the chosen pair (tw|cw) and the rejected pair (tw|cl) have different inputs (cw vs cl), so the reward model r(x,y) has different partition functions Z(cw) and Z(cl), and the logistic loss does not follow from the DPO derivation. The paper should either derive a principled preference objective for the input-side case or explicitly present L_input as a contrastive/regularization term distinct from DPO. This is load-bearing because the paper's novelty rests on 'bifocal preference optimization' as preference optimization.
  4. [Sections 5.1-5.3] No error bars, confidence intervals, or multi-seed results are reported. Several conclusions rely on tiny absolute WER differences; for example, in Table 2 the differences between Flipped vision (9.7/15.5/59.3), Random cropped vision (12.1/15.7/61.5), and Masked audio (9.4/15.7/59.2) are small on VisSpeech/Ego4D, and the statement in Section 5.3 that 'the Ego4D dataset benefits significantly' from spontaneous-based generation (59.9 to 59.0) is unsupported without significance testing.
  5. [Section 4.2] The How2 data split is not defined. The text states 'Following AVFormer, we use the 300-hour version of How2' but does not specify the train/development/test partition used for the SFT stage and for the WER evaluation. Since prior baselines in Table 1 may use different splits, the authors must report the exact split (e.g., the standard How2 dev5 partition) and the number of clips/utterances used.
  6. [Section 5.3 (Table 3)] The protocol for Table 3 is ambiguous. The caption says 'All models are fine-tuned on the Ego4D training set,' but the main evaluation of Ego4D in Table 1 is zero-shot. It is unclear whether OWSM-visual small is SFT on Ego4D for the same number of epochs as the How2 SFT, whether the preference data for this experiment are constructed from the Ego4D training set as described in the text, and what hyperparameters are used. The authors should specify the full training setup for Table 3.
minor comments (6)
  1. [Section 4.1] The phrase '1 V100 GPU' should be rewritten as 'one V100 GPU' or 'a single V100 GPU.'
  2. [Section 3.1] The notation τ i is used for an element of D, but the preceding definition lists triples (ai, vi, ti); consider writing the tuple explicitly to avoid ambiguity.
  3. [Figure 1] The labels 'Focal I' and 'Focal II' are confusing; consider using 'Input-side preference' and 'Output-side preference' directly in the figure.
  4. [Section 5.3] The word 'significantly' is used without statistical support (e.g., 'the Ego4D dataset benefits significantly'); replace it with 'numerically' or add significance tests.
  5. [Table 2] The mixture row labels the model as 'BPO-AVASR small,' but the reader must cross-reference Table 1 for the model definition; consider adding a footnote clarifying that this is the full method.
  6. [Table 1] The baseline 'OWSM-ft small' is cited as Peng et al. 2023b, but the text elsewhere refers to OWSM v3.1 (Peng et al. 2024); ensure the baseline reference matches the exact model version used.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the empirical claims are evaluated on held-out or zero-shot test sets, and the only overlapping-author citation (OWSM backbone) is not load-bearing because the paper provides a same-backbone no-BPO control.

full rationale

The paper's claimed derivation chain is empirical and self-contained. Preference pairs are constructed by manipulating How2 training inputs and ground-truth transcripts (Section 3.1), and all reported results (Tables 1-3) are evaluated on held-out How2 test data or on zero-shot VisSpeech/Ego4D, so no test WER is used to fit a parameter. The central SOTA comparison is a direct table result, and Table 2 includes the same-backbone SFT control (OWSM-visual small), which isolates the BPO gain as a small delta rather than defining it. The only overlapping-author citation with any load is OWSM v3.1 (Peng et al. 2024, same group), used as a pretrained backbone; because the paper provides a no-BPO control on that exact backbone and because OWSM is an independently trained external model, this is a model choice, not a circular justification. The input-side preference loss in Eq. 4 is not the standard DPO objective since the partition function Z(x) does not cancel across different input contexts cw and cl; this is a validity/attribution concern about whether the gain comes from preference alignment or a contrastive regularizer, but it is not circularity: the reported predictions are not equal to the training inputs by construction, and no benchmark result is assumed in the objective. No uniqueness theorem, ansatz-smuggling citation, or renamed-known-result pattern appears in the paper.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central claim rests on one ad hoc assumption (DPO validity with differing inputs), two domain assumptions about simulated errors matching real errors, and standard background results from DPO. No test-set parameters are fitted, so circularity burden is low.

free parameters (3)
  • alpha (CTC loss weight) = 0.3
    Weight for the CTC loss in the SFT objective (Eq. 3); chosen by hand, not swept.
  • beta (DPO temperature) = 0.1
    Temperature controlling deviation from the reference model in Eq. 4; chosen by hand, affects preference signal strength.
  • M (number of video frames) = 4
    Number of uniformly sampled frames per video for CLIP; chosen by hand and not ablated.
assumptions (4)
  • standard math Bradley-Terry model and DPO objective (Eq. 2) are valid for preference optimization.
    Background result cited from Rafailov et al. 2023; the paper builds on it for both preference losses.
  • ad hoc to paper The DPO objective remains valid when chosen and rejected pairs have different input contexts (Eq. 4, L_input).
    The standard DPO derivation assumes a fixed input x for both responses; Eq. 4 applies DPO to (tw|cw) vs (tw|cl) with cw != cl, and no alternative derivation is provided.
  • domain assumption ChatGPT-generated homophone, spontaneous, and vision-based rewrites approximate the real AV-ASR error distribution.
    Section 3.1 builds the output-side preference dataset with ChatGPT prompts; the transfer of improvements to VisSpeech and Ego4D is taken as evidence that the simulated errors match real errors.
  • domain assumption Masking content-word audio frames and flipping video frames create hard negatives that improve AV-ASR.
    Section 3.1, Type I and II; the paper does not show that flipping the entire frame is semantically a hard negative, only that it makes OCR harder.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Enhancing Audiovisual Speech Recognition through Bifocal Preference Optimization." pith.science (2026). https://pith.science/paper/37YYULFF

@misc{pith2026241219005,
  author       = {Pith},
  title        = {Pith review of: Enhancing Audiovisual Speech Recognition through Bifocal Preference Optimization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/37YYULFF}},
  note         = {Machine review of arXiv:2412.19005}
}
read the original abstract

Audiovisual Automatic Speech Recognition (AV-ASR) aims to improve speech recognition accuracy by leveraging visual signals. It is particularly challenging in unconstrained real-world scenarios across various domains due to noisy acoustic environments, spontaneous speech, and the uncertain use of visual information. Most previous works fine-tune audio-only ASR models on audiovisual datasets, optimizing them for conventional ASR objectives. However, they often neglect visual features and common errors in unconstrained video scenarios. In this paper, we propose using a preference optimization strategy to improve speech recognition accuracy for real-world videos. First, we create preference data via simulating common errors that occurred in AV-ASR from two focals: manipulating the audio or vision input and rewriting the output transcript. Second, we propose BPO-AVASR, a Bifocal Preference Optimization method to improve AV-ASR models by leveraging both input-side and output-side preference. Extensive experiments demonstrate that our approach significantly improves speech recognition accuracy across various domains, outperforming previous state-of-the-art models on real-world video speech recognition.

Figures

Figures reproduced from arXiv: 2412.19005 by the authors.

Figure 1
Figure 1. Overview of our proposed framework. We con￾struct bifocal preferences by augmenting transcript (T), au￾dio (A) or video (V), resulting in two types of preference pairs: input-side (A and V) and output-side (T). Through bi￾focal preference optimization, BPO-AVASR learns to gener￾ate transcripts that are better aligned with these preferences. Compared to standard ASR, performing AV-ASR on real￾world videos presents th… view at source ↗
Figure 2
Figure 2. The illustration of preference dataset constructing strategies: (a) Input-side preference construction: simulates errors by manipulating input audios and videos; (b) Output-side preference construction: simulates common errors by manipu￾lating ground truth transcripts. over SFT-based AV-ASR models. Also, we leverage bifo￾cal preference optimization, specifically designed for uncon￾strained AV-ASR tasks. Unlike DPO, … view at source ↗
Figure 3
Figure 3. Qualitative Results. We show the ground truth text (GT), and predictions from the OWSM-visual small(w/o DPO) and BPO-AVASR small. We show the enhanced performance in three scenarios: when vision provides content cues (top), when vision offers context clues (middle), and when speech is under spontaneous scenarios (bottom). Errors in the predicted words compared to the GT are highlighted in red. Faces are blurred for … view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Examples of generated dense captions using ShareCaptioner-Video. Strategy and ground truth transcript Prompt to ChatGPT Output from ChatGPT Type IV: Spontaneous-based Generation transcriptGT: We use marble bases that are imported directly from Italy. We have a speech f…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

38 extracted references · 27 canonical work pages

  1. [1]

    Afouras, T.; et al. 2022. Deep Audio-Visual Speech Recognition. IEEE Trans. Pattern Anal. , 44(12): 8717--8727

    work page 2022

  2. [2]

    A.; and Terry, M

    Bradley, R. A.; and Terry, M. E. 1952. Rank analysis of incomplete block designs: I. The method of paired comparisons. Biometrika, 39(3/4): 324--345

    1952

  3. [3]

    Bu, H.; et al. 2017. AISHELL-1: An open-source Mandarin speech corpus and a speech recognition baseline. In O-COCOSDA, 1--5

    work page 2017

  4. [4]

    Caglayan, O.; et al. 2019. Multimodal Grounding for Sequence-to-sequence Speech Recognition. In ICASSP, 8648--8652

    work page 2019

  5. [5]

    Chen, L.; Wei, X.; Li, J.; Dong, X.; Zhang, P.; Zang, Y.; Chen, Z.; Duan, H.; Lin, B.; Tang, Z.; et al. 2024. ShareGPT4Video: Improving Video Understanding and Generation with Better Captions. arXiv preprint arXiv:2406.04325

    arXiv 2024

  6. [6]

    Chiu, C.; Qin, J.; Zhang, Y.; Yu, J.; and Wu, Y. 2022. Self-supervised learning with random-projection quantizer for speech recognition. In Chaudhuri, K.; Jegelka, S.; Song, L.; Szepesv \' a ri, C.; Niu, G.; and Sabato, S., eds., International Conference on Machine Learning, ICML 2022, 17-23 July 2022, Baltimore, Maryland, USA , volume 162 of Proceedings ...

    work page 2022

  7. [7]

    S.; et al

    Chung, J. S.; et al. 2017. Lip Reading Sentences in the Wild. In CVPR, 3444--3453

    work page 2017

  8. [8]

    Gabeur, V.; et al. 2022. AVATAR: Unconstrained Audiovisual Speech Recognition. In Interspeech, 2818--2822

    work page 2022

Show all 38 references
  1. [9]

    Ghorbani, S.; et al. 2021. Listen, Look and Deliberate: Visual Context-Aware Speech Recognition Using Pre-Trained Text-Video Representations. In IEEE SLT Workshop, 621--628

    2021

  2. [10]

    Grauman, K.; et al. 2022. Ego4D: Around the World in 3, 000 Hours of Egocentric Video. In CVPR, 18973--18990

    2022

  3. [11]

    A.; and Est \` e ve, Y

    Hernandez, F.; Nguyen, V.; Ghannay, S.; Tomashenko, N. A.; and Est \` e ve, Y. 2018. TED-LIUM 3: Twice as Much Data and Corpus Repartition for Experiments on Speaker Adaptation. In Karpov, A.; Jokisch, O.; and Potapova, R., eds., Speech and Computer - 20th International Confer...

    2018

  4. [12]

    Kahn, J.; Rivi \` e re, M.; Zheng, W.; Kharitonov, E.; Xu, Q.; Mazar \' e , P.; Karadayi, J.; Liptchinsky, V.; Collobert, R.; Fuegen, C.; Likhomanenko, T.; Synnaeve, G.; Joulin, A.; Mohamed, A.; and Dupoux, E. 2020. Libri-Light: A Benchmark for ASR with Limited or No Supervisi...

    2020

  5. [13]

    Kim, K.; et al. 2022. E-Branchformer: Branchformer with Enhanced Merging for Speech Recognition. In IEEE SLT Workshop, 84--91

    2022

  6. [14]

    B.; et al

    Kumar, V. B.; et al. 2023. Visual Information Matters for ASR Error Correction. In ICASSP, 1--5

    2023

  7. [15]

    Li, J. 2021. Recent Advances in End-to-End Automatic Speech Recognition. arXiv:2111.01690

    2021 arXiv

  8. [16]

    Lu, Y.; Song, J.; Chang, X.; Bian, H.; Maiti, S.; and Watanabe, S. 2024. SynesLM: A Unified Approach for Audio-visual Speech Recognition and Translation via Language Model and Synthetic Data. arXiv:2408.00624

    2024 arXiv

  9. [17]

    Ma, P.; et al. 2021. End-To-End Audio-Visual Speech Recognition with Conformers. In ICASSP, 7613--7617

    2021

  10. [18]

    Miech, A.; et al. 2019. HowTo100M: Learning a Text-Video Embedding by Watching Hundred Million Narrated Video Clips. In 2019 IEEE/CVF ICCV , 2630--2640

    2019

  11. [19]

    Panayotov, V.; et al. 2015. Librispeech: An ASR corpus based on public domain audio books. In ICASSP, 5206--5210

    2015

  12. [20]

    Paraskevopoulos, G.; et al. 2020. Multimodal and Multiresolution Speech Recognition with Transformers. In ACL, 2381--2387

    2020

  13. [21]

    Peng, P.; et al. 2023 a . Prompting the Hidden Talent of Web-Scale Speech Models for Zero-Shot Task Generalization. In Interspeech, 396--400

    2023

  14. [22]

    Peng, Y.; et al. 2023 b . Reproducing Whisper-Style Training Using An Open-Source Toolkit And Publicly Available Data. In ASRU, 1--8

    2023

  15. [23]

    Peng, Y.; et al. 2024. OWSM v3.1: Better and Faster Open Whisper-Style Speech Models based on E-Branchformer. arXiv:2401.16658

    2024 arXiv

  16. [24]

    Prabhavalkar, R.; et al. 2024. End-to-End Speech Recognition: A Survey. IEEE ACM Trans. Audio Speech Lang. Process. , 32: 325--351

    2024

  17. [25]

    Pratap, V.; Xu, Q.; Sriram, A.; Synnaeve, G.; and Collobert, R. 2020. MLS: A Large-Scale Multilingual Dataset for Speech Research. In Meng, H.; Xu, B.; and Zheng, T. F., eds., 21st Annual Conference of the International Speech Communication Association, Interspeech 2020, Virtu...

    2020

  18. [26]

    Radford, A.; et al. 2021. Learning Transferable Visual Models From Natural Language Supervision. In ICML, volume 139, 8748--8763

    2021

  19. [27]

    Radford, A.; et al. 2023. Robust Speech Recognition via Large-Scale Weak Supervision. In ICML, volume 202, 28492--28518

    2023

  20. [28]

    Rafailov, R.; Hejna, J.; Park, R.; and Finn, C. 2024. From r to Q\( ^ * \): Your Language Model is Secretly a Q-Function. CoRR, abs/2404.12358

    2024 arXiv

  21. [29]

    D.; Ermon, S.; and Finn, C

    Rafailov, R.; Sharma, A.; Mitchell, E.; Manning, C. D.; Ermon, S.; and Finn, C. 2023. Direct Preference Optimization: Your Language Model is Secretly a Reward Model. In Oh, A.; Naumann, T.; Globerson, A.; Saenko, K.; Hardt, M.; and Levine, S., eds., Advances in Neural Informat...

    2023

  22. [30]

    Sanabria, R.; et al. 2018. How2: A Large-scale Dataset for Multimodal Language Understanding. abs/1811.00347

    2018 arXiv

  23. [31]

    H.; et al

    Seo, P. H.; et al. 2023. AVFormer: Injecting Vision into Frozen Speech Models for Zero-Shot AV-ASR . In CVPR, 22922--22931

    2023

  24. [32]

    Vaswani, A.; et al. 2017. Attention is All you Need. In NeurIPS, 5998--6008

    2017

  25. [33]

    Y.; Xu, N.; Zhang, S.; Poon, H.; and Chen, M

    Wang, F.; Zhou, W.; Huang, J. Y.; Xu, N.; Zhang, S.; Poon, H.; and Chen, M. 2024. mDPO: Conditional Preference Optimization for Multimodal Large Language Models. CoRR, abs/2406.11839

    2024 arXiv

  26. [34]

    Zeng, J.; Meng, F.; Yin, Y.; and Zhou, J. 2024. Teaching Large Language Models to Translate with Comparison. In Wooldridge, M. J.; Dy, J. G.; and Natarajan, S., eds., Thirty-Eighth AAAI Conference on Artificial Intelligence, AAAI 2024, Thirty-Sixth Conference on Innovative App...

    2024

  27. [35]

    Zhang, R.; Gui, L.; Sun, Z.; Feng, Y.; Xu, K.; Zhang, Y.; Fu, D.; Li, C.; Hauptmann, A.; Bisk, Y.; and Yang, Y. 2024. Direct Preference Optimization of Video Large Multimodal Models from Language Model Reward. arXiv:2404.01258

    2024 arXiv

  28. [36]

    Zhu, K.; Zhao, L.; Ge, Z.; and Zhang, X. 2024. Self-Supervised Visual Preference Alignment. CoRR, abs/2404.10501

    2024 arXiv

  29. [37]

    , " * write output.state after.block = add.period write newline

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint howpublished institution isbn journal key month note number organization pages publisher school series title type volume year label extra.label sort.label short.list INTEGERS output.state before.all...

  30. [38]

    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 gl...

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.