ComplexityWorld: Benchmarking Vision-Language Models on Verifiable Visual Decision Making

Hanyu Li; Ningxin Pan; Yehui Tang

REVIEW 2 major objections 5 minor 27 references

ComplexityWorld: Benchmarking Vision-Language Models on Verifiable Visual Decision Making

T0 review · 2 major / 5 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 ComplexityWorld presents evidence that current vision-language models can recognize images but often cannot turn visual evidence into a complete, constraint-satisfying decision: on 390 verifier-checked tasks, only GPT-5.6-Sol exceeds 40%…

desk verdict A solid benchmark with a real diagnostic result; exact scores are provisional until verifiers and images get independent validation. read the letter →

arxiv 2608.07584 v1 pith:WIPE5KEZ submitted 2026-08-05 cs.CV

Ningxin Pan , Hanyu Li , Yehui Tang This is my paper · ORCID

classification cs.CV

keywords vision-languagemodelsvisualdecisionmakingverifiablebenchmarkexecutableverifierconstraintsatisfactionacceptanceratevisual-to-decisionbottleneckrepresentationsensitivity

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

ComplexityWorld tests whether vision-language models can use an image as the sole evidence for a complete decision—a route, schedule, placement, or assignment whose parts must jointly satisfy global constraints—and return a solution a hidden verifier accepts. The paper builds 390 tasks spanning 39 visual worlds and 29 decision categories, each generated from a hidden structured specification and rendered as a scene. Under direct inference every evaluated model except GPT-5.6-Sol stays below 40% verifier acceptance rate, while GPT-5.6-Sol reaches 75.6%. Giving the same instances as explicit structured records lifts performance by 21 to 41 points, and equivalent problems in different visual presentations differ by up to 38.3 points. The paper concludes that a persistent visual-to-decision bottleneck remains: current models are much better at solving an explicit problem than at recovering the problem from pixels, and extra inference via agent scaffolds gives only smaller, model-dependent gains.

What carries the argument

The load-bearing machinery is a generate-render-verify pipeline. A seeded generator creates a solvable problem from a hidden structured specification; a visual-world adapter renders it as a 1600-by-1000 image using domain-specific names, icons, and layouts; and a task-specific verifier program checks every constraint, accepting any feasible solution rather than one serialized reference answer. The model sees only the image, instruction, and output schema, so the image is the only source of decision evidence. The same pipeline produces matched diagnostics: answer-free explicit-structure records hold the decision problem and verifier fixed while removing the image, and multiple adapters render identical hidden specifications in different visual forms. The primary metric is verifier acceptance rate (VAR), the fraction of returned decisions that parse, refer to valid entities, and satisfy the hidden verifier.

What would settle it

Present each model with the same task twice: once as the rendered image and once as a faithful textual transcription of the image's entities and constraints, then compare verifier acceptance rates; if the gap vanished, the bottleneck would be visual recovery, while a persistent gap would point to solution construction. In parallel, check the verifiers themselves by independently re-implementing a sample from the released task specifications and confirming agreement on the published 596 altered outputs, since the reported VAR is only meaningful if the verifiers enforce the intended constraints.

Watch

Extended reading notes

Core claim

The central claim is that reliable visual decision making requires two abilities that current VLMs do not have in equal measure: recovering the decision-relevant structure from visual evidence, and constructing a globally consistent solution from that structure. The paper's evidence is the gap between conditions: direct image inference yields low verifier acceptance rates (75.6% for GPT-5.6-Sol, 39.5% for Qwen3.7-Plus, 32.3% for Gemini-3.5-Flash, 24.4% for MiMo-v2.5), but replacing the image with an answer-free structured record of the same instances improves Qwen by 40.6 points and GPT by 21.4 points, and identical underlying problems presented in three matched visual forms produce gaps of up to 38.3 points. The same pattern appears across the eight constraint-signature families: strong performance on one family does not predict performance on another. The paper therefore concludes that strong solution construction in one representation does not imply reliable visual decision making across representations—success depends on how decision information is organized and presented, not just on model scale or reasoning effort.

Load-bearing premise

The whole result rests on the assumption that the verifiers enforce exactly the constraints the tasks intend and that every image shows everything needed without revealing the answer; the paper's own checks of this are behavioral (altered outputs and visual inspection), not an independent correctness proof.

Editorial extensions

If this is right

Success on perception-heavy benchmarks should not be read as success at decision making; a model can answer visual questions while failing to produce a verifier-accepted complete solution.
Benchmark scores are presentation-sensitive: reporting one rendering per problem can overstate or understate capability, because reorganizing identical decision information moved acceptance by tens of points.
Scaling inference alone is unlikely to close the gap: the two agent scaffolds tested improved their base models by 7.4 and 3.8 points, with the smaller gain statistically indistinguishable from zero.
Explicit structured inputs are a strong intervention, so hybrid systems that transcribe images into structured problem records before solving may outperform end-to-end VLMs on constrained decisions.
The released generators, frozen tasks, and verifiers give a reproducible target for training on verified outcomes, with the paper's suggested discipline of withholding whole visual-world adapters when testing generalization.

Reading between the lines

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

A testable extension the paper does not run: train on the frozen main panel and evaluate on the other available templates from the released 113-template library, holding out whole visual-world adapters, to measure out-of-distribution generalization of any verifier-trained model.
If the bottleneck is as persistent as measured, interface design becomes a first-order lever: presenting pairwise relations explicitly (adjacency tables, cost tables, traversability graphs) may be a cheap deployment fix, though the paper does not isolate which visual factor causes the gain because several change together.
The benchmark's distinction between decision categories and constraint-signature families could serve as a diagnostic grid for future model releases: reporting VAR per family would show whether a model's improvement comes from better visual extraction or better constraint construction, since the two map to different failure profiles.
One could turn the presentation-sensitivity result into a robustness metric: a model that solves the same problem in several renderings is more reliable than one that peaks in a single format, and the paper's all-three-success statistic is a step in that direction.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

First, the takeaway: this is a useful benchmark with a real diagnostic finding, but the exact scores should be treated as provisional until the verifiers and images get independent validation. The authors combine hidden structured specifications, an executable verifier that accepts any feasible solution, matched explicit-structure records, and equivalent-presentation controls. That's a thoughtful design. The central result—models do much better when the same decision problem is given as a structured record than as an image—is robust enough to survive some verifier looseness, and the 0/234 no-evidence control plus the 596-output sensitivity test show good discipline.

The main soft spot is the load-bearing premise: each verifier correctly implements the intended constraints, and each rendered image contains complete, unambiguous evidence. The paper concedes in Section 3.3 that its validation is behavioral—generation and verification share task-specific code. Seven of 295 altered decision outputs were accepted by verifiers; those are described as feasible alternatives, but without the release we cannot check whether they are true alternatives or loopholes. The blinded image-quality check is described only at a high level. The headline 75.6% vs. below-40% gap is big, so the bottleneck conclusion likely stands, but the exact numbers and per-category comparisons need the release to be credible.

Two smaller issues: the main table lacks confidence intervals, and the equivalent-problem study changes several visual factors together, so it establishes presentation sensitivity rather than isolating a cause. Those limits are disclosed, but they should be remembered when citing.

The paper is well-scoped, cites relevant work, and does not overclaim: it is a benchmark plus diagnostics, not a training study. It deserves a serious referee. My recommendation: send it to review, but ask for the release and for an independent audit of a sample of verifiers before the exact scores are relied on.

Referee Report

2 major / 5 minor

Summary. This paper introduces ComplexityWorld, a benchmark of 390 visual decision tasks drawn from 39 template-world pairs across 29 decision categories. Each task is generated from a hidden structured specification, rendered as a 1600-by-1000 image, and scored by an executable verifier that accepts any feasible solution, yielding a verifier acceptance rate (VAR, Eq. 1). Under direct inference, GPT-5.6-Sol reaches 75.6% VAR while the other three evaluated models remain below 40% in the observed sample. The authors find that replacing images with answer-free explicit-structure records raises Qwen3.7-Plus by 40.6 points and GPT-5.6-Sol by 21.4 points, that matched visual presentations produce gaps up to 38.3 points, and that agent scaffolds give smaller, model-dependent gains. They conclude that current VLMs exhibit a persistent visual-to-decision bottleneck that additional inference alone does not remove.

Significance. If the benchmark's verifiers and images are valid, this is a valuable contribution: it provides exact constraint-based evaluation with multiple acceptable solutions, controlled representation diagnostics, and a clear bottleneck finding with practical implications for VLM design. The paper is notably honest about its limitations, including the behavioral-only verifier check, the confounded presentation study, and the lack of causal attribution, which strengthens trust in its claims. The main risk is that the central metric inherits unverified assumptions about verifier correctness and image-evidence completeness; addressing these would make the leaderboard and bottleneck conclusions solid.

major comments (2)

[Section 3.3 (Construction Pipeline; verifier stress test)] The sole validation of verifier correctness is the shared-code sensitivity test, which the authors themselves describe as behavioral evidence rather than an independent correctness validation. Because VAR in Eq. (1) is the only scoring metric and the bottleneck claim in Sections 5.1 and 7 depends on every verifier enforcing exactly the constraints intended for all 390 tasks, this is load-bearing. Please add an independent check of verifier behavior for at least a random stratified sample of tasks (e.g., human adjudication of accepted/rejected outputs against the rendered images, or an independently reimplemented reference verifier), report agreement, and specify whether the seven accepted decision changes from the stress test are genuinely feasible alternative solutions under the stated task constraints.
[Section 3.3 (Rendering and QC; blinded check)] Image completeness and leakage are equally load-bearing: a low VAR on a task whose image omits or ambiguously presents a constraint would reflect a flawed task, not a model limitation. The blinded image-quality check is described only at a high level, with four example images reproduced in the Supplementary Material, and no repository link is provided. Please report the sample size, selection criteria, inter-annotator agreement, the full set of inspected images (or a stable link), and per-template pass/fail results for the automated leakage checks. This is necessary to rule out the alternative that some tasks do not present complete, unambiguous decision evidence.

minor comments (5)

[Table 1 and Abstract] Report 95% confidence intervals for the main VARs; for Qwen3.7-Plus, 154/390 = 39.5% has an approximate interval of [34.6%, 44.4%], so the abstract's claim that all non-GPT models 'remain below 40%' is not statistically robust.
[Section 3.3 and Section 7] The paper states that the frozen benchmark and construction pipeline will be released, but no repository URL, DOI, or other stable artifact link is given; without it, the reproducibility claims cannot be verified.
[Section 5.3 (Table 3)] Per-template results appear in Figure 3, but Table 3 aggregates them; add per-template confidence intervals and clarify that the Holm-adjusted McNemar correction applies to the two presentation comparisons within each template.
[Throughout] The metric name is rendered inconsistently as 'V AR' in the abstract and several places; use 'VAR' consistently.
[Section 5.2] The phrase 'answer-free explicit-structure record' is potentially confusing because the record contains all decision-relevant constraints; consider renaming it to 'solution-free structured record' to clarify that only the reference solution and verifier are excluded.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: benchmark scores are direct measurements against explicit verifiers; no fitted parameter is relabeled as a prediction and no self-citation is load-bearing.

full rationale

The paper's derivation chain is a benchmark measurement, not a derivation of predictions from fitted inputs. VAR (Eq. 1) is explicitly defined as the frequency with which a task-specific executable verifier accepts a model's complete decision; the direct-inference results in Table 1 are raw counts over the frozen 390-task panel, so 'GPT-5.6-Sol reaches 75.6%' is a direct measurement against a published evaluation rule. The visual-to-decision bottleneck claim is supported by the Section 5.2 diagnostic, which holds instances, public instruction, required outputs, parser, and verifier fixed and changes only the representation (image vs answer-free explicit-structure record), and by the Section 5.3 matched-presentation study, which holds the hidden specification and verifier fixed across three renderings. These are controlled empirical comparisons, not consequences of the definitions. No parameter is fitted and no fitted value is renamed as a prediction; no equation is equivalent by construction to another equation. The reference list contains no self-citations by Pan, Li, or Tang, so there is no load-bearing self-citation chain. The paper's own Section 3.3 caveat, 'Because generation and verification share task-specific code, it provides behavioral evidence rather than an independent correctness validation,' is a validity limitation about verifier correctness and image-evidence completeness, not a circular step; per the hard rules, such an assumption does not by itself establish circularity. Therefore no circular step meets the quoted-reduction standard, and the appropriate score is 0.

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

The central claim depends on the correctness of the verifiers and on the clarity of the rendered scenes, both of which are attested only by internal checks. No free parameters are fitted to data and no new physical or theoretical entities are introduced.

assumptions (3)

domain assumption The executable verifier for each task correctly implements the intended constraints.
All VAR results depend on this; the paper's stress test is behavioral because generator and verifier share task-specific code (Section 3.3).
domain assumption Rendered images present all decision-relevant entities and relations unambiguously without leaking the solution.
QC includes automated checks and blinded review of sampled low-VAR instances (Section 3.3).
domain assumption Ten instances per template-world pair and the six-template presentation study are sufficient for the reported aggregate and comparative conclusions.
Stated limitations in Section 6 acknowledge the limited instance counts and the small presentation-study scope.

how reviews work

0 comments

Cite this review

Pith. "Pith review of ComplexityWorld: Benchmarking Vision-Language Models on Verifiable Visual Decision Making." pith.science (2026). https://pith.science/paper/WIPE5KEZ

@misc{pith2026260807584,
  author       = {Pith},
  title        = {Pith review of: ComplexityWorld: Benchmarking Vision-Language Models on Verifiable Visual Decision Making},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WIPE5KEZ}},
  note         = {Machine review of arXiv:2608.07584}
}

read the original abstract

Vision-language models (VLMs) have made rapid progress in visual perception and increasingly support real-world tasks that depend on images. Many such tasks, however, require more than rec- ognizing what an image contains: a model must use visual evidence to make a complete decision whose parts jointly satisfy global constraints. We introduce COMPLEXITYWORLD, a benchmark of 390 tasks across 39 domain-inspired visual worlds and 29 decision categories. Each task is generated from a hidden structured specification, rendered as a visual scene, and scored by an exe- cutable verifier that accepts any feasible solution. Under direct inference, all evaluated models ex- cept GPT-5.6-Sol remain below 40% verifier ac- ceptance rate (VAR), while GPT-5.6-Sol reaches 75.6%. Performance improves substantially when the same decision information is made explicit in structured form, yet varies sharply across equiva- lent visual presentations. Agent scaffolds provide smaller, model-dependent gains. Together, these results reveal a persistent visual-to-decision bot- tleneck that additional inference alone does not remove.

Figures

Figures reproduced from arXiv: 2608.07584 by the authors.

**Figure 1.** Representative ComplexityWorld tasks organized by eight constraint-signature families. These family labels [PITH_FULL_IMAGE:figures/full_fig_p003_1.png] view at source ↗

**Figure 2.** Performance profiles across eight constraint-signature families on the frozen evaluation set. Each panel reports one [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Equivalent hidden specifications shown in three visual forms, labeled A, B, and C. [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗

Figures from the paper (1 more)

**Figure 4.** Figure 4: Example visual input. One matched answer-free explicit-structure entry is {"id":"T01","row":0,"column":0,"edges":[2,3,1,0,0,1]}. It specifies the tile’s grid position and six edge-colour indices; the full record contains no answer-bearing information. 60 matched groups…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

27 extracted references · 16 canonical work pages

[1]

2022 , publisher =

Masry, Ahmed and Long, Do Xuan and Tan, Jia Qing and Joty, Shafiq and Hoque, Enamul , booktitle =. 2022 , publisher =

work page 2022
[2]

Lu, Pan and Bansal, Hritik and Xia, Tony and Liu, Jiacheng and Li, Chunyuan and Hajishirzi, Hannaneh and Cheng, Hao and Chang, Kai-Wei and Galley, Michel and Gao, Jianfeng , booktitle =
[3]

Lawrence and Girshick, Ross , booktitle =

Johnson, Justin and Hariharan, Bharath and van der Maaten, Laurens and Fei-Fei, Li and Zitnick, C. Lawrence and Girshick, Ross , booktitle =. 2017 , doi =

work page 2017
[4]

2024 , doi =

Yue, Xiang and Ni, Yuansheng and Zheng, Tianyu and Zhang, Kai and Liu, Ruoqi and Zhang, Ge and Stevens, Samuel and Jiang, Dongfu and Ren, Weiming and Sun, Yuxuan and Wei, Cong and Yu, Botao and Yuan, Ruibin and Sun, Renliang and Yin, Ming and Zheng, Boyuan and Yang, Zhenzhu and Liu, Yibo and Huang, Wenhao and Sun, Huan and Su, Yu and Chen, Wenhu , booktit...

work page 2024
[5]

2025 , publisher =

Yue, Xiang and Zheng, Tianyu and Ni, Yuansheng and Wang, Yubo and Zhang, Kai and Tong, Shengbang and Sun, Yuxuan and Yu, Botao and Zhang, Ge and Sun, Huan and Su, Yu and Chen, Wenhu and Neubig, Graham , booktitle =. 2025 , publisher =

work page 2025
[6]

Journal of Data-centric Machine Learning Research , volume =

Benchmarking Robustness of Multimodal Image-Text Models under Distribution Shift , author =. Journal of Data-centric Machine Learning Research , volume =

work page
[7]

2019 , eprint =

On the Measure of Intelligence , author =. 2019 , eprint =

2019
[8]

Chen, Liang and Xie, Weichu and Liang, Yiyan and He, Hongfeng and Zhao, Hans and Yang, Zhibo and Huang, Zhiqi and Wu, Haoning and Lu, Haoyu and Y. Charles and Bao, Yiping and Fan, Yuantao and Li, Guopeng and Shen, Haiyang and Chen, Xuanzhong and Xu, Wendong and Si, Shuzheng and Cai, Zefan and Chai, Wenhao and Huang, Ziqi and Liu, Fangfu and Liu, Tianyu an...

arXiv

Show all 27 references

[9]

2502.00698 , archivePrefix =

Cai, Huanqia and Yang, Yijun and Hu, Winston , year =. 2502.00698 , archivePrefix =

arXiv
[10]

2024 , publisher =

Singh, Shubhankar and Chaurasia, Purvi and Varun, Yerram and Pandya, Pranshu and Gupta, Vatsal and Gupta, Vivek and Roth, Dan , booktitle =. 2024 , publisher =

2024
[11]

2025 , eprint =

Ren, Yufan and Tertikas, Konstantinos and Maiti, Shalini and Han, Junlin and Zhang, Tong and S. 2025 , eprint =

2025
[12]

2504.10885 , archivePrefix =

Zhang, Zeyu and Chen, Zijian and Zhang, Zicheng and Sun, Yuze and Tian, Yuan and Jia, Ziheng and Li, Chunyi and Liu, Xiaohong and Min, Xiongkuo and Zhai, Guangtao , year =. 2504.10885 , archivePrefix =

arXiv
[13]

2026 , eprint =

Pencil Puzzle Bench: A Benchmark for Multi-Step Verifiable Reasoning , author =. 2026 , eprint =

2026
[14]

2403.01777 , archivePrefix =

Fan, Lizhou and Hua, Wenyue and Li, Xiang and Zhu, Kaijie and Jin, Mingyu and Li, Lingyao and Ling, Haoyang and Chi, Jinkui and Wang, Jindong and Ma, Xin and Zhang, Yongfeng , year =. 2403.01777 , archivePrefix =

arXiv
[15]

2025 , publisher =

Mayer, Julius and Ballout, Mohamad and Jassim, Serwan and Nosrat Nezami, Farbod and Bruni, Elia , booktitle =. 2025 , publisher =

2025
[16]

2507.23382 , archivePrefix =

Ji, Yiyan and Chen, Haoran and Chen, Qiguang and Wu, Chengyue and Qin, Libo and Che, Wanxiang , year =. 2507.23382 , archivePrefix =

arXiv
[17]

2602.08367 , archivePrefix =

Wang, Zexuan and Yang, Chenghao and Que, Yingqi and Yang, Zhenzhu and Yuan, Huaqing and Wang, Yiwen and Jiang, Zhengxuan and Fang, Shengjie and Wu, Zhenhe and Wang, Zhaohui and Yao, Zhixin and Liu, Jiashuo and Ren, Jincheng and Li, Yuzhen and Yang, Yang and Liu, Jiaheng and Ji...
[18]

2026 , eprint =

Li, Zhong and Huang, Qi and Zhu, Yuxuan and Mohammadi Amiri, Mohammad and van Stein, Niki and B. 2026 , eprint =

2026
[19]

2606.19787 , archivePrefix =

Li, Jiajun and Cai, Mingshu and Li, Yixuan and Ding, Yu and Hou, Ran and Nie, Guanyu and Han, Xiongwei and Wang, Wanyuan , year =. 2606.19787 , archivePrefix =

arXiv
[20]

2025 , eprint =

Assessing the Value of Visual Input: A Benchmark of Multimodal Large Language Models for Robotic Path Planning , author =. 2025 , eprint =

2025
[21]

Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , booktitle =
[22]

2303.11381 , archivePrefix =

Yang, Zhengyuan and Li, Linjie and Wang, Jianfeng and Lin, Kevin and Azarnasab, Ehsan and Ahmed, Faisal and Liu, Zicheng and Liu, Ce and Zeng, Michael and Wang, Lijuan , year =. 2303.11381 , archivePrefix =

arXiv
[23]

Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =

Sur. Proceedings of the IEEE/CVF International Conference on Computer Vision , pages =. 2023 , doi =

2023
[24]

Advances in Neural Information Processing Systems , year =

Visual Sketchpad: Sketching as a Visual Chain of Thought for Multimodal Language Models , author =. Advances in Neural Information Processing Systems , year =
[25]

Xie, Tianbao and Zhang, Danyang and Chen, Jixuan and Li, Xiaochuan and Zhao, Siheng and Cao, Ruisheng and Hua, Toh Jing and Cheng, Zhoujun and Shin, Dongchan and Lei, Fangyu and Liu, Yitao and Xu, Yiheng and Zhou, Shuyan and Savarese, Silvio and Xiong, Caiming and Zhong, Victo...
[26]

Complexity of Computer Computations , editor =

Reducibility Among Combinatorial Problems , author =. Complexity of Computer Computations , editor =. 1972 , publisher =

1972
[27]

1979 , publisher =

Computers and Intractability: A Guide to the Theory of NP-Completeness , author =. 1979 , publisher =

1979

Pith tools

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

[1] [1]

2022 , publisher =

Masry, Ahmed and Long, Do Xuan and Tan, Jia Qing and Joty, Shafiq and Hoque, Enamul , booktitle =. 2022 , publisher =

work page 2022

[2] [2]

Lu, Pan and Bansal, Hritik and Xia, Tony and Liu, Jiacheng and Li, Chunyuan and Hajishirzi, Hannaneh and Cheng, Hao and Chang, Kai-Wei and Galley, Michel and Gao, Jianfeng , booktitle =

[3] [3]

Lawrence and Girshick, Ross , booktitle =

Johnson, Justin and Hariharan, Bharath and van der Maaten, Laurens and Fei-Fei, Li and Zitnick, C. Lawrence and Girshick, Ross , booktitle =. 2017 , doi =

work page 2017

[4] [4]

2024 , doi =

Yue, Xiang and Ni, Yuansheng and Zheng, Tianyu and Zhang, Kai and Liu, Ruoqi and Zhang, Ge and Stevens, Samuel and Jiang, Dongfu and Ren, Weiming and Sun, Yuxuan and Wei, Cong and Yu, Botao and Yuan, Ruibin and Sun, Renliang and Yin, Ming and Zheng, Boyuan and Yang, Zhenzhu and Liu, Yibo and Huang, Wenhao and Sun, Huan and Su, Yu and Chen, Wenhu , booktit...

work page 2024

[5] [5]

2025 , publisher =

Yue, Xiang and Zheng, Tianyu and Ni, Yuansheng and Wang, Yubo and Zhang, Kai and Tong, Shengbang and Sun, Yuxuan and Yu, Botao and Zhang, Ge and Sun, Huan and Su, Yu and Chen, Wenhu and Neubig, Graham , booktitle =. 2025 , publisher =

work page 2025

[6] [6]

Journal of Data-centric Machine Learning Research , volume =

Benchmarking Robustness of Multimodal Image-Text Models under Distribution Shift , author =. Journal of Data-centric Machine Learning Research , volume =

work page

[7] [7]

2019 , eprint =

On the Measure of Intelligence , author =. 2019 , eprint =

2019

[8] [8]

Chen, Liang and Xie, Weichu and Liang, Yiyan and He, Hongfeng and Zhao, Hans and Yang, Zhibo and Huang, Zhiqi and Wu, Haoning and Lu, Haoyu and Y. Charles and Bao, Yiping and Fan, Yuantao and Li, Guopeng and Shen, Haiyang and Chen, Xuanzhong and Xu, Wendong and Si, Shuzheng and Cai, Zefan and Chai, Wenhao and Huang, Ziqi and Liu, Fangfu and Liu, Tianyu an...

arXiv