Pith. sign in

REVIEW 3 major objections 4 minor 37 references

This paper claims that separating a model's verification plan from exact tool execution—via a Mathematical Tool Flow interface—makes LLMs revise wrong math answers more reliably than reflection or hand-written verification programs, with av

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 · deepseek-v4-flash

2026-08-03 04:47 UTC pith:7PPAEAXU

load-bearing objection The MTF interface is a genuinely new and clean idea, but the headline accuracy gains rest on a small, single-judge evaluation that needs an audit before I'd trust the numbers. the 3 major comments →

arxiv 2607.29549 v1 pith:7PPAEAXU submitted 2026-07-31 cs.AI

AMTFV: Agentic Mathematical Tool-Flow Verification for LLM Self-Correction

Rui Zou , Yutao Zhu , Mengqi Wei , Ji-Rong Wen This is my paper
classification cs.AI
keywords mathematical verificationself-correctionlarge language modelstool-augmented reasoningMathematical Tool Flowagentic verificationexact computationanswer revision
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.

AMTFV tries to establish that the main weakness in LLM mathematical self-correction is not a lack of tools but the premature coupling of verification planning with low-level code execution. It introduces Mathematical Tool Flow (MTF), an interrupt–execute–resume interface through which the model specifies what to compute as a mathematical object, then pauses while a toolbox agent executes exact computation and returns structured results. The paper argues this decoupling allows verification targets like global minimality, exhaustive enumeration, and symbolic equivalence to be checked explicitly, and reports that across five hard math datasets and seven model configurations, AMTFV outperforms representative baselines, with the largest gains on problems requiring more verification steps. A sympathetic reader would take from this that verification reliability in LLMs can be improved by making computational intent explicit and delegating exact execution, rather than by asking the model to reflect in natural language or to write ad hoc programs.

Core claim

The core claim is that separating mathematical verification modeling from concrete execution improves self-correction. In AMTFV, a verification agent first constructs a verification workflow and encodes the mathematical objects and computational intent as an MTF request—a context, an object to compute or verify, and the desired return operation. A mathematical toolbox agent then selects an appropriate backend tool, generates an executable call, dispatches it, and returns structured results. These results drive candidate-answer adjudication, answer revision, and verification-workflow revision. The authors report that AMTFV ranks highest in average accuracy under all seven base-model settings,

What carries the argument

The central mechanism is the MTF fragment τ = ⟨Γ, M, ρ⟩, where Γ is context (variables, domains, parameter assumptions), M is the mathematical object to compute, construct, or verify (a set, expression, equation system, recurrence, constraint system), and ρ is the return operation (cardinality, symbolic simplification, feasible solutions, global optimum, satisfiability). MTF uses an interrupt–execute–resume pattern: the LLM emits a computation request, pauses, the mathematical toolbox agent translates the request into an executable call using tools for symbolic computation, enumeration, exact rational arithmetic, and constraint solving, and the structured result is appended to the execution

Load-bearing premise

The comparison rests on one automatic answer-equivalence judge: answers that are neither exact-string nor float matches are judged by a single LLM evaluator, and the 170-problem benchmark is reported without confidence intervals or significance tests; if that judge or the answer-extraction step systematically favors AMTFV's structured, set-valued, or rounded output formats, the reported gains could be an artifact.

What would settle it

Re-score every final answer by exact string match against the gold answer (or have humans adjudicate all non-exact matches) while keeping the same initial responses, baselines, and iteration budgets; then compare AMTFV's accuracy gain against the strongest baseline. If the gain shrinks to near zero or reverses, the claimed advantage lives in the evaluator rather than in verification. Additionally, compute bootstrap confidence intervals over the 170 problems to see whether the 8.3-percentage-point gap is plausibly nonzero.

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

If this is right

  • If the decoupling claim holds, verification can target global conditions such as minimality, exhaustive enumeration, and symbolic equivalence, which natural-language reflection tends to miss and ad hoc code often checks only for local consistency.
  • Gains concentrated in medium- and high-verification-complexity samples suggest that the framework's value grows with the number of computational interactions required, not with simpler checks.
  • The higher W→C and lower C→W rates imply that a well-structured verification loop can improve accuracy without the overcorrection risk that often accompanies reflection-based rewriting.
  • Consistent ranking across seven model configurations from three model families suggests the benefit transfers across base models and reasoning modes, not just one strong model.

Where Pith is reading between the lines

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

  • A direct test of the paper's design claim would be to swap the toolbox backend for a different exact-computation engine (for example, a different symbolic library) while keeping MTF and the agents identical; if the gains persist, decoupling is the active ingredient, and if they vanish, the specific toolbox is doing more work than the interface.
  • The reported correlation between tool-call count and empirical difficulty (r=0.67) hints that MTF call counts could serve as a cheap, automatic difficulty label for hard math problems, although the paper itself only uses the count as a complexity signal.
  • Because the benchmark contains 170 problems and the reported gains are a few problems per setting, the 8.3-percentage-point figure is best read as an estimate to be confirmed on a larger sample or with human-verified answer equivalence before relying on its precise magnitude.

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 / 4 minor

Summary. The paper proposes AMTFV, an agentic framework for mathematical answer verification and self-correction. At its core is a Mathematical Tool Flow (MTF) interface that lets an LLM verification agent specify mathematical objects and computational intent in a structured request, which a separate mathematical toolbox agent compiles into exact tool calls (SymPy, Fraction, itertools, Z3). This is intended to decouple verification modeling from low-level implementation. The method is evaluated on five mathematical datasets (170 problems total) and seven DeepSeek/GPT/Gemini model configurations, comparing against natural-language reflection, checklist-based correction, self-consistency, and program-verification baselines (ProgCo, ProgCo-Py). The main claims are that AMTFV achieves the highest average accuracy in all evaluated settings, with gains over the strongest baseline of up to 8.3 percentage points, and that gains concentrate on samples of medium and high verification complexity, with correction-state analysis indicating the gains come from correcting errors rather than corrupting correct answers.

Significance. If the empirical claims hold, the MTF design is a meaningful contribution to test-time verification and self-correction for mathematical reasoning. The paper provides a clear architectural separation between verification modeling and exact computation, and the case studies (AIME24 minimality, AIME25 counting, AMO lattice enumeration, BRUMO25 exact double sum) concretely illustrate failure modes of natural-language reflection and ad hoc program verification that the proposed method addresses. The design avoids fitting constants to the benchmark, shares identical initial responses across methods, and the supplementary material states that raw outputs are included, which supports reproducibility. These are genuine strengths. The main limitation is that the comparative evaluation rests on a small, single-run protocol with a single LLM answer adjudicator, so the magnitude and sometimes the existence of the reported gains are not yet established to the standard expected for the paper's central claim.

major comments (3)
  1. [Supplementary A.5] The final-answer labels for all answers that are neither exact-string nor float matches are produced by a single DeepSeek-Flash evaluator at temperature 0. AMTFV is explicitly designed to return structured, set-valued, exact-rational, or multi-component outputs, whereas ProgCo and ProgCo-Py more often emit a single number. If this evaluator is systematically more lenient toward AMTFV's output format, the reported advantage could be an artifact of the evaluation rather than of verification quality. The differences are small in absolute terms: the headline +8.3pp on DeepSeek-Flash corresponds to roughly 14 of 170 problems, and several cells differ by only 2–6 problems. No human audit, inter-annotator agreement, or alternate evaluator is reported. Because the supplementary material states that raw outputs are included, a manual or independent re-judgment of the disputed cases is feasible an
  2. [Eq. (7), 'Verification Complexity Analysis'] The Low/Medium/High verification-complexity bins are defined by the number of MTF tool calls that AMTFV itself makes for the instance. The claim that AMTFV's gains are largest in the Medium and High bins is therefore partly circular: the bins are a function of the very method whose performance is being characterized. The reported correlation with empirical difficulty (Pearson r=0.67) mitigates this, but it does not remove the endogeneity, since tool-call count is still determined by AMTFV's own behavior. This analysis should either be relabeled as a descriptive property of AMTFV's execution, or the complexity measure should be defined independently of the method (e.g., by problem features or required proof/verification structure).
  3. [Experiments, 'Main Results' and 'Architectural Variant Analysis'] The evaluation uses 170 problems in total, with per-dataset n=30 or 50 for AMO, and reports a single deterministic run per method with no confidence intervals or significance tests. Many of the claimed improvements are small: on DeepSeek-Flash-Think AMTFV averages 89.4 vs. 88.2 for ProgCo-Py (a 1.2pp difference, i.e., about 2 problems), and in several GPT/Gemini settings the per-cell differences are 2–4 problems. Without uncertainty quantification, the claim that AMTFV 'outperforms representative baselines overall' is only a point estimate. The paired data structure (same initial responses, same problems) allows simple tests such as McNemar's test or bootstrap confidence intervals; these should be reported.
minor comments (4)
  1. [Supplementary Tables 4 and 5] The Avg columns report changes relative to Initial Score, but the Initial Score row is not shown in these tables. Add the Initial Score values so the reader can verify the claimed gains.
  2. [Main text, Fig. 6 and Eq. (7)] The verification-complexity bin definition appears only in the supplementary material. Since this definition is central to the complexity analysis, at least a one-sentence summary should appear in the main text or in the Figure 6 caption.
  3. [Supplementary Table 6] The AMO row says the maximum isosceles-free subset size is 5, while the final AMTFV answer is 6. This is correct (n=6 is the smallest number guaranteeing an isosceles triangle), but the table's phrasing 'maximum isosceles-free subset: 5' next to 'AMTFV: 6' is confusing; clarify the relationship.
  4. [General] The paper uses future-dated model names and API references (e.g., DeepSeek V4, GPT-5.4) that may not be familiar to readers; it would help to state explicitly that these are the model identifiers current as of the paper's writing, and to include access dates in the references as done for other URLs.

Circularity Check

0 steps flagged

No significant circularity: the headline accuracy comparison is an external empirical benchmark, no parameters are fitted, and the self-referential complexity stratification does not reduce the central claims.

full rationale

The paper's central claim is empirical: AMTFV is compared with published baselines (reflex, Self-Refine, Reflexion, TICK, CoT-Tool, ProgCo) on fixed initial responses from external base models; final-answer accuracy is computed by a stated three-stage equivalence protocol. No constant or parameter is fitted to the benchmark results, so nothing is 'predicted' from its own fitted inputs. The method components (MTF, toolbox agent) are described procedurally, and the worked examples (e.g., symbolic-expression check, double-sum case) execute exact computations that are independent of the final accuracy table. The references contain no self-citations that carry a uniqueness or ansatz justification; all load-bearing citations (ProgCo, SymPy, datasets) are external prior work. The only potentially self-referential element is the verification-complexity binning: Low/Medium/High bins are defined by AMTFV's own MTF call count (Eq. 7), so the statement that AMTFV gains most in medium/high complexity is a characterization of its own process signal rather than an independently measured difficulty label. However, this is an interpretation/construct-validity issue, not an equation-level circularity: gain is not equal to call count by construction, and the main comparison to external baselines does not depend on these bins. The evaluation-protocol concerns (single DeepSeek-Flash judge, small sample, no significance tests) are threats to validity, not circularity, and are outside this pass.

Axiom & Free-Parameter Ledger

2 free parameters · 4 axioms · 1 invented entities

No numerical parameters are fitted to the benchmark results; the main free choices are experimental budgets and bin thresholds. The central comparison rests on fairness of shared initialization and equivalence evaluation, plus the assumption that exact-tool verification is more reliable than language reflection.

free parameters (2)
  • max_turn = 3
    Outer correction/verification rounds are capped at 3 for all iterative methods; this hand-set budget directly affects the reported accuracies.
  • verification_complexity_bin_thresholds = Low ≤1, Medium 2–3, High ≥4 MTF calls
    Hand-set cutoffs in Eq. (7) define the complexity analysis and drive the claim of larger gains on medium/high complexity samples.
axioms (4)
  • domain assumption Using exact mathematical tools (SymPy, Fraction, Z3) for verification produces more reliable judgments than natural-language reflection or ad hoc code.
    This is the design motivation in the Introduction and Method; if false, AMTFV's mechanism adds no benefit.
  • domain assumption The LLM can faithfully express verification intents as MTF requests (context–object–return) and the toolbox agent can compile them into correct tool calls.
    Invoked throughout the Method section; if the model hallucinates objects or miscompiles calls, tool results cannot support adjudication.
  • domain assumption Shared initial responses/candidates and the three-stage equivalence evaluator produce unbiased accuracy measurements across methods.
    Supplementary A.2 and A.5; the central comparison depends on this fairness premise.
  • domain assumption The five datasets and their gold answers are correct and representative of mathematical reasoning difficulty.
    Experiments setup; the small 170-problem sample makes this assumption load-bearing.
invented entities (1)
  • Mathematical Tool Flow (MTF) interrupt–execute–resume interface no independent evidence
    purpose: Intermediate representation separating mathematical verification modeling from concrete tool execution; the central new abstraction of the method.
    MTF is a design artifact introduced by the paper. It has no external falsifiable handle beyond the paper's own benchmark results; its value rests entirely on the empirical comparison.

pith-pipeline@v1.3.0-daily-deepseek · 19487 in / 10785 out tokens · 105658 ms · 2026-08-03T04:47:36.929851+00:00 · methodology

0 comments
read the original abstract

Large language models have demonstrated strong mathematical problem-solving capabilities, yet reliably verifying their candidate answers remains challenging. Existing representative methods mainly revise outputs through natural-language reflection or assist verification by directly generating verification programs; the former may not reliably support exact computation, whereas the latter prematurely couples mathematical modeling with low-level implementation. We propose AMTFV (Agentic Mathematical Tool-Flow Verification). By introducing Mathematical Tool Flow (MTF) as an interrupt--execute--resume interface, AMTFV decouples verification modeling from concrete execution and supports exact computation through a mathematical toolbox. Specifically, the verification agent first constructs a verification workflow, encodes the mathematical objects and computational intent requiring reliable execution in an MTF request, and sends it to the mathematical toolbox agent. The latter parses the request, generates executable calls, and dispatches them to the backend for exact computation. Tool outputs then support candidate-answer adjudication, answer revision, and verification-workflow revision. We evaluate AMTFV on five challenging mathematical reasoning datasets with seven model configurations from DeepSeek, GPT, and Gemini. Experimental results show that AMTFV outperforms the representative baselines evaluated in this study overall; under an individual model configuration, it improves average accuracy over the strongest baseline by up to 8.3 percentage points, with larger gains on samples of medium and high verification complexity.

Figures

Figures reproduced from arXiv: 2607.29549 by Ji-Rong Wen, Mengqi Wei, Rui Zou, Yutao Zhu.

Figure 1
Figure 1. Figure 1: Comparison of natural-language reflection, direct [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the AMTFV framework. cal computation from natural-language reasoning; and ToRA integrates natural-language reasoning with tool calls for mathematical problem solving (Gao et al. 2023; Chen et al. 2023; Gou et al. 2024). Tool-augmented mathematical agents such as AgentMath and R1-Code-Interpreter likewise use code interpreters or tool calls for complex mathematical tasks (Luo et al. 2026; Liu et… view at source ↗
Figure 3
Figure 3. Figure 3: Average accuracy of verification-enhanced methods under each base-model setting. The horizontal axis shows the [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Overall average-accuracy trajectories under dif [PITH_FULL_IMAGE:figures/full_fig_p007_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Row-normalized correction-state transition matri [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Verification–correction performance under different iteration budgets. The three panels correspond to ProgCo, ProgCo [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Relationship between empirical difficulty and verifi [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗

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

37 extracted references · 1 canonical work pages

  1. [1]

    2025 , address =

    Song, Xiaoshuai and Wu, Yanan and Wang, Weixun and Liu, Jiaheng and Su, Wenbo and Zheng, Bo , booktitle =. 2025 , address =

    2025

  2. [2]

    Advances in Neural Information Processing Systems , volume =

    Self-Refine: Iterative Refinement with Self-Feedback , author =. Advances in Neural Information Processing Systems , volume =. 2023 , url =

    2023

  3. [3]

    Advances in Neural Information Processing Systems , volume =

    Reflexion: Language Agents with Verbal Reinforcement Learning , author =. Advances in Neural Information Processing Systems , volume =. 2023 , url =

    2023

  4. [4]

    NeurIPS 2024 Workshop on Language Gamification , year =

    Cook, Jonathan and Rockt. NeurIPS 2024 Workshop on Language Gamification , year =

    2024

  5. [5]

    International Conference on Learning Representations , year =

    Self-Consistency Improves Chain of Thought Reasoning in Language Models , author =. International Conference on Learning Representations , year =

  6. [6]

    2025 , publisher =

    Abedi Firouzjaei, Hassan , title =. 2025 , publisher =. doi:10.5281/zenodo.15331821 , url =

    work page doi:10.5281/zenodo.15331821 2025

  7. [7]

    DeepSeek V4 Preview Release , year =

  8. [8]

    Gemini 2.5 Flash-Lite , year =

  9. [9]

    Gemini API OpenAI Compatibility , year =

  10. [10]

    Second Conference on Language Modeling , year =

    A Sober Look at Progress in Language Model Reasoning: Pitfalls and Paths to Reproducibility , author =. Second Conference on Language Modeling , year =

  11. [11]

    2025 , eprint =

    ScaleDiff: Scaling Difficult Problems for Advanced Mathematical Reasoning , author =. 2025 , eprint =

    2025

  12. [12]

    2026 , address =

    Liu, Junlin and An, Shengnan and Zhou, Shuang and Ma, Dan and Lin, Yehao and Lv, Xinxuan and Wang, Xuanlin and Li, Xiaoyu and Wang, Ziwen and Cao, Xuezhi and Cai, Xunliang , booktitle =. 2026 , address =

    2026

  13. [13]

    Advances in Neural Information Processing Systems , volume =

    Chain-of-Thought Prompting Elicits Reasoning in Large Language Models , author =. Advances in Neural Information Processing Systems , volume =. 2022 , url =

    2022

  14. [14]

    2021 , eprint =

    Training Verifiers to Solve Math Word Problems , author =. 2021 , eprint =. doi:10.48550/arXiv.2110.14168 , url =

    Pith/arXiv arXiv doi:10.48550/arxiv.2110.14168 2021

  15. [15]

    Transactions of the Association for Computational Linguistics , volume =

    Automatically Correcting Large Language Models: Surveying the Landscape of Diverse Automated Correction Strategies , author =. Transactions of the Association for Computational Linguistics , volume =. 2024 , address =

    2024

  16. [16]

    When Can

    Kamoi, Ryo and Zhang, Yusen and Zhang, Nan and Han, Jiawei and Zhang, Rui , journal =. When Can. 2024 , address =

    2024

  17. [17]

    The Twelfth International Conference on Learning Representations , year =

    Large Language Models Cannot Self-Correct Reasoning Yet , author =. The Twelfth International Conference on Learning Representations , year =

  18. [18]

    2024 , address =

    Tyen, Gladys and Mansoor, Hassan and Carbune, Victor and Chen, Peter and Mak, Tony , booktitle =. 2024 , address =

    2024

  19. [19]

    2023 , url =

    Yao, Shunyu and Zhao, Jeffrey and Yu, Dian and Du, Nan and Shafran, Izhak and Narasimhan, Karthik and Cao, Yuan , booktitle =. 2023 , url =

    2023

  20. [20]

    Advances in Neural Information Processing Systems , volume =

    Toolformer: Language Models Can Teach Themselves to Use Tools , author =. Advances in Neural Information Processing Systems , volume =. 2023 , url =

    2023

  21. [21]

    Making Language Models Better Tool Learners with Execution Feedback , author =. Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , month = jun, pages =. 2024 , address =

    2024

  22. [22]

    2023 , volume =

    Gao, Luyu and Madaan, Aman and Zhou, Shuyan and Alon, Uri and Liu, Pengfei and Yang, Yiming and Callan, Jamie and Neubig, Graham , booktitle =. 2023 , volume =

    2023

  23. [23]

    Transactions on Machine Learning Research , year =

    Program of Thoughts Prompting: Disentangling Computation from Reasoning for Numerical Reasoning Tasks , author =. Transactions on Machine Learning Research , year =

  24. [24]

    2024 , url =

    Gou, Zhibin and Shao, Zhihong and Gong, Yeyun and Shen, Yelong and Yang, Yujiu and Huang, Minlie and Duan, Nan and Chen, Weizhu , booktitle =. 2024 , url =

    2024

  25. [25]

    and Paprocki, Mateusz and

    Meurer, Aaron and Smith, Christopher P. and Paprocki, Mateusz and. PeerJ Computer Science , volume =. 2017 , url =

    2017

  26. [26]

    2409.12122 , archivePrefix =

    Yang, An and Zhang, Beichen and Hui, Binyuan and Gao, Bofei and Yu, Bowen and Li, Chengpeng and Liu, Dayiheng and Tu, Jianhong and Zhou, Jingren and Lin, Junyang and Lu, Keming and Xue, Mingfeng and Lin, Runji and Liu, Tianyu and Ren, Xingzhang and Zhang, Zhenru , year =. 2409.12122 , archivePrefix =

    Pith/arXiv arXiv

  27. [27]

    2025 , doi =

    Guo, Daya and others , journal =. 2025 , doi =

    2025

  28. [28]

    Proceedings of the AAAI Conference on Artificial Intelligence , volume =

    MathSmith: Towards Extremely Hard Mathematical Reasoning by Forging Synthetic Problems with a Reinforced Policy , author =. Proceedings of the AAAI Conference on Artificial Intelligence , volume =. 2026 , doi =

    2026

  29. [29]

    2025 , address =

    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 , booktitle =. 2025 , address =

    2025

  30. [30]

    Boosting

    Zhao, Xutong and Xu, Tengyu and Wang, Xuewei and Chen, Zhengxing and Jin, Di and Tan, Liang and Lin, Yen-Ting and Yu, Zishun and Zhao, Zhuokai and He, Yun and Wang, Sinong and Fang, Han and Chandar, Sarath and Zhu, Chen , booktitle =. Boosting. 2025 , url =

    2025

  31. [31]

    The Fourteenth International Conference on Learning Representations , year =

    AgentMath: Empowering Mathematical Reasoning for Large Language Models via Tool-Augmented Agent , author =. The Fourteenth International Conference on Learning Representations , year =

  32. [32]

    R1-Code-Interpreter:

    Liu, Yueying and Chen, Yongchao and Zhou, Junwei and Hao, Yilun and Wang, Jingquan and Zhang, Yang and Li, Na and Fan, Chuchu , booktitle =. R1-Code-Interpreter:. 2026 , url =

    2026

  33. [33]

    2025 , eprint =

    Large Language Models and Mathematical Reasoning Failures , author =. 2025 , eprint =

    2025

  34. [34]

    and Burger, Doug and Wang, Chi , booktitle =

    Wu, Qingyun and Bansal, Gagan and Zhang, Jieyu and Wu, Yiran and Li, Beibin and Zhu, Erkang and Jiang, Li and Zhang, Xiaoyun and Zhang, Shaokun and Liu, Jiale and Awadallah, Ahmed Hassan and White, Ryen W. and Burger, Doug and Wang, Chi , booktitle =. 2024 , url =

    2024

  35. [35]

    The Twelfth International Conference on Learning Representations , year =

    Hong, Sirui and Zhuge, Mingchen and Chen, Jonathan and Zheng, Xiawu and Cheng, Yuheng and Wang, Jinlin and Zhang, Ceyao and Wang, Zili and Yau, Steven Ka Shing and Lin, Zijuan and Zhou, Liyang and Ran, Chenyu and Xiao, Lingfeng and Wu, Chenglin and Schmidhuber, J. The Twelfth International Conference on Learning Representations , year =

  36. [36]

    Chen, Weize and Su, Yusheng and Zuo, Jingwei and Yang, Cheng and Yuan, Chenfei and Chan, Chi-Min and Yu, Heyang and Lu, Yaxi and Hung, Yi-Hsin and Qian, Chen and Qin, Yujia and Cong, Xin and Xie, Ruobing and Liu, Zhiyuan and Sun, Maosong and Zhou, Jie , booktitle =

  37. [37]

    2026 , howpublished =

    2026