Pith. sign in

REVIEW 3 major objections 5 minor 92 references

Vision-language models already do backward prediction; training them to recover which of their own reasonings precede a given answer yields label-free self-reflection and higher accuracy on hard visual tasks.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · grok-4.5

2026-07-13 06:44 UTC pith:LO64BBRY

load-bearing objection Solid unsupervised self-reflection plug-in for VLMs that turns the model’s own samples into a backward-prediction training signal; gains are real, neuroscience framing is light. the 3 major comments →

arxiv 2607.07361 v2 pith:LO64BBRY submitted 2026-07-08 cs.CV

BUS: Brain-Inspired Unsupervised Self-Reflection via Backward Prediction for Multimodal Reasoning

Jiacheng Yang , Tongying Xiao , Yunkai Dang , Cong Wang , Yuekun Yang , Qi Fan , Tianyu Ding , Wenbin Li
show 2 more authors
Feng Miao Yang Gao
This is my paper
classification cs.CV
keywords vision-language modelsself-reflectionbackward predictionunsupervised trainingmultimodal reasoninglabel-free learningfine-grained visual reasoning
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Complex visual questions demand consistent, fine-grained reasoning, yet today’s vision-language models often produce unreliable chains of thought and existing self-reflection fixes need large annotated datasets. This paper first shows that mainstream models already possess a brain-like ability called backward prediction: given a future answer they can identify which earlier reasoning steps were likely to produce it. Building on that observation the authors introduce BUS, a training loop that samples many reasoning–answer pairs from the model itself, then asks the model to recover the reasonings that precede each answer. Because the correct predecessors are known from the sampling step, the model generates its own learning signal and can be fine-tuned with ordinary supervised or reinforcement-learning methods without any human labels. On high-resolution visual benchmarks the resulting models gain several percentage points over the same base model, supporting the claim that strengthening backward prediction is a practical route to better reflective reasoning.

Core claim

Mainstream vision-language models already perform backward prediction analogous to the brain’s use of predecessor representations. By sampling multiple reasoning–answer pairs and then training the model to identify which reasonings lead to a known answer, an explicit self-verification signal is obtained on unlabeled data. The resulting framework, BUS, therefore improves reflective reasoning and task accuracy without annotated reflection data and works as a plug-in for both supervised fine-tuning and reinforcement learning.

What carries the argument

BUS (Brain-inspired Unsupervised Self-reflection): after forward sampling of reasoning–answer pairs, a new query is built that asks the model which of those reasonings can lead to a given answer category; matching the known predecessors supplies the unsupervised target used by SFT or GRPO, encouraging path–answer consistency via a Bayes-derived objective.

Load-bearing premise

Teaching a model to recover which of its own sampled reasonings lead to a sampled answer will improve true answer accuracy on new questions, even when many of those sampled answers are themselves incorrect.

What would settle it

Measure both backward-prediction accuracy on the constructed predecessor queries and ordinary answer accuracy on the same held-out images after BUS training; if predecessor recovery rises sharply while task accuracy stays flat or falls, the claimed transfer from self-verification to correctness is falsified.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Self-reflection can be trained without costly human-annotated reflection datasets.
  • The same plug-in works with both supervised fine-tuning and reinforcement-learning pipelines and across different base models.
  • Gains transfer to out-of-distribution high-resolution benchmarks, not only the training distribution.
  • Even training solely on incorrect sampled answers still yields gains, because the model must still learn logical path–answer associations.
  • Larger base models continue to benefit, indicating the method scales with model size.

Where Pith is reading between the lines

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

  • The same predecessor-recovery objective could be applied to pure language models to obtain chain-of-thought self-correction without process supervision.
  • If backward prediction is already latent, a lightweight test-time prompt that forces predecessor queries might give free reflection gains with no fine-tuning at all.
  • Systematic errors could still be reinforced when many sampled answers share the same wrong reasoning pattern; monitoring answer diversity during sampling is a natural safeguard to test.
  • Combining BUS with visual-grounding or region-focus methods may amplify gains, because better path–answer consistency could stabilize attention on the right image regions.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

3 major / 5 minor

Summary. The paper argues that mainstream VLMs already exhibit backward-prediction behavior (predecessor-like choice biases) analogous to human neuroscience findings, and introduces BUS, a label-free plug-in that exploits this for self-reflective training. After repeated forward sampling of reasoning–answer pairs on unlabeled images, the model is trained (via SFT or GRPO) to recover which of its own reasonings precede a given answer category, yielding an unsupervised consistency signal (Eqs. 5–8). Initialized from Qwen3-VL-8B, BUS reports gains of +5.8% on MME-RealWorld-Lite (in-distribution, unsupervised), +7.7–8.0% on HR-Bench 4K/8K, +6.3% on V*, and smaller lifts on general math/reasoning suites, remaining competitive with some supervised self-reflection baselines while requiring no ground-truth labels.

Significance. If the mechanism transfers as claimed, BUS offers a practical, annotation-free route to improve fine-grained multimodal reasoning and is model-agnostic (SFT/RL-compatible). The empirical package is a clear strength: consistent gains over the same base model, an incorrect-only ablation (Table 3), sensitivity to sample count n, scaling to 32B, and out-of-distribution transfer. The theoretical framing via Bayes and a KL-style objective (Eq. 8) is transparent. These elements make the work a useful contribution to unsupervised self-reflection even if the neuroscience analogy remains interpretive rather than mechanistic.

major comments (3)
  1. [§3, Fig. 2] §3 / Fig. 2: The pilot only demonstrates choice bias (≥65%) under an artificial divergent state-space with planted rewards that make SR and PR expected values differ. This is an in-context decision probe, not evidence that VLMs maintain or use predecessor representations during natural visual reasoning, nor that the same computation improves visual grounding. The leap from this probe to “backward prediction is key to advancing reflective reasoning” (abstract/conclusion) is therefore under-supported and should be tempered or replaced by a more direct test (e.g., intervention on PR-like structure and measured effect on grounding errors).
  2. [§4, Eqs. 5–8, Table 3] §4, Eqs. 5–8 and Stage II: Training targets a'_g are constructed solely from the model’s own forward samples (a'_g = {y_i | a_i = c_j}). The resulting objective (Eq. 8) aligns p_θ(c|y,x) to the empirical joint of those samples. Table 3 shows that even the Incorrect-only variant still raises MME-RW-Lite overall from 48.6% to 50.7%, which is consistent with tighter answer–reasoning association but does not by itself establish that the recovered paths become more factually correct on held-out images. Without additional diagnostics (faithfulness metrics, error-type breakdowns, or comparison against a pure self-consistency baseline that does not use the backward formulation), the causal link from the claimed mechanism to the reported OOD accuracy gains remains the weakest load-bearing assumption.
  3. [Abstract, §1, Fig. 3] Abstract and §1 claim that BUS supplies “explicit reflective behavior during test time,” yet the framework (Fig. 3) and all reported evaluations use standard forward inference after training. If test-time reflection is only implicit (better forward policy), the contrast with prior self-reflection methods is overstated and should be clarified or demonstrated with an explicit test-time backward procedure and its incremental benefit.
minor comments (5)
  1. [Table 1] Table 1: Several self-reflection and supervised baselines report incomplete cells (—). For fairness, either fill the missing numbers under a common evaluation protocol or explicitly note that those entries are unavailable.
  2. [Eq. (6)] Eq. (6) partial-credit reward is sensible but its sensitivity (threshold for “subset,” effect of |a'|/|a'_g|) is not ablated; a short note or curve would strengthen the RL variant.
  3. [Fig. 5] Fig. 5 visualization is helpful; adding the base model’s full selected choice and a short quantitative summary of how often BUS corrects the base model’s backward errors would make the qualitative claim more precise.
  4. [Abstract] Minor wording: “facilitateself-reflectivereasoning” and similar concatenated tokens appear in the abstract/intro (likely PDF extraction artifacts); clean for camera-ready.
  5. [§2] Related Work could more sharply distinguish BUS from other unsupervised or self-consistency methods (e.g., self-consistency decoding, process-reward models without labels) to avoid overstating novelty of the “label-free” claim alone.

Circularity Check

1 steps flagged

Self-generated a'_g labels make the unsupervised signal self-referential by design, but external-benchmark gains are not forced by construction.

specific steps
  1. self definitional [Sec. 4 BUS Framework (definition of a'_g and correctness)]
    "the ground truth answer of x′_I&T is the previously sampled reasoning that precedes c_j, i.e., a′_g={y_i|a_i=c_j}. The model’s backward prediction is considered correct if a′=a′_g. Therefore, by comparing a′ and a′_g, we can provide explicit learning signals on data without any external supervision."

    a'_g is defined directly as the subset of the model's own forward samples that match the chosen answer category. The 'self-verification' signal and the subsequent SFT/GRPO updates (Eqs. 5–6) are therefore self-referential by construction: the model is trained to recover which of its own reasonings produced its own answers. This does not force external-task accuracy, but the unsupervised learning signal itself reduces to the model's current distribution.

full rationale

The paper's core derivation is an unsupervised self-consistency loop (forward sample {(y_i,a_i)}, construct backward query whose 'ground truth' a'_g is exactly the subset of those y_i that produced the chosen answer category, then SFT/GRPO on that signal). Equations 5–8 simply rewrite the resulting objective as alignment of p_θ(c|y,x) to the empirical joint of the model's own samples (Bayes + entropy + KL). This is intentionally self-referential and label-free, yet it does not algebraically force the reported accuracy lifts on MME-RealWorld-Lite, HR-Bench, V*, MathVerse, etc.; those remain ordinary empirical measurements on held-out data. The neuroscience pilot (Fig. 2) is an independent choice-bias experiment under artificial rewards and is not used to derive the training objective. No parameters are fitted to a subset and then re-predicted, no uniqueness theorem is imported from overlapping authors, and no known result is merely renamed. The mild circularity is therefore confined to the construction of the training signal itself; the central performance claims stay externally falsifiable and are not reduced to the inputs by definition. Score 2 reflects that single, non-load-bearing self-reference.

Axiom & Free-Parameter Ledger

3 free parameters · 3 axioms · 2 invented entities

The central claim rests on (1) treating VLMs as already capable of predecessor-style backward prediction, (2) treating self-generated reasoning–answer consistency as a sufficient unsupervised objective for better task accuracy, and (3) a small set of training knobs (sample count, partial-credit reward). No new physical entity is postulated; the main invented object is the BUS training procedure itself.

free parameters (3)
  • n (number of repeated samples per question) = 8
    Set to 8 in main experiments; sensitivity table uses 2/4/8. Directly controls diversity of reasonings and strength of the backward signal.
  • GRPO partial-credit reward R(a', a'_g) = 0 or |a'|/|a'_g|
    Hand-designed: 0 if any incorrect path selected, else |a'|/|a'_g|. Shapes the RL learning signal for multi-path selection.
  • GRPO group size G and other RL/SFT hyperparameters
    Referenced via TRL/GRPO but not fully specified in the main text; required to reproduce training dynamics.
axioms (3)
  • domain assumption Mainstream VLMs already perform backward prediction (predecessor representations) for decision-making, analogous to human OFC/HPC/striatum mechanisms.
    Section 3 pilot: ≥65% of choices match the backward expected-value ranking after in-context transition exposure. Load-bearing for the “brain-inspired” framing.
  • ad hoc to paper Consistency between a model’s own sampled reasonings and answers is a valid unsupervised learning signal that improves true visual-reasoning accuracy, including when answers are wrong.
    Stage II construction of a'_g and Table 3 BUS-GRPO-Incorrect; without this, label-free gains would not be expected.
  • standard math Bayes rewrite p(y|c,x) ∝ p(c|y,x)p(y|x) justifies optimizing log p(c|y,x) as the BUS objective.
    Equations 7–8; standard probability identity used to motivate the loss.
invented entities (2)
  • BUS (Brain-inspired Unsupervised Self-reflection) training loop no independent evidence
    purpose: Turn forward samples into backward-prediction questions that supply SFT/RL targets without external labels.
    Core method object of the paper; not previously defined in the cited literature.
  • Predecessor-representation interpretation of VLM reasoning paths no independent evidence
    purpose: Map neuroscience PRs onto sampled reasonings that “precede” an answer category.
    Used to motivate Stage II; independent evidence is only the authors’ pilot probe, not external PR measurements in VLMs.

pith-pipeline@v1.1.0-grok45 · 22209 in / 3141 out tokens · 35803 ms · 2026-07-13T06:44:12.165071+00:00 · methodology

0 comments
read the original abstract

Current Vision-Language Models (VLMs) often struggle to handle complex visual tasks that require consistent and fine-grained reasoning. Recent methods aim to train models to facilitate self-reflective reasoning, i.e., reviewing and improving the generated reasoning. However, they require large volumes of annotated data and lack explicit reflective behavior during test time. By contrast, humans perform explicit and efficient self-reflection through mechanisms such as backward prediction, i.e., predicting which current states are likely to precede a given future state. Inspired by neuroscience, this work proposes a novel solution to address these challenges. We first observe and investigate the phenomenon that mainstream VLMs can perform backward prediction, similar to the human brain. A label-free training framework named Brain-inspired Unsupervised Self-reflection (BUS) is proposed to leverage and exploit backward prediction capability to enhance reflective reasoning in complex visual tasks. BUS enables self-verification of reflective reasoning based on backward prediction, providing explicit learning signals under unsupervised conditions. In this way, BUS eliminates reliance on annotated data while improving reasoning performance. Designed as a model-agnostic plug-in, our framework is compatible with popular fine-tuning methods, such as Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL). Initialized from Qwen3-VL-8B, it improves HR-Bench-8K (+8.0%), HR-Bench-4K (+7.7%), V* Bench (+6.3%), and MME-RealWorld-Lite (+5.8%), proving backward prediction is key to advancing reflective reasoning.

Figures

Figures reproduced from arXiv: 2607.07361 by Cong Wang, Feng Miao, Jiacheng Yang, Qi Fan, Tianyu Ding, Tongying Xiao, Wenbin Li, Yang Gao, Yuekun Yang, Yunkai Dang.

Figure 1
Figure 1. Figure 1: Backward prediction process in question-answering [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: VLMs perform backward prediction. (a) Task environment. States are represented by images, and darker arrows denote [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: BUS Framework. BUS achieves effective label-free training through self-verification of reflective reasoning. In Stage I, we generate multiple reasoning–answer pairs through repeated sampling. In Stage II, we guide the model to perform brain-inspired backward prediction and fine-tune it on unlabeled data, removing the need for annotated datasets. 4 Brain-Inspired Unsupervised Self-Reflection In this section… view at source ↗
Figure 4
Figure 4. Figure 4: Answer accuracy compared with the larger foun [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Visualization results of Qwen3-VL-8B and our BUS-8B on a constructed backward-prediction question. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Backward prediction capability compared with the [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

92 extracted references · 22 linked inside Pith

  1. [2]

    Bai, S.; Chen, K.; Liu, X.; Wang, J.; Ge, W.; Song, S.; Dang, K.; Wang, P.; Wang, S.; Tang, J.; Zhong, H.; Zhu, Y.; Yang, M.; Li, Z.; Wan, J.; Wang, P.; Ding, W.; Fu, Z.; Xu, Y.; Ye, J.; Zhang, X.; Xie, T.; Cheng, Z.; Zhang, H.; Yang, Z.; Xu, H.; and Lin, J. 2025 b . Qwen2.5-VL Technical Report. arXiv:2502.13923

    Pith/arXiv arXiv 2025

  2. [4]

    Chen, L.; Li, J.; Dong, X.; Zhang, P.; Zang, Y.; Chen, Z.; Duan, H.; Wang, J.; Qiao, Y.; Lin, D.; and Zhao, F. 2024. Are We on the Right Way for Evaluating Large Vision-Language Models? In Advances in Neural Information Processing Systems, volume 37, 27056--27087

    2024

  3. [6]

    P.; and Press, C

    de Lange , F. P.; and Press, C. 2026. Forward and backward prediction in learning and perception. Current Opinion in Neurobiology, 96: 103144

    2026

  4. [7]

    Deng, Y.; Bansal, H.; Yin, F.; Peng, N.; Wang, W.; and Chang, K.-W. 2025. Openvlthinker: An early exploration to complex vision-language reasoning via iterative self-improvement. arXiv e-prints, arXiv--2503

    2025

  5. [10]

    Ding, Y.; and Zhang, R. 2025. Sherlock: Self-Correcting Reasoning in Vision-Language Models. In Advances in Neural Information Processing Systems, volume 38, 101638--101672

    2025

  6. [11]

    Fan, Z.; Zhang, J.; Li, R.; Zhang, J.; Chen, R.; Hu, H.; Wang, K.; Wang, P.; Qu, H.; Zhou, S.; et al. 2026. Vlm-3r: Vision-language models augmented with instruction-aligned 3d reconstruction. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition, 31054--31065

    2026

  7. [12]

    Google DeepMind . 2025. Gemini-2.5-Pro. https://deepmind.google/models/gemini/pro/

    2025

  8. [13]

    Guo, D.; Yang, D.; Zhang, H.; Song, J.; Wang, P.; Zhu, Q.; Xu, R.; Zhang, R.; Ma, S.; Bi, X.; et al. 2025. DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning. Nature, 645(8081): 633--638

    2025

  9. [18]

    Jian, P.; Wu, J.; Sun, W.; Wang, C.; Ren, S.; and Zhang, J. 2025. Look Again, Think Slowly: Enhancing Visual Reflection in Vision-Language Models. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 9251--9270

    2025

  10. [19]

    Lu, P.; Bansal, H.; Xia, T.; Liu, J.; Li, C.; Hajishirzi, H.; Cheng, H.; Chang, K.-W.; Galley, M.; and Gao, J. 2024. MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts. In International Conference on Learning Representations, volume 2024, 23439--23554

    2024

  11. [20]

    Ma, R.; Wang, P.; Liu, C.; Liu, X.; Chen, J.; Zhang, B.; Zhou, X.; Du, N.; and Li, J. 2025. S ^2 R : Teaching LLM s to Self-verify and Self-correct via Reinforcement Learning. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 22632--22654

    2025

  12. [21]

    Namboodiri, V. M. K.; and Stuber, G. D. 2021. The learning of prospective and retrospective cognitive maps within neural circuits. Neuron, 109(22): 3552--3575

    2021

  13. [22]

    OpenAI . 2025. OpenAI-o3. https://openai.com/index/introducing-o3-and-o4-mini/

    2025

  14. [23]

    V.; He, C.; and Wu, L

    Pan, Z.; Li, Y.; Lin, H.; Pei, Q.; Tang, Z.; Wu, W.; Ming, C.; Zhao, H. V.; He, C.; and Wu, L. 2025. LEMMA : Learning from Errors for M athe M atical Advancement in LLM s. In Findings of the Association for Computational Linguistics: ACL 2025, 11615--11639

    2025

  15. [24]

    Qiao, R.; Tan, Q.; Dong, G.; MinhuiWu, M.; Sun, C.; Song, X.; Wang, J.; GongQue, Z.; Lei, S.; Zhang, Y.; Wei, Z.; Zhang, M.; Qiao, R.; Zong, X.; Xu, Y.; Yang, P.; Bao, Z.; Diao, M.; Li, C.; and Zhang, H. 2025. We-Math: Does Your Large Multimodal Model Achieve Human-like Mathematical Reasoning? In Proceedings of the 63rd Annual Meeting of the Association f...

    2025

  16. [25]

    B.; and Eldar, E

    Sharp, P. B.; and Eldar, E. 2024. Humans adaptively deploy forward and backward prediction. Nature human behaviour, 8(9): 1726--1737

    2024

  17. [30]

    von Werra, L.; Belkada, Y.; Tunstall, L.; Beeching, E.; Thrush, T.; Lambert, N.; Huang, S.; Rasul, K.; and Gallouédec, Q. 2020. TRL: Transformer Reinforcement Learning. https://github.com/huggingface/trl

    2020

  18. [31]

    Wan, Z.; Dou, Z.; Liu, C.; Zhang, Y.; Cui, D.; Zhao, Q.; Shen, H.; Xiong, J.; Xin, Y.; Jiang, Y.; Tao, C.; He, Y.; Zhang, M.; and Yan, S. 2025. SRPO: Enhancing Multimodal LLM Reasoning via Reflection-Aware Reinforcement Learning. In Advances in Neural Information Processing Systems, volume 38, 153676--153713

    2025

  19. [33]

    Wang, H.; Qu, C.; Huang, Z.; Chu, W.; Lin, F.; and Chen, W. 2025 b . VL-Rethinker: Incentivizing Self-Reflection of Vision-Language Models with Reinforcement Learning. In Advances in Neural Information Processing Systems, volume 38, 30865--30891

    2025

  20. [35]

    Wang, W.; Ding, L.; Zeng, M.; Zhou, X.; Shen, L.; Luo, Y.; Yu, W.; and Tao, D. 2025 d . Divide, conquer and combine: A training-free framework for high-resolution image perception in multimodal large language models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 39, 7907--7915

    2025

  21. [40]

    Wu, P.; and Xie, S. 2024. V?: Guided Visual Search as a Core Mechanism in Multimodal LLMs. In Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), 13084--13094

    2024

  22. [41]

    Yang, S.; Niu, Y.; Liu, Y.; Ye, Y.; Lin, B.; and Yuan, L. 2026. Look-back: Implicit visual re-focusing in mllm reasoning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 40, 11694--11702

    2026

  23. [42]

    Yao, H.; Huang, J.; Wu, W.; Zhang, J.; Wang, Y.; Liu, S.; Wang, Y.; Song, Y.; Feng, H.; Shen, L.; and Tao, D. 2025. Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Search. In Advances in Neural Information Processing Systems, volume 38, 29918--29952

    2025

  24. [45]

    Zhang, Q.; Wang, D.; Qian, H.; Li, Y.; Zhang, T.; Huang, M.; Xu, K.; Li, H.; Yan, L.; and Qiu, H. 2025 a . Understanding the Dark Side of LLM s' Intrinsic Self-Correction. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 27066--27101

    2025

  25. [46]

    Zhang, R.; Jiang, D.; Zhang, Y.; Lin, H.; Guo, Z.; Qiu, P.; Zhou, A.; Lu, P.; Chang, K.-W.; Qiao, Y.; Gao, P.; and Li, H. 2025 b . MATHVERSE: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems? In Computer Vision -- ECCV 2024, 169--186. Springer Nature Switzerland

    2025

  26. [47]

    Zhang, Y.; Zhang, H.; Tian, H.; Fu, C.; Zhang, S.; Wu, J.; Li, F.; Wang, K.; Wen, Q.; Zhang, Z.; Wang, L.; and Jin, R. 2025 c . MME-RealWorld: Could Your Multimodal LLM Challenge High-Resolution Real-World Scenarios that are Difficult for Humans? In International Conference on Learning Representations, volume 2025, 89655--89701

    2025

  27. [52]

    Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education

    Clancey, William J. Communication, Simulation, and Intelligent Agents: Implications of Personal Intelligent Machines for Medical Education. Proceedings of the Eighth International Joint Conference on Artificial Intelligence (IJCAI-83)

  28. [53]

    Classification Problem Solving

    Clancey, William J. Classification Problem Solving. Proceedings of the Fourth National Conference on Artificial Intelligence

  29. [54]

    , title =

    Robinson, Arthur L. , title =. 1980 , doi =. https://science.sciencemag.org/content/208/4447/1019.full.pdf , journal =

    1980

  30. [55]

    New Ways to Make Microcircuits Smaller---Duplicate Entry

    Robinson, Arthur L. New Ways to Make Microcircuits Smaller---Duplicate Entry. Science

  31. [56]

    Clancey and Glenn Rennels , abstract =

    Diane Warner Hasling and William J. Clancey and Glenn Rennels , abstract =. Strategic explanations for a diagnostic consultation system , journal =. 1984 , issn =. doi:https://doi.org/10.1016/S0020-7373(84)80003-6 , url =

    doi:10.1016/s0020-7373(84)80003-6 1984

  32. [57]

    and Rennels, Glenn R

    Hasling, Diane Warner and Clancey, William J. and Rennels, Glenn R. and Test, Thomas. Strategic Explanations in Consultation---Duplicate. The International Journal of Man-Machine Studies

  33. [58]

    Poligon: A System for Parallel Problem Solving

    Rice, James. Poligon: A System for Parallel Problem Solving

  34. [59]

    Transfer of Rule-Based Expertise through a Tutorial Dialogue

    Clancey, William J. Transfer of Rule-Based Expertise through a Tutorial Dialogue

  35. [60]

    The Engineering of Qualitative Models

    Clancey, William J. The Engineering of Qualitative Models

  36. [61]

    2023 , eprint=

    Attention Is All You Need , author=. 2023 , eprint=

    2023

  37. [62]

    Pluto: The 'Other' Red Planet

    NASA. Pluto: The 'Other' Red Planet

  38. [63]

    Nature , volume=

    DeepSeek-R1 incentivizes reasoning in LLMs through reinforcement learning , author=. Nature , volume=. 2025 , publisher=

    2025

  39. [64]

    Science , volume=

    Mesolimbic dopamine release conveys causal associations , author=. Science , volume=. 2022 , publisher=

    2022

  40. [65]

    Neuron , volume=

    The learning of prospective and retrospective cognitive maps within neural circuits , author=. Neuron , volume=. 2021 , publisher=

    2021

  41. [66]

    Forward and backward prediction in learning and perception , journal =

    Floris P. Forward and backward prediction in learning and perception , journal =. 2026 , issn =

    2026

  42. [67]

    Nature human behaviour , volume=

    Humans adaptively deploy forward and backward prediction , author=. Nature human behaviour , volume=. 2024 , publisher=

    2024

  43. [68]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Vlm-3r: Vision-language models augmented with instruction-aligned 3d reconstruction , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  44. [69]

    GitHub repository , howpublished =

    Leandro von Werra and Younes Belkada and Lewis Tunstall and Edward Beeching and Tristan Thrush and Nathan Lambert and Shengyi Huang and Kashif Rasul and Quentin Gallouédec , title =. GitHub repository , howpublished =. 2020 , publisher =

    2020

  45. [70]

    2025 , howpublished=

    OpenAI-o3 , author=. 2025 , howpublished=

    2025

  46. [71]

    2025 , howpublished=

    Gemini-2.5-Pro , author=. 2025 , howpublished=

    2025

  47. [72]

    5: Visual Agentic Intelligence , author=

    Kimi K2. 5: Visual Agentic Intelligence , author=. arXiv preprint arXiv:2602.02276 , year=

    Pith/arXiv arXiv

  48. [73]

    arXiv preprint arXiv:2601.19798 , year=

    Youtu-VL: Unleashing Visual Potential via Unified Vision-Language Supervision , author=. arXiv preprint arXiv:2601.19798 , year=

    arXiv

  49. [74]

    arXiv preprint arXiv:2602.11858 , year=

    Zooming without Zooming: Region-to-Image Distillation for Fine-Grained Multimodal Perception , author=. arXiv preprint arXiv:2602.11858 , year=

    arXiv

  50. [75]

    5: Advancing open-source multimodal models in versatility, reasoning, and efficiency , author=

    Internvl3. 5: Advancing open-source multimodal models in versatility, reasoning, and efficiency , author=. arXiv preprint arXiv:2508.18265 , year=

    Pith/arXiv arXiv

  51. [76]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Look-back: Implicit visual re-focusing in mllm reasoning , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  52. [77]

    Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Search , volume =

    Yao, Huanjin and Huang, Jiaxing and Wu, Wenhao and Zhang, Jingyi and Wang, Yibo and Liu, Shunyu and Wang, Yingjie and Song, YuXin and Feng, Haocheng and Shen, Li and Tao, Dacheng , booktitle =. Mulberry: Empowering MLLM with o1-like Reasoning and Reflection via Collective Monte Carlo Tree Search , volume =

  53. [78]

    Look Again, Think Slowly: Enhancing Visual Reflection in Vision-Language Models

    Jian, Pu and Wu, Junhong and Sun, Wei and Wang, Chen and Ren, Shuo and Zhang, Jiajun. Look Again, Think Slowly: Enhancing Visual Reflection in Vision-Language Models. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025

    2025

  54. [79]

    arXiv preprint arXiv:2602.18746 , year=

    MIRROR: Multimodal Iterative Reasoning via Reflection on Visual Regions , author=. arXiv preprint arXiv:2602.18746 , year=

    Pith/arXiv arXiv

  55. [80]

    arXiv preprint arXiv:2602.13949 , year=

    Experiential reinforcement learning , author=. arXiv preprint arXiv:2602.13949 , year=

    arXiv

  56. [81]

    Sherlock: Self-Correcting Reasoning in Vision-Language Models , volume =

    Ding, Yi and Zhang, Ruqi , booktitle =. Sherlock: Self-Correcting Reasoning in Vision-Language Models , volume =

  57. [82]

    arXiv preprint arXiv:2501.04519 , year=

    RStar-math: Small LLMs can master math reasoning with self-evolved deep thinking , author=. arXiv preprint arXiv:2501.04519 , year=

    Pith/arXiv arXiv

  58. [83]

    arXiv preprint arXiv:2504.02902 , year=

    Beyond Accuracy: The Role of Calibration in Self-Improving Large Language Models , author=. arXiv preprint arXiv:2504.02902 , year=

    Pith/arXiv arXiv

  59. [84]

    S ^2 R : Teaching LLM s to Self-verify and Self-correct via Reinforcement Learning

    Ma, Ruotian and Wang, Peisong and Liu, Cheng and Liu, Xingyan and Chen, Jiaqi and Zhang, Bang and Zhou, Xin and Du, Nan and Li, Jia. S ^2 R : Teaching LLM s to Self-verify and Self-correct via Reinforcement Learning. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025

    2025

  60. [85]

    arXiv preprint arXiv:2501.11425 , year=

    Agent-r: Training language model agents to reflect via iterative self-training , author=. arXiv preprint arXiv:2501.11425 , year=

    Pith/arXiv arXiv

  61. [86]

    Vicky and He, Conghui and Wu, Lijun

    Pan, Zhuoshi and Li, Yu and Lin, Honglin and Pei, Qizhi and Tang, Zinan and Wu, Wei and Ming, Chenlin and Zhao, H. Vicky and He, Conghui and Wu, Lijun. LEMMA : Learning from Errors for M athe M atical Advancement in LLM s. Findings of the Association for Computational Linguistics: ACL 2025. 2025

    2025

  62. [87]

    Understanding the Dark Side of LLM s' Intrinsic Self-Correction

    Zhang, Qingjie and Wang, Di and Qian, Haoting and Li, Yiming and Zhang, Tianwei and Huang, Minlie and Xu, Ke and Li, Hewu and Yan, Liu and Qiu, Han. Understanding the Dark Side of LLM s' Intrinsic Self-Correction. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025

    2025

  63. [88]

    arXiv preprint arXiv:2507.06261 , year=

    Gemini 2.5: Pushing the frontier with advanced reasoning, multimodality, long context, and next generation agentic capabilities , author=. arXiv preprint arXiv:2507.06261 , year=

    Pith/arXiv arXiv

  64. [89]

    arXiv preprint arXiv:2601.03267 , year=

    Openai gpt-5 system card , author=. arXiv preprint arXiv:2601.03267 , year=

    Pith/arXiv arXiv

  65. [90]

    arXiv preprint arXiv:2410.21276 , year=

    Gpt-4o system card , author=. arXiv preprint arXiv:2410.21276 , year=

    Pith/arXiv arXiv

  66. [91]

    arXiv preprint arXiv:2511.05271 , year=

    Deepeyesv2: Toward agentic multimodal model , author=. arXiv preprint arXiv:2511.05271 , year=

    Pith/arXiv arXiv

  67. [92]

    thinking with images

    Deepeyes: Incentivizing "thinking with images" via reinforcement learning , author=. arXiv preprint arXiv:2505.14362 , year=

    Pith/arXiv arXiv

  68. [93]

    arXiv preprint arXiv:2603.14117 , year=

    Improving visual reasoning with iterative evidence refinement , author=. arXiv preprint arXiv:2603.14117 , year=

    arXiv

  69. [94]

    arXiv preprint arXiv:2602.14482 , year=

    TikArt: Stabilizing Aperture-Guided Fine-Grained Visual Reasoning with Reinforcement Learning , author=. arXiv preprint arXiv:2602.14482 , year=

    arXiv

  70. [95]

    arXiv preprint arXiv:2507.07999 , year=

    Traceable evidence enhanced visual grounded reasoning: Evaluation and methodology , author=. arXiv preprint arXiv:2507.07999 , year=

    arXiv

  71. [96]

    arXiv preprint arXiv:2508.11630 , year=

    Thyme: Think beyond images , author=. arXiv preprint arXiv:2508.11630 , year=

    Pith/arXiv arXiv

  72. [97]

    arXiv preprint arXiv:2505.15966 , year=

    Pixel reasoner: Incentivizing pixel-space reasoning with curiosity-driven reinforcement learning , author=. arXiv preprint arXiv:2505.15966 , year=

    Pith/arXiv arXiv

  73. [98]

    arXiv preprint arXiv:2604.03307 , year=

    V-Reflection: Transforming MLLMs from Passive Observers to Active Interrogators , author=. arXiv preprint arXiv:2604.03307 , year=

    Pith/arXiv arXiv

  74. [99]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =

    Wu, Penghao and Xie, Saining , title =. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR) , month =. 2024 , pages =

    2024

  75. [100]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume=

    Divide, conquer and combine: A training-free framework for high-resolution image perception in multimodal large language models , author=. Proceedings of the AAAI Conference on Artificial Intelligence , volume=

  76. [101]

    MME-RealWorld: Could Your Multimodal LLM Challenge High-Resolution Real-World Scenarios that are Difficult for Humans? , url =

    Zhang, YiFan and Zhang, Huanyu and Tian, Haochen and Fu, Chaoyou and Zhang, Shuangqing and Wu, Junfei and Li, Feng and Wang, Kun and Wen, Qingsong and Zhang, Zhang and Wang, Liang and Jin, Rong , booktitle =. MME-RealWorld: Could Your Multimodal LLM Challenge High-Resolution Real-World Scenarios that are Difficult for Humans? , url =

  77. [102]

    Are We on the Right Way for Evaluating Large Vision-Language Models? , volume =

    Chen, Lin and Li, Jinsong and Dong, Xiaoyi and Zhang, Pan and Zang, Yuhang and Chen, Zehui and Duan, Haodong and Wang, Jiaqi and Qiao, Yu and Lin, Dahua and Zhao, Feng , booktitle =. Are We on the Right Way for Evaluating Large Vision-Language Models? , volume =

  78. [103]

    We-Math: Does Your Large Multimodal Model Achieve Human-like Mathematical Reasoning?

    Qiao, Runqi and Tan, Qiuna and Dong, Guanting and MinhuiWu, MinhuiWu and Sun, Chong and Song, Xiaoshuai and Wang, Jiapeng and GongQue, Zhuoma and Lei, Shanglin and Zhang, YiFan and Wei, Zhe and Zhang, Miaoxuan and Qiao, Runfeng and Zong, Xiao and Xu, Yida and Yang, Peiqing and Bao, Zhimin and Diao, Muxi and Li, Chen and Zhang, Honggang. We-Math: Does Your...

    2025

  79. [104]

    MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts , volume =

    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 =. MathVista: Evaluating Mathematical Reasoning of Foundation Models in Visual Contexts , volume =

  80. [105]

    MATHVERSE: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?

    Zhang, Renrui and Jiang, Dongzhi and Zhang, Yichi and Lin, Haokun and Guo, Ziyu and Qiu, Pengshuo and Zhou, Aojun and Lu, Pan and Chang, Kai-Wei and Qiao, Yu and Gao, Peng and Li, Hongsheng. MATHVERSE: Does Your Multi-modal LLM Truly See the Diagrams in Visual Math Problems?. Computer Vision -- ECCV 2024. 2025

    2024

Showing first 80 references.