Pith. sign in

REVIEW 3 major objections 3 minor 42 references

An Explainable Machine Learning Framework for Railway Predictive Maintenance using Data Streams from the Metro Operator of Portugal

T0 review · 3 major / 3 minor · reviewed 2026-08-05 · deepseek-v4-flash

Pith's one-line read The abstract promises explainable railway fault prediction; the body delivers a different study.

desk verdict The abstract is about railway predictive maintenance, but the full text is a different paper (StructVRM), so the reported F-measure and accuracy are unsupported by the submission. read the letter →

arxiv 2508.05388 v1 pith:UFRMHPCL submitted 2025-08-07 cs.AI

classification cs.AI
keywords predictivemaintenancerailwayfaultpredictiononlinemachinelearningexplainabilityclassimbalanceMetroPTdatasetdatastreamdocumentmismatch
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This submission tries to establish that an online processing pipeline can predict railway equipment faults in real time, build statistical and frequency features on the fly, and explain each alarm in natural language and visuals. The abstract reports above 98% F-measure and above 99% accuracy on data from Portugal's metro operator, under class imbalance and noise. That would matter because maintenance crews could act on early failure signs with understandable justifications, reducing downtime and cost. However, the full text after the abstract is a different paper: it is titled StructVRM and describes a structured verifiable reward model for multimodal reasoning, with no railway data, no feature-construction module, and no explainability module. The claimed railway results therefore have no experimental support in the supplied document.

What carries the argument

For the stated abstract claim, the carrying mechanism is the online pipeline: a pre-processing module that computes statistical and frequency features on the fly, incremental classifiers, and an explainability module that renders decisions as natural language and visuals. For the body actually supplied, the carrying mechanism is different: a trained verifier that outputs a vector of sub-question-level 0/1 scores, whose mean becomes the reward signal in reinforcement learning. Putting the two together, the document has no single mechanism connecting the claimed railway results to any described implementation.

What would settle it

Read the text after the abstract: it is titled StructVRM and contains no MetroPT dataset, no online feature construction, no explainability module, and no railway experiments; the absence of the claimed 98% F-measure and 99% accuracy results in the supplied body is directly observable and settles that the abstract's railway claims are not substantiated by the document.

Watch

Extended reading notes

Core claim

On its own terms, the paper claims to be the first to perform online fault prediction with natural language and visual explainability, using a pipeline of sample pre-processing, incremental machine-learning classification, and outcome explanation, with performance above 98% F-measure and above 99% accuracy on the MetroPT dataset. The supplied full text, however, does not implement this claim. The body is the StructVRM paper, which proposes a model-based verifier that scores each sub-question of a multimodal reasoning problem instead of giving one binary reward, and it reports results on public multimodal reasoning benchmarks rather than on any railway data. No railway predictive-maintenance

Load-bearing premise

The load-bearing premise is that the submitted document is one coherent preprint whose full text implements the abstract's railway pipeline; that premise fails on inspection, and the headline accuracy metric would also need per-class measures to be meaningful for rare faults.

Editorial extensions

If this is right

  • If the claimed pipeline works, maintenance crews could be alerted in real time, with each alarm explained in plain language and visuals, before a fault disrupts service.
  • An F-measure above 98% would mean the system keeps false alarms low while catching nearly all real faults, the balance the paper says protects service availability.
  • Reported robustness under class imbalance and noise would let the same pipeline run on raw operational data streams rather than clean, curated data.
  • Accuracy above 99% would support shifting maintenance from scheduled or reactive work to proactive decisions, cutting cost and improving safety.

Reading between the lines

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

  • This reader's inference: the submission is a composite of two unrelated preprints—the abstract's railway claim and the body's reward-model study are never connected, and no railway experiments appear in the body.
  • Even if a matching implementation existed, aggregate accuracy above 99% would be a weak headline for a rare-fault dataset; per-class precision and recall on faults would settle whether the method actually detects failures.
  • A directly testable extension: run the promised online feature construction and explainability modules on the public MetroPT data with a streaming train/test split, reporting fault-class F1, alarm latency, and explanation faithfulness.
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 / 3 minor

Summary. The manuscript presents, in its title and abstract, an explainable online predictive-maintenance pipeline for the MetroPT railway dataset from the metro operator of Porto, Portugal. The claimed contributions are a sample pre-processing module that builds statistical and frequency features on the fly, an incremental classification module with machine learning models, an explainability module providing natural-language and visual explanations, and experimental results above 98% F-measure and 99% accuracy that are robust to class imbalance and noise. However, the full text of the submission is a different paper: it is titled 'StructVRM: Aligning Multimodal Reasoning with Structured and Verifiable Reward Models,' carries the arXiv header '2508.05383v1 [cs.AI] 7 Aug 2025,' lists ByteDance Seed China as the affiliation, and contains no railway data, no MetroPT processing pipeline, no incremental classification experiments, and no explainability evaluation. The quantitative claims in the abstract therefore have no corresponding methodology, experiments, tables, or code anywhere in the submitted document.

Significance. If the intended railway predictive-maintenance paper were present and its claimed results were reproducible, the contribution could be practically significant for intelligent transportation systems: real-time fault prediction with explainable alarms would address an operational need. The abstract also makes a plausible novelty claim about online fault prediction with natural-language and visual explainability. However, the submission as it stands provides no evidence for any of these claims. The body is an unrelated vision-language reward-modeling paper whose experiments are on multimodal STEM benchmarks; none of its tables, ablations, or error analyses bear on railway fault prediction. Because the central object of review—the railway pipeline—is entirely absent, the significance of the claimed result cannot be assessed, and the manuscript in its current form cannot be considered a coherent scientific submission.

major comments (3)
  1. [Title/Abstract vs. Full Text] The central claim of the paper—an online predictive-maintenance pipeline for the MetroPT railway dataset with F-measure above 98% and accuracy above 99%—appears only in the abstract. The full text, from Section 1 through Section 9, is a different paper titled 'StructVRM: Aligning Multimodal Reasoning with Structured and Verifiable Reward Models,' with its own header 'arXiv:2508.05383v1 [cs.AI] 7 Aug 2025.' There is no section describing the sample pre-processing module, incremental classification, explainability module, MetroPT dataset, or any railway predictive-maintenance experiment. This is not a local omission but the absence of the entire object described in the abstract, and it invalidates the paper's stated contribution.
  2. [Sections 7.1–7.6 (Experiments)] The only experiments in the manuscript are on multimodal reasoning benchmarks (VLM2 Bench, ScienceQA, CMMMU, STEM-Bench, etc.) for the Seed-StructVRM model. None of these experiments involve fault prediction, class imbalance, noise robustness, or explainability in a railway setting. Consequently the abstract's claims of robustness to imbalance and noise, and its claim that 'explanations effectively reflect the decision-making process,' are unsupported by any table, figure, or derivation in the submission.
  3. [Header and Authorship] The full text carries the arXiv identifier 2508.05383v1 and a ByteDance Seed China affiliation, not the submission's ID 2508.05388. It also does not list any metro operator of Porto or railway-related affiliation. This internal inconsistency indicates that the attached body is not the manuscript described by the abstract. At a minimum, the document cannot be evaluated as a coherent preprint, because the reader cannot determine which content is intended to constitute the paper.
minor comments (3)
  1. [Abstract] The abstract states this is 'the first to perform online fault prediction with natural language and visual explainability,' but the full text contains no related-work discussion of railway predictive maintenance, so the novelty claim cannot be checked.
  2. [Full Text] The full text's references ([1]–[42]) are all about vision-language models and reward modeling; none concern MetroPT, railway maintenance, or data streams from transportation systems. If the intended manuscript was accidentally replaced, the reference list should be from the railway paper.
  3. [General] There is no data availability statement, no code release, and no reproducibility description for the claimed railway pipeline. Even for the unrelated StructVRM experiments, no evaluation code or trained-verifier weights are provided.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity detectable: the submitted body is a different paper, and the railway-maintenance derivation chain it advertises is absent, so no claim can be reduced to its own inputs.

full rationale

The provided full text is not the railway predictive-maintenance paper promised by the abstract. It is titled 'StructVRM: Aligning Multimodal Reasoning with Structured and Verifiable Reward Models', carries the header arXiv:2508.05383v1, and presents a vision-language reward-modeling method. Consequently, the claimed MetroPT pipeline (sample pre-processing, incremental classification, explainability, >98% F-measure, >99% accuracy) has no equations, no fitted parameters, and no experimental section in the provided manuscript to audit for circularity. A missing methodology is a manuscript-integrity or correctness problem, not a circularity problem, and per the rules I cannot manufacture a circular step without quotable equations that reduce to themselves. To the extent the body can be assessed on its own terms, StructVRM is self-contained rather than circular: the verifier is trained on LLM-generated structured annotations and validated against human expert judgments (96.83% agreement); the reward is defined as a normalized mean of sub-question scores (Eqs. 1-4); rule-based rewards are explicitly distinguished from the model-based verifier; and the ablations compare configurations with and without StructVRM and without RL. The use of the verifier to filter CoT data used in later training is a training-loop design choice, not a definitional equivalence, and the final benchmark results are evaluated on external benchmarks rather than on the verifier's own training labels. No load-bearing self-citation chain or imported uniqueness theorem is present. Therefore, on the circularity axis, the appropriate finding is 0: no circular step can be identified in the provided text.

Assumptions & free parameters 1 free parameters · 2 assumptions · 1 invented entities

The central claims of the abstract rest entirely on unstated implementation details: classifier choices, feature parameters, the online update scheme, and the evaluation protocol. None of these are auditable because the body of the submission is a different paper. This ledger is accordingly a placeholder: it records the unknowns a real audit would need to resolve, not fitted values that can be inspected.

free parameters (1)
  • Unspecified model hyperparameters and feature-construction settings
    The abstract does not name the classifiers, their hyperparameters, or the statistical and frequency feature parameters of the 'dedicated sample pre-processing module'; none are auditable in the provided text.
assumptions (2)
  • domain assumption The submitted document is a coherent preprint whose body implements the abstract's pipeline
    Invoked by the abstract's claim that experiments 'were performed with the MetroPT data set'; the body does not contain these experiments, so the assumption appears to fail.
  • domain assumption MetroPT fault labels are reliable ground truth for operational faults
    The reported accuracy and F-measure presume the dataset's labels reflect real faults; not verifiable from the abstract alone.
invented entities (1)
  • None identified
    purpose: n/a
    No new particles, mediators, forces, or analogous entities are introduced in the abstract; the claimed 'novel pipeline' is a method, not a postulated entity.

how reviews work

0 comments
Cite this review

Pith. "Pith review of An Explainable Machine Learning Framework for Railway Predictive Maintenance using Data Streams from the Metro Operator of Portugal." pith.science (2026). https://pith.science/paper/UFRMHPCL

@misc{pith2026250805388,
  author       = {Pith},
  title        = {Pith review of: An Explainable Machine Learning Framework for Railway Predictive Maintenance using Data Streams from the Metro Operator of Portugal},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/UFRMHPCL}},
  note         = {Machine review of arXiv:2508.05388}
}
read the original abstract

This work contributes to a real-time data-driven predictive maintenance solution for Intelligent Transportation Systems. The proposed method implements a processing pipeline comprised of sample pre-processing, incremental classification with Machine Learning models, and outcome explanation. This novel online processing pipeline has two main highlights: (i) a dedicated sample pre-processing module, which builds statistical and frequency-related features on the fly, and (ii) an explainability module. This work is the first to perform online fault prediction with natural language and visual explainability. The experiments were performed with the MetroPT data set from the metro operator of Porto, Portugal. The results are above 98 % for F-measure and 99 % for accuracy. In the context of railway predictive maintenance, achieving these high values is crucial due to the practical and operational implications of accurate failure prediction. In the specific case of a high F-measure, this ensures that the system maintains an optimal balance between detecting the highest possible number of real faults and minimizing false alarms, which is crucial for maximizing service availability. Furthermore, the accuracy obtained enables reliability, directly impacting cost reduction and increased safety. The analysis demonstrates that the pipeline maintains high performance even in the presence of class imbalance and noise, and its explanations effectively reflect the decision-making process. These findings validate the methodological soundness of the approach and confirm its practical applicability for supporting proactive maintenance decisions in real-world railway operations. Therefore, by identifying the early signs of failure, this pipeline enables decision-makers to understand the underlying problems and act accordingly swiftly.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

42 extracted references · 6 canonical work pages

  1. [1]

    Dong Guo, Faming Wu, Feida Zhu, Fuxing Leng, Guang Shi, Haobin Chen, Haoqi Fan, Jian Wang, Jianyu Jiang, Jiawei Wang, et al. Seed1. 5-vl technical report.arXiv preprint arXiv:2505.07062, 2025

  2. [2]

    Learning to reason with llms, 2024

    OpenAI. Learning to reason with llms, 2024

  3. [3]

    Gemini 2.5: Our most intelligent ai model, 2025

    Google DeepMind. Gemini 2.5: Our most intelligent ai model, 2025

  4. [4]

    Mathvista: Evaluating mathematical reasoning of foundation models in visual contexts

    Pan Lu, Hritik Bansal, Tony Xia, Jiacheng Liu, Chunyuan Li, Hannaneh Hajishirzi, Hao Cheng, Kai-Wei Chang, Michel Galley, and Jianfeng Gao. Mathvista: Evaluating mathematical reasoning of foundation models in visual contexts. arXiv preprint arXiv:2310.02255, 2023

  5. [5]

    Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems? In European Conference on Computer Vision, pages 169–186

    Renrui Zhang, Dongzhi Jiang, Yichi Zhang, Haokun Lin, Ziyu Guo, Pengshuo Qiu, Aojun Zhou, Pan Lu, Kai-Wei Chang, Yu Qiao, et al. Mathverse: Does your multi-modal llm truly see the diagrams in visual math problems? In European Conference on Computer Vision, pages 169–186. Springer, 2024

  6. [6]

    Solidgeo: Measuring multimodal spatial math reasoning in solid geometry.arXiv preprint arXiv:2505.21177, 2025

    Peijie Wang, Chao Yang, Zhong-Zhi Li, Fei Yin, Dekang Ran, Mi Tian, Zhilong Ji, Jinfeng Bai, and Cheng-Lin Liu. Solidgeo: Measuring multimodal spatial math reasoning in solid geometry.arXiv preprint arXiv:2505.21177, 2025

  7. [7]

    We-math: Does your large multimodal model achieve human-like mathematical reasoning? arXiv preprint arXiv:2407.01284, 2024

    Runqi Qiao, Qiuna Tan, Guanting Dong, Minhui Wu, Chong Sun, Xiaoshuai Song, Zhuoma GongQue, Shanglin Lei, Zhe Wei, Miaoxuan Zhang, et al. We-math: Does your large multimodal model achieve human-like mathematical reasoning? arXiv preprint arXiv:2407.01284, 2024

  8. [8]

    Polymath: A challenging multi-modal mathematical reasoning benchmark.arXiv preprint arXiv:2410.14702, 2024

    Himanshu Gupta, Shreyas Verma, Ujjwala Anantheswaran, Kevin Scaria, Mihir Parmar, Swaroop Mishra, and Chitta Baral. Polymath: A challenging multi-modal mathematical reasoning benchmark.arXiv preprint arXiv:2410.14702, 2024

Show all 42 references
  1. [9]

    Mathscape: Evaluating mllms in multimodal math scenarios through a hierarchical benchmark

    Minxuan Zhou, Hao Liang, Tianpeng Li, Zhiyu Wu, Mingan Lin, Linzhuang Sun, Yaqi Zhou, Yan Zhang, Xiaoqin Huang, Yicong Chen, et al. Mathscape: Evaluating mllms in multimodal math scenarios through a hierarchical benchmark. arXiv preprint arXiv:2408.07543, 2024

  2. [10]

    R-bench: Are your large multimodal model robust to real-world corruptions?arXiv preprint arXiv:2410.05474, 2024

    Chunyi Li, Jianbo Zhang, Zicheng Zhang, Haoning Wu, Yuan Tian, Wei Sun, Guo Lu, Xiaohong Liu, Xiongkuo Min, Weisi Lin, et al. R-bench: Are your large multimodal model robust to real-world corruptions?arXiv preprint arXiv:2410.05474, 2024

  3. [11]

    Olympiadbench: A challenging benchmark for promoting agi with olympiad-level bilingual multimodal scientific problems

    Chaoqun He, Renjie Luo, Yuzhuo Bai, Shengding Hu, Zhen Thai, Junhao Shen, Jinyi Hu, Xu Han, Yujie Huang, Yuxiang Zhang, et al. Olympiadbench: A challenging benchmark for promoting agi with olympiad-level bilingual multimodal scientific problems. InProceedings of the 62nd Annua...

  4. [12]

    Scibench: Evaluating college-level scientific problem-solving abilities of large language models

    Xiaoxuan Wang, Ziniu Hu, Pan Lu, Yanqiao Zhu, Jieyu Zhang, Satyen Subramaniam, Arjun R Loomba, Shichang Zhang, Yizhou Sun, and Wei Wang. Scibench: Evaluating college-level scientific problem-solving abilities of large language models. InInternational Conference on Machine Lear...

  5. [13]

    Scemqa: A scientific college entrance level multimodal question answering benchmark

    Zhenwen Liang, Kehan Guo, Gang Liu, Taicheng Guo, Yujun Zhou, Tianyu Yang, Jiajun Jiao, Renjie Pi, Jipeng Zhang, and Xiangliang Zhang. Scemqa: A scientific college entrance level multimodal question answering benchmark. InProceedings ofthe 62nd AnnualMeeting ofthe Association ...

  6. [14]

    Vnhsge: Vietnamese high school graduation examination dataset for large language models

    Xuan-Quy Dao, Ngoc-Bich Le, The-Duy Vo, Xuan-Dung Phan, Bac-Bien Ngo, Van-Tien Nguyen, Thi-My-Thanh Nguyen, and Hong-Phuoc Nguyen. Vnhsge: Vietnamese high school graduation examination dataset for large language models. arXiv preprint arXiv:2305.12199, 2023

  7. [15]

    Chemvlm: Exploring the power of multimodal large language models in chemistry area

    Junxian Li, Di Zhang, Xunzhi Wang, Zeying Hao, Jingdi Lei, Qian Tan, Cai Zhou, Wei Liu, Yaotian Yang, Xinrui Xiong, et al. Chemvlm: Exploring the power of multimodal large language models in chemistry area. In Proceedings of the AAAI Conference on Artificial Intelligence, numb...

  8. [16]

    Chemrxivquest: A curated chemistry question-answer database extracted from chemrxiv preprints.arXiv preprint arXiv:2505.05232, 2025

    Mahmoud Amiri and Thomas Bocklitz. Chemrxivquest: A curated chemistry question-answer database extracted from chemrxiv preprints.arXiv preprint arXiv:2505.05232, 2025

  9. [17]

    Evaluating the symbol binding ability of large language models for multiple-choice questions in vietnamese general education

    Duc-Vu Nguyen and Quoc-Nam Nguyen. Evaluating the symbol binding ability of large language models for multiple-choice questions in vietnamese general education. InProceedings of the 12th International Symposium on Information and Communication Technology, pages 379–386, 2023. 17

  10. [18]

    R1-vl: Learning to reason with multimodal large language models via step-wise group relative policy optimization.arXiv preprint arXiv:2503.12937, 2025

    Jingyi Zhang, Jiaxing Huang, Huanjin Yao, Shunyu Liu, Xikun Zhang, Shijian Lu, and Dacheng Tao. R1-vl: Learning to reason with multimodal large language models via step-wise group relative policy optimization.arXiv preprint arXiv:2503.12937, 2025

  11. [19]

    Reason-rft: Reinforcement fine-tuning for visual reasoning.arXiv preprint arXiv:2503.20752, 2025

    Huajie Tan, Yuheng Ji, Xiaoshuai Hao, Minglan Lin, Pengwei Wang, Zhongyuan Wang, and Shanghang Zhang. Reason-rft: Reinforcement fine-tuning for visual reasoning.arXiv preprint arXiv:2503.20752, 2025

  12. [20]

    Vlm-r1: A stable and generalizable r1-style large vision-language model.arXiv preprint arXiv:2504.07615, 2025

    Haozhan Shen, Peng Liu, Jingcheng Li, Chunxin Fang, Yibo Ma, Jiajia Liao, Qiaoli Shen, Zilun Zhang, Kangjia Zhao, Qianqian Zhang, et al. Vlm-r1: A stable and generalizable r1-style large vision-language model.arXiv preprint arXiv:2504.07615, 2025

  13. [21]

    Visual-rft: Visual reinforcement fine-tuning.arXiv preprint arXiv:2503.01785, 2025

    Ziyu Liu, Zeyi Sun, Yuhang Zang, Xiaoyi Dong, Yuhang Cao, Haodong Duan, Dahua Lin, and Jiaqi Wang. Visual-rft: Visual reinforcement fine-tuning.arXiv preprint arXiv:2503.01785, 2025

  14. [22]

    Perception-r1: Pioneering perception policy with reinforcement learning.arXiv preprint arXiv:2504.07954, 2025

    En Yu, Kangheng Lin, Liang Zhao, Jisheng Yin, Yana Wei, Yuang Peng, Haoran Wei, Jianjian Sun, Chunrui Han, Zheng Ge, et al. Perception-r1: Pioneering perception policy with reinforcement learning.arXiv preprint arXiv:2504.07954, 2025

  15. [23]

    aha moment

    Hengguang Zhou, Xirui Li, Ruochen Wang, Minhao Cheng, Tianyi Zhou, and Cho-Jui Hsieh. R1-zero’s" aha moment" in visual reasoning on a 2b non-sft model.arXiv preprint arXiv:2503.05132, 2025

  16. [24]

    Kimi Team, Angang Du, Bofei Gao, Bowei Xing, Changjiu Jiang, Cheng Chen, Cheng Li, Chenjun Xiao, Chenzhuang Du, Chonghua Liao, et al. Kimi k1. 5: Scaling reinforcement learning with llms.arXiv preprint arXiv:2501.12599, 2025

  17. [25]

    Dapo: An open-source llm reinforcement learning system at scale

    Qiying Yu, Zheng Zhang, Ruofei Zhu, Yufeng Yuan, Xiaochen Zuo, Yu Yue, Weinan Dai, Tiantian Fan, Gaohong Liu, Lingjun Liu, et al. Dapo: An open-source llm reinforcement learning system at scale. arXiv preprint arXiv:2503.14476, 2025

  18. [26]

    Demystifying long chain-of-thought reasoning in llms.arXiv preprint arXiv:2502.03373, 2025

    Edward Yeo, Yuxuan Tong, Morry Niu, Graham Neubig, and Xiang Yue. Demystifying long chain-of-thought reasoning in llms.arXiv preprint arXiv:2502.03373, 2025

  19. [27]

    Open r1: A fully open reproduction of deepseek-r1, january 2025

    Hugging Face. Open r1: A fully open reproduction of deepseek-r1, january 2025. URL https://github. com/huggingface/open-r1, page 9, 2025

  20. [28]

    Rethinking rl scaling for vision language models: A transparent, from-scratch framework and comprehensive evaluation scheme.arXiv preprint arXiv:2504.02587, 2025

    Yan Ma, Steffi Chern, Xuyang Shen, Yiran Zhong, and Pengfei Liu. Rethinking rl scaling for vision language models: A transparent, from-scratch framework and comprehensive evaluation scheme.arXiv preprint arXiv:2504.02587, 2025

  21. [29]

    Vapo: Efficient and reliable reinforcement learning for advanced reasoning tasks.arXiv preprint arXiv:2504.05118, 2025

    Yu Yue, Yufeng Yuan, Qiying Yu, Xiaochen Zuo, Ruofei Zhu, Wenyuan Xu, Jiaze Chen, Chengyi Wang, TianTian Fan, Zhengyin Du, et al. Vapo: Efficient and reliable reinforcement learning for advanced reasoning tasks.arXiv preprint arXiv:2504.05118, 2025

  22. [30]

    Vl-rethinker: Incentivizing self-reflection of vision-language models with reinforcement learning.arXiv preprint arXiv:2504.08837, 2025

    Haozhe Wang, Chao Qu, Zuming Huang, Wei Chu, Fangzhen Lin, and Wenhu Chen. Vl-rethinker: Incentivizing self-reflection of vision-language models with reinforcement learning.arXiv preprint arXiv:2504.08837, 2025

  23. [31]

    Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025

    Daya Guo, Dejian Yang, Haowei Zhang, Junxiao Song, Ruoyu Zhang, Runxin Xu, Qihao Zhu, Shirong Ma, Peiyi Wang, Xiao Bi, et al. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.arXiv preprint arXiv:2501.12948, 2025

  24. [32]

    Rlpr: Extrapolating rlvr to general domains without verifiers.arXiv preprint arXiv:2506.18254, 2025

    Tianyu Yu, Bo Ji, Shouli Wang, Shu Yao, Zefan Wang, Ganqu Cui, Lifan Yuan, Ning Ding, Yuan Yao, Zhiyuan Liu, et al. Rlpr: Extrapolating rlvr to general domains without verifiers.arXiv preprint arXiv:2506.18254, 2025

  25. [33]

    Grpo-lead: A difficulty-aware reinforcement learning approach for concise mathematical reasoning in language models.arXiv preprint arXiv:2504.09696, 2025

    Jixiao Zhang and Chunsheng Zuo. Grpo-lead: A difficulty-aware reinforcement learning approach for concise mathematical reasoning in language models.arXiv preprint arXiv:2504.09696, 2025

  26. [34]

    Math- shepherd: Verify and reinforce llms step-by-step without human annotations.arXiv preprint arXiv:2312.08935, 2023

    Peiyi Wang, Lei Li, Zhihong Shao, RX Xu, Damai Dai, Yifei Li, Deli Chen, Yu Wu, and Zhifang Sui. Math- shepherd: Verify and reinforce llms step-by-step without human annotations.arXiv preprint arXiv:2312.08935, 2023

  27. [35]

    rstar-math: Small llms can master math reasoning with self-evolved deep thinking.arXiv preprint arXiv:2501.04519, 2025

    Xinyu Guan, Li Lyna Zhang, Yifei Liu, Ning Shang, Youran Sun, Yi Zhu, Fan Yang, and Mao Yang. rstar-math: Small llms can master math reasoning with self-evolved deep thinking.arXiv preprint arXiv:2501.04519, 2025

  28. [36]

    R-prm: Reasoning-driven process reward modeling.arXiv preprint arXiv:2503.21295, 2025

    Shuaijie She, Junxiao Liu, Yifeng Liu, Jiajun Chen, Xin Huang, and Shujian Huang. R-prm: Reasoning-driven process reward modeling.arXiv preprint arXiv:2503.21295, 2025. 18

  29. [37]

    Mm-prm: Enhancing multimodal mathematical reasoning with scalable step-level supervision.arXiv preprint arXiv:2505.13427, 2025

    Lingxiao Du, Fanqing Meng, Zongkai Liu, Zhixiang Zhou, Ping Luo, Qiaosheng Zhang, and Wenqi Shao. Mm-prm: Enhancing multimodal mathematical reasoning with scalable step-level supervision.arXiv preprint arXiv:2505.13427, 2025

  30. [38]

    Reasonflux-prm: Trajectory-aware prms for long chain-of-thought reasoning in llms.arXiv preprint arXiv:2506.18896, 2025

    Jiaru Zou, Ling Yang, Jingwen Gu, Jiahao Qiu, Ke Shen, Jingrui He, and Mengdi Wang. Reasonflux-prm: Trajectory-aware prms for long chain-of-thought reasoning in llms.arXiv preprint arXiv:2506.18896, 2025

  31. [39]

    Judgelm: Fine-tuned large language models are scalable judges

    Lianghui Zhu, Xinggang Wang, and Xinlong Wang. Judgelm: Fine-tuned large language models are scalable judges. arXiv preprint arXiv:2310.17631, 2023

  32. [40]

    Rm-r1: Reward modeling as reasoning.arXiv preprint arXiv:2505.02387, 2025

    Xiusi Chen, Gaotang Li, Ziqi Wang, Bowen Jin, Cheng Qian, Yu Wang, Hongru Wang, Yu Zhang, Denghui Zhang, Tong Zhang, et al. Rm-r1: Reward modeling as reasoning.arXiv preprint arXiv:2505.02387, 2025

  33. [41]

    Gram: A generative foundation reward model for reward generalization.arXiv preprint arXiv:2506.14175, 2025

    Chenglong Wang, Yang Gan, Yifu Huo, Yongyu Mu, Qiaozhi He, Murun Yang, Bei Li, Tong Xiao, Chunliang Zhang, Tongran Liu, et al. Gram: A generative foundation reward model for reward generalization.arXiv preprint arXiv:2506.14175, 2025

  34. [42]

    Reasongrm: Enhancing generative reward models through large reasoning models.arXiv preprint arXiv:2506.16712, 2025

    Bin Chen, Xinzge Gao, Chuanrui Hu, Penghang Yu, Hua Zhang, and Bing-Kun Bao. Reasongrm: Enhancing generative reward models through large reasoning models.arXiv preprint arXiv:2506.16712, 2025. 19

Pith tools

Reviewed August 5, 2026 · model on record in the stance chip above.