Pith. sign in

REVIEW 3 major objections 4 minor 1 cited by

A Survey of Deep Learning for Geometry Problem Solving

T0 review · 3 major / 4 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This survey organizes deep learning for geometry problem solving into a three-tier task taxonomy and a three-axis method taxonomy.

desk verdict A genuinely useful mapping of the geometry-solving literature, but the paper's own numbers undermine its comprehensiveness claim; the taxonomy and tables carry it. read the letter →

arxiv 2507.11936 v6 pith:5NVP47J4 submitted 2025-07-16 cs.CL cs.AIcs.CVcs.LG

Jianzhe Ma , Wenxuan Wang , Qin Jin This is my paper
classification cs.CLcs.AIcs.CVcs.LG
keywords geometryproblemsolvingdeeplearningmultimodallargelanguagemodelstheoremprovinggeometricnumericalcalculationneural-symbolicreasoningmathematicalbenchmarkssurvey
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

Geometry problem solving (GPS) is a multimodal mathematical task that requires reading diagrams, parsing formal statements, and performing proof or calculation. This survey aims to be a comprehensive reference for deep learning work on GPS, collecting more than 310 papers and organizing them into fundamental, core, and composite tasks, with methods classified by architecture, training stage, and inference stage. It also analyzes evaluation metrics, benchmarks representative models on Geometry3K, GeoQA, and MathVista-GPS, and identifies four trends: reinforcement learning helps, neural-symbolic methods stay competitive on symbolic benchmarks, high-quality large-scale data matters, and test-time scaling is promising. If the survey's organization is right, a newcomer can use it to locate the datasets, methods, and open problems of the field in one place.

What carries the argument

The carrying device is the survey's three-level task taxonomy (fundamental, core, composite) combined with a method taxonomy over architecture, training stage, and inference stage. This taxonomy is what lets the survey compare otherwise disparate systems: it assigns each dataset to a task level, places every reviewed method on the architecture/training/inference axes, and turns the field's gaps (few theorem-proving datasets, little solid and analytic geometry, mostly multiple-choice evaluation, scarce reasoning-process annotations) into visible vacancies rather than scattered observations.

What would settle it

Searching with related terms such as 'math word problem', 'diagram parsing', or 'geometric reasoning' and counting GPS papers absent from this survey would test the claim that about 310 papers suffice; if dozens of relevant papers are missing, the comprehensiveness claim weakens. A second test is to inspect the reference list for a known relevant paper and see whether the survey discusses it.

Watch

Extended reading notes

Core claim

The central claim is that the field has a stable structure that can be described by a task taxonomy and a method taxonomy. Tasks fall into fundamental abilities (diagram understanding, semantic parsing, relation extraction, knowledge prediction), core tasks (theorem proving and numerical calculation), and composite tasks (mathematical reasoning); methods fall into architectures (encoder-decoder with text encoder, diagram encoder, fusion module, decoder, and knowledge module) and strategies at training time (pre-training, supervised fine-tuning, reinforcement learning) and inference time (test-time scaling, knowledge-augmented inference). Within this structure, the paper reports that state-of-the-art results come from large high-quality data and RL, that neural-symbolic solvers still lead the symbolic Geometry3K benchmark, and that diagram perception is the current bottleneck, with adding diagrams sometimes lowering accuracy. It concludes that data, evaluation, and perception gaps define the field's open agenda.

Load-bearing premise

The paper's comprehensiveness rests on a literature search that used one keyword and one round of snowballing, and the paper itself says this may not fully represent the field.

Editorial extensions

If this is right

  • Researchers can use the taxonomy to choose an underserved niche, since the paper identifies theorem proving, solid and analytic geometry, non-English datasets, and reasoning-process annotations as thin.
  • Because most benchmarks are multiple-choice, reported accuracy may overstate ability; the paper's call for option-free and harder evaluation implies current leaderboards need re-reading.
  • If the four performance trends hold, future GPS systems will likely pair large high-quality training data and reinforcement learning with neural-symbolic components rather than rely on a single architecture.
  • As foundation models improve and existing benchmarks saturate, progress will depend on process-based metrics and efficiency measures, not just answer accuracy.

Reading between the lines

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

  • Editorial inference: a literature search built on one keyword and one snowballing round likely undercounts work published under names like 'math word problem' or 'diagram parsing', so the 310-paper corpus is probably a lower bound and the taxonomy may need periodic re-screening.
  • Editorial inference: the fundamental/core/composite structure is general enough that it could be transferred to other diagram-based reasoning domains, such as physics or chart-based question answering.
  • Editorial inference: because the performance table mixes benchmarks with different question formats, answer grading, and image types, a controlled benchmark that fixes these variables would directly test the reported trends.
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.

Referee Report

3 major / 4 minor

Summary. This paper surveys deep-learning approaches to geometry problem solving (GPS). It proposes a three-level task taxonomy (fundamental, core, and composite tasks), reviews the associated datasets, architectures, training-stage and inference-stage methods, and discusses automatic and manual evaluation metrics. It also compares a small set of representative models on three benchmarks and closes with challenges and future directions. The manuscript provides dense dataset and architecture tables in the appendix and points to a community-maintained GitHub list of relevant papers.

Significance. The survey addresses a genuine gap: broader mathematical-reasoning surveys treat geometry only as a subtopic, and prior GPS-specific surveys are narrower in scope. The proposed taxonomy is reasonable, and the appendix tables (Tables 2–5) offer a practical consolidation of datasets and system components that researchers entering the area will likely find useful. The paper is a survey rather than a derivation, so circularity is not at issue; its value rests on the completeness and accuracy of its coverage. The main risk is that the advertised corpus size and the search protocol are not internally consistent, which directly weakens the central claim of being a comprehensive reference. With the coverage claims repaired, the survey would be a solid entry point to the field.

major comments (3)
  1. [§1, Figure 5 (Appendix)] The introduction claims that the survey collected "more than 310 academic papers" on deep learning for GPS, but the paper's own Figure 5 reports annual counts of 1, 5, 3, 6, 15, 25, 112, and 83 for 2018 through April 2025, which sum to only 250. The figure does not cover the 2026-dated references that appear elsewhere in the bibliography, and no alternative definition of the corpus is provided. Because the abstract and introduction stake the paper's value on being "a comprehensive and practical reference," this internal inconsistency in the corpus count is load-bearing and must be resolved, either by updating the figure and the count or by qualifying the comprehensiveness claim.
  2. [§1, Limitations] The search protocol described in Section 1 (a single Google Scholar keyword "geometry" plus one round of forward and backward snowballing) is not sufficient to substantiate the claim of comprehensive coverage, and the manuscript's own Limitations section concedes that the survey "may not fully represent the development process of the entire field." Please provide a more detailed and transparent protocol—database and query strings, snowballing directions, inclusion/exclusion criteria, screening counts, and the relationship between the Figure 5 subset and the claimed >310 papers—and move the coverage caveat into the main text where it qualifies the central claim.
  3. [§5.1, Table 1] Table 1 reports state-of-the-art and second-best results on Geometry3K, GeoQA, and MathVista, but the numbers are taken from the original papers without specifying the evaluation protocol used in each case (e.g., answer extraction, multiple-choice versus free-form output, model version, or prompt format). The four trends stated in §5.1, such as "neural-symbolic methods demonstrate superior performance on symbolic-oriented tasks," rely on cross-model comparisons between numbers that may not be directly comparable. Please either state the conditions under which each score was obtained or limit the trends to subsets with a verified common evaluation setting.
minor comments (4)
  1. [§3.1.2, §3.3.1, References] The citation "Sinha et al." appears without a year in §3.1.2 (e.g., "Trinh et al., 2024; Sinha et al.; Chervonyi et al., 2025"), and the corresponding reference entry also lacks a year and venue. The citation "Tey" in §3.2.2 and §3.3.1 similarly points to a bibliography entry with no year and no publication venue. These entries should be completed or removed.
  2. [Table 3] The composite-datasets table contains a row labeled "MATH()(2024)" with an apparently lost version tag, and several cells are left blank rather than marked with "—" or "N/A", which makes the table harder to parse and less reliable as a reference.
  3. [Figure 5] The figure caption says "data for 2025 is up to April," but the x-axis label reads "(Jan. Apr.)" without the year, and the figure should either be extended to include the 2026 references cited in the text or explicitly state that the count is only through April 2025.
  4. [Throughout] There are numerous mechanical formatting artifacts, including "LLaV A" instead of "LLaVA", "MA VIS" instead of "MAVIS", "DEBRUP DAS" in the reference list, and the semicolon after "Tey". A careful copyedit would improve the manuscript's usability as a reference.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: the survey organizes external literature and makes no derivation or prediction claims that reduce to its own inputs.

full rationale

This is a survey paper, not a derivation. Its central claims are taxonomic and descriptive: it summarizes tasks, methods, evaluation metrics, datasets, and performance results from the external literature. There are no fitted parameters, no predicted quantities that reduce to fitted values, and no formal derivation chain whose conclusion is built into its premises. The authors do not appear to rely on their own prior work as load-bearing evidence; the reference list contains no obvious self-citations by Ma, Wang, or Jin, and the taxonomy in Sections 2–3 is presented as an organizational framework rather than as a mathematically forced result. The Skeptic's concern about the comprehensiveness claim is a legitimate correctness and transparency issue: Section 1 reports 'more than 310 academic papers,' while Appendix Figure 5 shows yearly counts summing to 250 for 2018 through April 2025, and the Limitations section concedes the survey 'may not fully represent the development process of the entire field.' However, a discrepancy in the reported corpus size is not circularity; it does not show that any conclusion is equivalent to its inputs by construction, and it does not involve fitting, self-citation, or renaming. The paper is self-contained as an organizing review: its conclusions about field trends are supported by cited external results and benchmarks, not by the survey's own definitions. The Limitations statement is explicitly acknowledged and does not hide a circular step. Accordingly, no specific circular step can be quoted, and the honest finding is no significant circularity.

Assumptions & free parameters 0 free parameters · 2 assumptions · 0 invented entities

The central claim is the comprehensiveness of the survey. It rests on two domain assumptions: that the collected papers are representative, and that the proposed taxonomy is a useful organization. There are no free parameters or invented entities because this is a review.

assumptions (2)
  • domain assumption The collected 310 papers are representative of the deep learning for GPS literature.
    The survey's comprehensiveness claim depends on this; the search used one round of snowballing and a single Google Scholar keyword, which may not be exhaustive.
  • domain assumption The proposed taxonomy (fundamental, core, composite tasks; architecture, training, inference) is a useful and accurate way to organize the field.
    The survey is structured around this taxonomy, but it is a scholarly choice rather than a proven result.

how reviews work

0 comments
Cite this review

Pith. "Pith review of A Survey of Deep Learning for Geometry Problem Solving." pith.science (2026). https://pith.science/paper/5NVP47J4

@misc{pith2026250711936,
  author       = {Pith},
  title        = {Pith review of: A Survey of Deep Learning for Geometry Problem Solving},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5NVP47J4}},
  note         = {Machine review of arXiv:2507.11936}
}
read the original abstract

Geometry problem solving, a crucial aspect of mathematical reasoning, is vital across various domains, including education, the assessment of AI's mathematical abilities, and multimodal capability evaluation. The recent surge in deep learning technologies, particularly the emergence of multimodal large language models, has significantly accelerated research in this area. This paper presents a survey of the applications of deep learning in geometry problem solving, including (i) a comprehensive summary of the relevant tasks in geometry problem solving; (ii) a thorough review of related deep learning methods; (iii) a detailed analysis of evaluation metrics and methods; and (iv) a critical discussion of state-of-the-art performance, existing challenges, and promising future directions. Our objective is to offer a comprehensive and practical reference of deep learning for geometry problem solving, thereby fostering further advancements in this field. We maintain a list of relevant papers: https://github.com/majianz/dl4gps.

Figures

Figures reproduced from arXiv: 2507.11936 by the authors.

Figure 1
Figure 1. Overview of the survey’s structure. Early deep learning applications primarily fo￾cused on combining deep learning modules with symbolic systems to solve geometry problems us￾ing formal languages, with the deep learning mod￾ule acting as part of the overall system (Liu et al., 2019a; Lu et al., 2021). Neural-symbolic meth￾ods have gradually become the mainstream ap￾proach in deep learning for GPS. In recent years, w… view at source ↗
Figure 2
Figure 2. Taxonomy of tasks and datasets for geometry problem solving. [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. An example of geometry theorem proving and [PITH_FULL_IMAGE:figures/full_fig_p003_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Taxonomy of deep learning methods for geometry problem solving. [PITH_FULL_IMAGE:figures/full_fig_p004_4.png]
Figure 5
Figure 5. Figure 5: Papers on deep learning for geometry problem [PITH_FULL_IMAGE:figures/full_fig_p027_5.png]

Discussion (0). Sign in to comment.

Forward citations

Cited by 1 Pith paper

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. Draw2Think: Harnessing Geometry Reasoning through Constraint Engine Interaction

    cs.CV 2026-05 unverdicted novelty 7.0 of 10

    Draw2Think recasts geometric reasoning as agentic interaction with a constraint engine, achieving 95.9% predicate-level construction fidelity and up to 16.4% accuracy gains on solid geometry tasks.

Reference graph

Works this paper leans on

29 extracted references · 17 canonical work pages · cited by 1 Pith paper

  1. [4]

    InProceedings of the International Conference on Internet Multimedia Computing and Service, pages 201–206

    Plane geometry diagram retrieval by using hierarchical searching strategy. InProceedings of the International Conference on Internet Multimedia Computing and Service, pages 201–206. Wenbin Gan, Xinguo Yu, Chao Sun, Bin He, and Ming- shu Wang. 2017. Understanding plane geometry problems by integrating relations extracted from text and diagram. InImage and ...

    arXiv 2017

  2. [5]

    InThe Twelfth Inter- national Conference on Learning Representations

    Tora: A tool-integrated reasoning agent for mathematical problem solving. InThe Twelfth Inter- national Conference on Learning Representations. Aaron Grattafiori, Abhimanyu Dubey, Abhinav Jauhri, Abhinav Pandey, Abhishek Kadian, Ahmad Al- Dahle, Aiesha Letman, Akhil Mathur, Alan Schelten, Alex Vaughan, and 1 others. 2024. The llama 3 herd of models.arXiv ...

    arXiv 2024

  3. [8]

    InProceedings of the 21st International Conference on Intelligent User Interfaces, pages 419–430

    Analyticalink: An interactive learning environ- ment for math word problem solving. InProceedings of the 21st International Conference on Intelligent User Interfaces, pages 419–430. Deepak Kapur. 1986. Using gröbner bases to reason about geometry problems.Journal of Symbolic Com- putation, 2(4):399–408. Mehran Kazemi, Hamidreza Alvari, Ankit Anand, Jialin...

    work page 1986

  4. [9]

    MultiLingPoT: Enhancing Mathematical Reasoning with Multilingual Program Fine-tuning

    Automatically building diagrams for olympiad geometry problems. InCADE, pages 577–588. Sergey Kurbatov, Igor Fominykh, and Aleksandr V orobyev. 2021. Cognitive patterns for semantic presentation of natural-language descriptions of well- formalizable problems. InRussian Conference on Artificial Intelligence, pages 317–330. Springer. Sergey S Kurbatov and I...

    work page Pith review arXiv 2021

  5. [11]

    InProceedings of the 15th International Conference on Education Technology and Computers, pages 308– 314

    Estimating answer strategies using online handwritten data: A study using geometry problems. InProceedings of the 15th International Conference on Education Technology and Computers, pages 308– 314. Chibuike Onuoha, Yusuke Haga, Dilshad Ferdousi, and Truong Cong Thang. 2025. Multimodal large lan- guage models for high school mathematical reason- ing: Impa...

    work page 2025

  6. [13]

    In2016 23rd International Conference on Pattern Recognition (ICPR), pages 1125–1130

    Improving pgf retrieval effectiveness with ac- tive learning. In2016 23rd International Conference on Pattern Recognition (ICPR), pages 1125–1130. IEEE. Gollam Rabby, Farhana Keya, Parvez Zamil, and Sören Auer. 2024. Mc-nest–enhancing mathematical rea- soning in large language models with a monte carlo nash equilibrium self-refine tree.arXiv preprint arXi...

    arXiv 2024

  7. [14]

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu

    Direct preference optimization: Your lan- guage model is secretly a reward model.Advances in Neural Information Processing Systems, 36:53728– 53741. Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. 2020. Exploring the lim- its of transfer learning with a unified text-to-text trans...

    arXiv 2020

  8. [15]

    Beyond Captioning: Task-Specific Prompting for Improved VLM Performance in Mathematical Reasoning

    Math-llava: Bootstrapping mathematical rea- soning for multimodal large language models. In Findings of the Association for Computational Lin- guistics: EMNLP 2024, pages 4663–4680. Ayush Singh, Mansi Gupta, Shivank Garg, Abhinav Ku- mar, and Vansh Agrawal. 2024. Beyond captioning: Task-specific prompting for improved vlm perfor- mance in mathematical rea...

    work page Pith review arXiv 2024

Show all 29 references
  1. [17]

    InProceedings of the 2024 Confer- ence on Empirical Methods in Natural Language Processing, pages 16618–16635

    Explicit memory learning with expectation maximization. InProceedings of the 2024 Confer- ence on Empirical Methods in Natural Language Processing, pages 16618–16635. Alex Young, Bei Chen, Chao Li, Chengen Huang, Ge Zhang, Guanwei Zhang, Guoyin Wang, Heng Li, Jiangcheng Zhu, J...

    2024 arXiv

  2. [22]

    Chengke Zou, Xingang Guo, Rui Yang, Junyu Zhang, Bin Hu, and Huan Zhang

    An example-based approach to 3d man-made object reconstruction from line drawings.Pattern Recognition, 60:543–553. Chengke Zou, Xingang Guo, Rui Yang, Junyu Zhang, Bin Hu, and Huan Zhang. 2025. Dynamath: A dy- namic visual benchmark for evaluating mathematical reasoning robust...

    2017

  3. [24]

    builds a pre-trained dataset containing more than 120,000 plane geometry images and tunes the visual encoder-decoder architecture using the mask auto-encoding scheme to obtain a visual encoder that fully understands geometric diagrams. Addi- tionally, some GPS work uses relate...

    2021

  4. [25]

    C Encoder-Decoder Architecture for Geometry Problem Solving In this section, we further elaborate on the deep learning components of the encoder-decoder archi- tecture used for GPS

    is a 173-problem geometric autoformaliza- tion dataset designed to test whether AI can under- stand mathematical problems and solutions written by humans and convert them into formal theorems and proofs. C Encoder-Decoder Architecture for Geometry Problem Solving In this secti...

    2022

  5. [26]

    encodes diagrams through VQV AE (Van Den Oord et al., 2017). Some other studies use a CNN-Transformer hy- brid architecture to integrate the functions of a text encoder and a diagram encoder into a multimodal encoder (Li et al., 2024g; Lin et al., 2024). C.3 Multimodal Fusion ...

    2019

  6. [27]

    Other studies employ pre-trained language models as decoders

    or GRU (Tsai et al., 2021; Jian et al., 2023b; Zhang et al., 2023a; Li et al., 2024g; Zhang et al., 2024b,d) as decoders in deep learning systems, which may also integrate attention mechanisms. Other studies employ pre-trained language models as decoders. For example, Liang et...

    2023

  7. [28]

    In addi- tion, Zhang et al

    as the decoder (Shengyuan and Xiuqin, 2024; Zhuang et al., 2025; Zhang et al., 2025a). In addi- tion, Zhang et al. (2025d) use MAmmoTH2 (Yue et al., 2024b), Zhang et al. (2025h) choose Yi- 1.5 (Young et al., 2024), and Cho et al. (2025) use Llama 3 (Grattafiori et al., 2024). ...

    2025

  8. [29]

    Answer Verifier.Ensuring the correctness of the solution logic is one of the key steps in solv- ing geometry problems

    to guide the training of theorem predictors. Answer Verifier.Ensuring the correctness of the solution logic is one of the key steps in solv- ing geometry problems. In addition to the tradi- tional rule-based verification method (Zhang et al., 2024d), Pan et al. (2025) introduc...

    2025

  9. [859]

    Shuai Peng, Di Fu, Liangcai Gao, Xiuqin Zhong, Hong- guang Fu, and Zhi Tang

    IEEE. Shuai Peng, Di Fu, Liangcai Gao, Xiuqin Zhong, Hong- guang Fu, and Zhi Tang. 2024. Multimath: Bridging visual and mathematical reasoning for large language models.arXiv preprint arXiv:2409.00147. Shuai Peng, Di Fu, Yijun Liang, Liangcai Gao, and Zhi Tang. 2023. Geodrl: A...

    2024 arXiv

  10. [1643]

    Ming-Liang Zhang, Fei Yin, and Cheng-Lin Liu

    International Joint Conferences on Artificial Intelligence Organization. Ming-Liang Zhang, Fei Yin, and Cheng-Lin Liu. 2023a. A multi-modal neural geometric solver with textual clauses parsed from diagram. InProceedings of the Thirty-Second International Joint Conference on Ar...

    arXiv

  11. [2004]

    1https://www.geogebra.org This task is also related to GPS

    and constrained numerical optimization (Ye et al., 2020). 1https://www.geogebra.org This task is also related to GPS. GeoX (Xia et al.,

    2020

  12. [2015]

    In2015 International Confer- ence of Educational Innovation through Technology (EITT), pages 46–50

    Automatic reconstruction of plane geometry figures in documents. In2015 International Confer- ence of Educational Innovation through Technology (EITT), pages 46–50. IEEE. Xinguo Yu, Yixing Geng, and Zihan Feng. 2021b. Solv- ing solid geometric calculation problems in text. In ...

    2021

  13. [2016]

    In2016 12th IAPR Workshop on Document Analysis Systems (DAS), pages 245–250

    Analysis of stroke intersection for overlap- ping pgf elements. In2016 12th IAPR Workshop on Document Analysis Systems (DAS), pages 245–250. IEEE. Zhe Chen, Jiannan Wu, Wenhai Wang, Weijie Su, Guo Chen, Sen Xing, Muyan Zhong, Qinglong Zhang, Xizhou Zhu, Lewei Lu, and 1 others....

    2025 arXiv

  14. [2017]

    In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 795– 804

    Beyond sentential semantic parsing: Tackling the math sat with a cascade of tree transducers. In Proceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, pages 795– 804. Joy Hsu, Jiajun Wu, and Noah Goodman. 2022. Geo- clidean: Few-shot generaliz...

    2017 arXiv

  15. [2018]

    Maizhen Ning, Qiu-Feng Wang, Kaizhu Huang, and Xiaowei Huang

    Computer-assisted theorem proving in syn- thetic geometry.Handbook of Geometric Constraint Systems Principles, pages 25–73. Maizhen Ning, Qiu-Feng Wang, Kaizhu Huang, and Xiaowei Huang. 2023. A symbolic characters aware model for solving geometry problems. InProceed- ings of t...

    2023

  16. [2019]

    InProceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 6281–6290

    Deep modular co-attention networks for visual question answering. InProceedings of the IEEE/CVF conference on computer vision and pattern recogni- tion, pages 6281–6290. Albert S Yue, Lovish Madaan, Ted Moskovitz, DJ Strouse, and Aaditya K Singh. 2024a. Harp: A challenging hum...

    2024 arXiv

  17. [2021]

    InInternational Conference on Learning Representations

    An image is worth 16x16 words: Transformers for image recognition at scale. InInternational Conference on Learning Representations. Jinghao Dou, Xiaopan Lyu, Xinguo Yu, and Hao Wu

  18. [2023]

    Litian Huang, Xinguo Yu, Feng Xiong, Bin He, Sheng- bing Tang, and Jiawen Fu

    Solving algebraic problems with geometry di- agrams using syntax-semantics diagram understand- ing.Computers, Materials & Continua, 77(1). Litian Huang, Xinguo Yu, Feng Xiong, Bin He, Sheng- bing Tang, and Jiawen Fu. 2024. Hologram rea- soning for solving algebra problems with...

    2024 arXiv

  19. [2024]

    In2024 International Conference on Intelligent Education and Intelligent Research (IEIR), pages 1–6

    An enhanced relation-flow algorithm for solv- ing number line problems. In2024 International Conference on Intelligent Education and Intelligent Research (IEIR), pages 1–6. IEEE. Yifan Du, Zikang Liu, Yifan Li, Wayne Xin Zhao, Yuqi Huo, Bingning Wang, Weipeng Chen, Zheng Liu, ...

    2025 arXiv

  20. [2025]

    Tengjin Weng, Jingyi Wang, Wenhao Jiang, and Zhong Ming

    Feynman: Knowledge-infused diagramming agent for scaling visual reasoning data. Tengjin Weng, Jingyi Wang, Wenhao Jiang, and Zhong Ming. 2025. Visnumbench: Evaluating number sense of multimodal large language models. InProceed- ings of the IEEE/CVF International Conference on ...

    2025

  21. [2026]

    Dongxiang Zhang

    Proposing and solving olympiad geometry with guided tree search.Nature Machine Intelligence, pages 1–12. Dongxiang Zhang. 2022. Deep learning in automatic math word problem solvers. InAI in Learning: De- signing the Future, pages 233–246. Springer Interna- tional Publishing Ch...

    2022 arXiv

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.