pith. machine review for the scientific record. sign in

arxiv: 2605.02819 · v1 · submitted 2026-05-04 · 💻 cs.AI

Recognition: 3 theorem links

· Lean Theorem

SCPRM: A Schema-aware Cumulative Process Reward Model for Knowledge Graph Question Answering

Jiujiu Chen , Yazheng Liu , Sihong Xie , Hui Xiong

Authors on Pith no claims yet

Pith reviewed 2026-05-08 17:51 UTC · model grok-4.3

classification 💻 cs.AI
keywords knowledge graph question answeringprocess reward modelrisk compensationschema distanceMonte Carlo tree searchmulti-hop reasoningcumulative rewards
0
0 comments X

The pith

SCPRM evaluates KG reasoning steps with prefix conditioning and schema distance to supply cumulative plus future rewards.

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

Standard process reward models often assign high scores to flawed multi-hop paths because later correct steps offset earlier errors, an effect worsened in knowledge graphs that contain many routes between the same entities. The paper shows that conditioning rewards on the full reasoning prefix while adding a schema-distance signal to the implicit target parsed from the query supplies both cumulative and forward-looking guidance. This combination is embedded in Monte Carlo Tree Search to steer exploration away from risky paths. On medical, legal, and CWQ benchmarks the resulting SCPRM-MCTS method raises Hits@k by 1.18 percent on average over strong baselines, with particular value in risk-sensitive domains.

Core claim

We propose a Schema-aware Cumulative Process Reward Model (SCPRM) that evaluates reasoning paths by conditioning on the reasoning prefix and incorporating schema distance between current reasoning step and the implicit target parsed from the query, which provides cumulative and future rewards to guide the path explorations. We further integrate SCPRM into Monte Carlo Tree Search (MCTS) as SCPRM-MCTS to conduct multi-hop reasoning on KGs for question answering tasks. Across medical and legal KGQA and CWQ, SCPRM-MCTS improves the performance of Hits@k by an average of 1.18 percent over strong baselines, demonstrating more accurate and risk-sensitive reasoning evaluation.

What carries the argument

The Schema-aware Cumulative Process Reward Model (SCPRM), which augments step-wise rewards by prefix conditioning and schema-distance future signals to the query-derived target.

Load-bearing premise

Schema distance to an implicit target parsed from the query can be computed reliably and supplies a meaningful future-reward signal that actually mitigates risk compensation without introducing new errors.

What would settle it

An ablation that removes the schema-distance term from SCPRM and measures whether Hits@k gains disappear, or manual inspection of scored paths showing that early risky steps followed by corrections still receive high cumulative rewards.

Figures

Figures reproduced from arXiv: 2605.02819 by Hui Xiong, Jiujiu Chen, Sihong Xie, Yazheng Liu.

Figure 1
Figure 1. Figure 1: Comparison of reward models for reasoning based on medical KGQA. Red rewards indicate view at source ↗
Figure 2
Figure 2. Figure 2: The full pipeline of SCPRM training and working with MCTS for reasoning on KGQA. view at source ↗
Figure 3
Figure 3. Figure 3: Hits@k performance (%) of different reward models across datasets. view at source ↗
Figure 4
Figure 4. Figure 4: Performance (%) of different noisy ratios of SCPRM across datasets. view at source ↗
read the original abstract

Large language models excel at complex reasoning, yet evaluating their intermediate steps remains challenging. Although process reward models provide step-wise supervision, they often suffer from a risk compensation effect, where incorrect steps are offset by later correct ones, assigning high rewards to flawed reasoning paths. This issue is further exacerbated in knowledge graph (KG) reasoning, as there may exist multiple paths between the start and end entities in the KGs, and a risky step can make the reasoning path flawed. Those limitations are problematic in risk-sensitive tasks such as medical and legal KG reasoning. To address the issues, we propose a Schema-aware Cumulative Process Reward Model (SCPRM) that evaluates reasoning paths by conditioning on the reasoning prefix , and incorporating schema distance between current reasoning step and the implicit target parsed from the query, which provides cumulative and future rewards to guide the path explorations. We further integrate SCPRM into Monte Carlo Tree Search (MCTS) as SCPRM-MCTS to conduct multi-hop reasoning on KGs for question answering (QA) tasks. Across medical and legal KGQA and CWQ, SCPRM-MCTS improves the performance of Hits@k by an average of 1.18% over strong baselines, demonstrating more accurate and risk-sensitive reasoning evaluation.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 0 minor

Summary. The paper proposes a Schema-aware Cumulative Process Reward Model (SCPRM) for knowledge graph question answering that evaluates reasoning paths by conditioning on the reasoning prefix and adding schema distance between the current step and an implicit target entity parsed from the query. This supplies both cumulative and future rewards to mitigate risk compensation in process reward models. The model is integrated into Monte Carlo Tree Search as SCPRM-MCTS for multi-hop KG reasoning. On medical and legal KGQA tasks plus CWQ, SCPRM-MCTS reports an average 1.18% improvement in Hits@k over strong baselines.

Significance. If the empirical gains hold and the schema-distance signal proves reliable, the work could strengthen process supervision for risk-sensitive KGQA in domains such as medicine and law by reducing the chance that flawed intermediate steps receive high rewards. The MCTS integration provides a concrete mechanism for guiding path search with future-oriented information.

major comments (2)
  1. [Abstract] Abstract: the central claim of a 1.18% average Hits@k improvement is stated without any reference to experimental protocol, baseline definitions, statistical significance tests, ablation results, or error analysis. This information is load-bearing for assessing whether the reported gain supports the model’s effectiveness.
  2. [Method] Method section (SCPRM definition): the implicit target is described as “parsed from the query” and schema distance is used to supply future reward, yet no formal definition, parsing procedure, or validation of parsing accuracy is supplied. Because this distance is the source of the future-reward term, any systematic parsing error would directly corrupt the reward signal and undermine the risk-compensation mitigation claim.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the thoughtful and constructive comments. We address each major comment below, indicating planned revisions to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: the central claim of a 1.18% average Hits@k improvement is stated without any reference to experimental protocol, baseline definitions, statistical significance tests, ablation results, or error analysis. This information is load-bearing for assessing whether the reported gain supports the model’s effectiveness.

    Authors: We agree that the abstract would benefit from additional context on the evaluation. In the revised version we will expand the abstract to briefly note the datasets (medical and legal KGQA tasks plus CWQ), the use of Hits@k against strong baselines, and that the reported gains are supported by ablation studies and statistical significance tests presented in the experimental section. Detailed protocols, baseline definitions, and error analysis remain in the main body due to abstract length constraints. revision: partial

  2. Referee: [Method] Method section (SCPRM definition): the implicit target is described as “parsed from the query” and schema distance is used to supply future reward, yet no formal definition, parsing procedure, or validation of parsing accuracy is supplied. Because this distance is the source of the future-reward term, any systematic parsing error would directly corrupt the reward signal and undermine the risk-compensation mitigation claim.

    Authors: We acknowledge that a formal definition and explicit parsing procedure are required for reproducibility and to substantiate the future-reward claim. In the revised manuscript we will add (1) a mathematical definition of schema distance, (2) a step-by-step description of the query-parsing procedure used to obtain the implicit target entity, and (3) empirical validation of parsing accuracy together with a brief discussion of how parsing errors would affect the reward signal. revision: yes

Circularity Check

0 steps flagged

No circularity: model proposal and empirical gains are independent of self-defined quantities

full rationale

The paper defines SCPRM as a new process reward model that conditions on reasoning prefixes and adds a schema-distance term to an implicit target parsed from the query. It then reports measured Hits@k improvements on external KGQA benchmarks against independent baselines. No equations, uniqueness theorems, or fitted parameters are shown to reduce by construction to the authors' own inputs; the reported 1.18% gain is an external-task measurement rather than a self-referential prediction. The derivation chain therefore remains self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 1 invented entities

Based solely on the abstract, the central claim rests on the untested effectiveness of schema-distance future rewards and their integration with MCTS; no free parameters, axioms, or invented entities are described in detail.

axioms (1)
  • domain assumption Knowledge graphs possess usable schemas that allow distance computation between reasoning steps and an implicit target entity.
    The model relies on schema distance to supply future rewards.
invented entities (1)
  • Implicit target parsed from the query no independent evidence
    purpose: To enable schema-distance calculation for future-reward estimation without knowing the final answer explicitly.
    The abstract states that the query is parsed to an implicit target for distance measurement.

pith-pipeline@v0.9.0 · 5525 in / 1506 out tokens · 79146 ms · 2026-05-08T17:51:14.846542+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

48 extracted references · 9 canonical work pages · 4 internal anchors

  1. [1]

    J. Baek, A. F. Aji, and A. Saffari. Knowledge-augmented language model prompting for zero-shot knowledge graph question answering. InProceedings of the 1st Workshop on Natural Language Reasoning and Structured Explanations (NLRSE), pages 78–106, 2023

    2023

  2. [2]

    K. D. Bollacker, C. Evans, P. K. Paritosh, T. Sturge, and J. Taylor. Freebase: a collaboratively created graph database for structuring human knowledge. InProceedings of the ACM SIGMOD International Conference on Management of Data, SIGMOD 2008, Vancouver, BC, Canada, June 10-12, 2008, pages 1247–1250. ACM, 2008

    2008

  3. [3]

    Bordes, N

    A. Bordes, N. Usunier, A. García-Durán, J. Weston, and O. Yakhnenko. Translating embeddings for modeling multi-relational data. InAdvances in Neural Information Processing Systems 26: 27th Annual Conference on Neural Information Processing Systems 2013. Proceedings of a meeting held December 5-8, 2013, Lake Tahoe, Nevada, United States, pages 2787–2795, 2013

    2013

  4. [4]

    C. Bu, G. Chang, Z. Chen, C. Dang, Z. Wu, Y . He, and X. Wu. Query-driven multimodal graphrag: Dynamic local knowledge graph construction for online reasoning. InFindings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025, Findings of ACL, pages 21360–21380. Association for Computational Linguistics, 2025

    2025

  5. [5]

    H. Cai, S. Zhao, L. Zhang, X. Shen, Q. Xu, W. Shen, Z. Wen, and T. Ban. Unilaw-r1: A large language model for legal reasoning with reinforcement learning and iterative inference. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 18128–18142, 2025

    2025

  6. [6]

    Training Verifiers to Solve Math Word Problems

    K. Cobbe, V . Kosaraju, M. Bavarian, M. Chen, H. Jun, L. Kaiser, M. Plappert, J. Tworek, J. Hilton, R. Nakano, C. Hesse, and J. Schulman. Training verifiers to solve math word problems.CoRR, abs/2110.14168, 2021

    work page internal anchor Pith review arXiv 2021

  7. [7]

    H. Cui, T. Peng, R. Han, J. Han, and L. Liu. Path-based multi-hop reasoning over knowledge graph for answering questions via adversarial reinforcement learning.Knowl. Based Syst., 276:110760, 2023

    2023

  8. [8]

    S. Cui, S. Yu, H.-Y . Huang, Y .-C.-D. Lin, Y . Huang, B. Zhang, J. Xiao, H. Zuo, J. Wang, and Z. Li. mirtarbase 2025: updates to the collection of experimentally validated microrna–target interactions.Nucleic Acids Res., 53(D1):D147–D156, 2025

    2025

  9. [9]

    R. Das, S. Dhuliawala, M. Zaheer, L. Vilnis, I. Durugkar, A. Krishnamurthy, A. Smola, and A. McCallum. Go for a walk and arrive at the answer: Reasoning over paths in knowledge bases using reinforcement learning. In6th International Conference on Learning Representations, ICLR 2018, Vancouver, BC, Canada, April 30 - May 3, 2018, Conference Track Proceedings, 2018

    2018

  10. [10]

    DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning

    DeepSeek-AI. Deepseek-r1: Incentivizing reasoning capability in llms via reinforcement learning.CoRR, abs/2501.12948, 2025

    work page internal anchor Pith review arXiv 2025

  11. [11]

    Rubrics as Rewards: Reinforcement Learning Beyond Verifiable Domains

    A. Gunjal, A. Wang, E. Lau, V . Nath, B. Liu, and S. Hendryx. Rubrics as rewards: Reinforcement learning beyond verifiable domains.CoRR, abs/2507.17746, 2025

    work page internal anchor Pith review arXiv 2025

  12. [12]

    P. E. Hart, N. J. Nilsson, and B. Raphael. A formal basis for the heuristic determination of minimum cost paths.IEEE transactions on Systems Science and Cybernetics, 4(2):100–107, 1968. 10

    1968

  13. [13]

    Huang, J

    Z. Huang, J. Shi, Y . Gao, C. Cui, S. Zhang, J. Li, Y . Zhou, and Q. Cui. HMDD v3.0: a database for experimentally supported human microrna-disease associations.Nucleic Acids Res., 47(Database-Issue):D1013–D1017, 2019

    2019

  14. [14]

    Reinforcement learning with rubric anchors

    Z. Huang, Y . Zhuang, G. Lu, Z. Qin, H. Xu, T. Zhao, R. Peng, J. Hu, Z. Shen, X. Hu, X. Gu, P. Tu, J. Liu, W. Chen, Y . Fu, Z. Fan, Y . Gu, Y . Wang, Z. Yang, J. Li, and J. Zhao. Reinforcement learning with rubric anchors.CoRR, abs/2508.12790, 2025

    work page arXiv 2025

  15. [15]

    Jiang, Y

    S. Jiang, Y . Liao, Z. Chen, Y . Zhang, Y . Wang, and Y . Wang. Meds3: Towards medical slow thinking with self-evolved soft dual-sided process supervision. InProceedings of the AAAI Conference on Artificial Intelligence, volume 40, pages 31319–31327, 2026

    2026

  16. [16]

    J. Li, P. Cao, Y . Chen, J. Xu, H. Li, X. Jiang, K. Liu, and J. Zhao. Towards better chain-of- thought: A reflection on effectiveness and faithfulness. InFindings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025, Findings of ACL, pages 10747–10765. Association for Computational Linguistics, 2025

    2025

  17. [17]

    Q. Li, X. Dai, X. Li, W. Zhang, Y . Wang, R. Tang, and Y . Yu. Codeprm: Execution feedback- enhanced process reward model for code generation. InFindings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025, pages 8169–

    2025

  18. [18]

    Association for Computational Linguistics, 2025

    2025

  19. [19]

    S. Li, Z. Liu, Z. Gui, H. Chen, and W. Zhang. Enrich-on-graph: Query-graph alignment for complex reasoning with LLM enriching. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, EMNLP 2025, Suzhou, China, November 4-9, 2025, pages 7672–7692. Association for Computational Linguistics, 2025

    2025

  20. [20]

    Li and Y

    W. Li and Y . Li. Process reward model with q-value rankings. InThe Thirteenth International Conference on Learning Representations, ICLR 2025, Singapore, April 24-28, 2025, 2025

    2025

  21. [21]

    Lightman, V

    H. Lightman, V . Kosaraju, Y . Burda, H. Edwards, B. Baker, T. Lee, J. Leike, J. Schulman, I. Sutskever, and K. Cobbe. Let’s verify step by step. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024, 2024

    2024

  22. [22]

    X. V . Lin, R. Socher, and C. Xiong. Multi-hop knowledge graph reasoning with reward shaping. InProceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, Brussels, Belgium, October 31 - November 4, 2018, pages 3243–3253. Association for Computational Linguistics, 2018

    2018

  23. [23]

    L. Luo, Y . Li, G. Haffari, and S. Pan. Reasoning on graphs: Faithful and interpretable large lan- guage model reasoning. InThe Twelfth International Conference on Learning Representations, ICLR 2024, Vienna, Austria, May 7-11, 2024, 2024

    2024

  24. [24]

    C. Ma, Y . Chen, T. Wu, A. Khan, and H. Wang. Large language models meet knowledge graphs for question answering: Synthesis and opportunities. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, EMNLP 2025, Suzhou, China, November 4-9, 2025, pages 24578–24597. Association for Computational Linguistics, 2025

    2025

  25. [25]

    Mavromatis and G

    C. Mavromatis and G. Karypis. GNN-RAG: graph neural retrieval for efficient large language model reasoning on knowledge graphs. InFindings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025, Findings of ACL, pages 16682–16699. Association for Computational Linguistics, 2025

    2025

  26. [26]

    T. Mu, A. Helyar, J. Heidecke, J. Achiam, A. Vallone, I. Kivlichan, M. Lin, A. Beutel, J. Schul- man, and L. Weng. Rule based rewards for language model safety. InAdvances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024, 2024

    2024

  27. [27]

    A. Ni, S. Iyer, D. Radev, V . Stoyanov, W. Yih, S. I. Wang, and X. V . Lin. LEVER: learning to verify language-to-code generation with execution. InInternational Conference on Machine Learning, ICML 2023, 23-29 July 2023, Honolulu, Hawaii, USA, volume 202 ofProceedings of Machine Learning Research, pages 26106–26128, 2023. 11

    2023

  28. [28]

    Ouyang, J

    L. Ouyang, J. Wu, X. Jiang, D. Almeida, C. L. Wainwright, P. Mishkin, C. Zhang, S. Agarwal, K. Slama, A. Ray, J. Schulman, J. Hilton, F. Kelton, L. Miller, M. Simens, A. Askell, P. Welinder, P. F. Christiano, J. Leike, and R. Lowe. Training language models to follow instructions with human feedback. InAdvances in Neural Information Processing Systems 35: ...

    2022

  29. [29]

    X. Song, S. Zhang, and T. Yu. Rekg-mcts: Reinforcing LLM reasoning on knowledge graphs via training-free monte carlo tree search. InFindings of the Association for Computational Linguistics, ACL 2025, Vienna, Austria, July 27 - August 1, 2025, pages 9288–9306. Association for Computational Linguistics, 2025

    2025

  30. [30]

    J. Sun, C. Xu, L. Tang, S. Wang, C. Lin, Y . Gong, L. Ni, H.-Y . Shum, and J. Guo. Think-on- graph: Deep and responsible reasoning of large language model on knowledge graph. InThe Twelfth International Conference on Learning Representations, 2024

    2024

  31. [31]

    Z. Sun, Z. Deng, J. Nie, and J. Tang. Rotate: Knowledge graph embedding by relational rotation in complex space. In7th International Conference on Learning Representations, ICLR 2019, New Orleans, LA, USA, May 6-9, 2019, 2019

    2019

  32. [32]

    Talmor and J

    A. Talmor and J. Berant. The web as a knowledge-base for answering complex questions. In Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, NAACL-HLT 2018, New Orleans, Louisiana, USA, June 1-6, 2018, Volume 1 (Long Papers), pages 641–651. Association for Computat...

    2018

  33. [33]

    X. Tan, X. Wang, Q. Liu, X. Xu, X. Yuan, and W. Zhang. Paths-over-graph: Knowledge graph empowered large language model reasoning. InProceedings of the ACM on Web Conference 2025, WWW 2025, Sydney, NSW, Australia, 28 April 2025- 2 May 2025, pages 3505–3522. ACM, 2025

    2025

  34. [34]

    Turpin, J

    M. Turpin, J. Michael, E. Perez, and S. R. Bowman. Language models don’t always say what they think: Unfaithful explanations in chain-of-thought prompting. InAdvances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023, 2023

    2023

  35. [35]

    Solving math word problems with process- and outcome-based feedback

    J. Uesato, N. Kushman, R. Kumar, H. F. Song, N. Y . Siegel, L. Wang, A. Creswell, G. Irving, and I. Higgins. Solving math word problems with process- and outcome-based feedback.CoRR, abs/2211.14275, 2022

    work page Pith review arXiv 2022

  36. [36]

    H. Wang, C. Gao, Q. Xu, B. Liu, G. Hussein, H. R. Korsapati, M. E. Labban, K. Iheasirim, M. Hassan, and G. Anil. Process-supervised reward models for verifying clinical note generation: A scalable approach guided by domain expertise. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 19138–19158, 2025

    2025

  37. [37]

    P. Wang, L. Li, Z. Shao, R. Xu, D. Dai, Y . Li, D. Chen, Y . Wu, and Z. Sui. Math-shepherd: Verify and reinforce llms step-by-step without human annotations. InProceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), ACL 2024, Bangkok, Thailand, August 11-16, 2024, pages 9426–9439, 2024

    2024

  38. [38]

    Y . Wang, S. Fan, M. Wang, S. Gao, C. Wang, and N. Yin. Damr: Efficient and adaptive context- aware knowledge graph question answering with llm-guided mcts.CoRR, abs/2508.00719, 2025

    work page arXiv 2025

  39. [39]

    J. Wei, X. Wang, D. Schuurmans, M. Bosma, B. Ichter, F. Xia, E. H. Chi, Q. V . Le, and D. Zhou. Chain-of-thought prompting elicits reasoning in large language models. InAdvances in Neural Information Processing Systems 35: Annual Conference on Neural Information Processing Systems 2022, NeurIPS 2022, New Orleans, LA, USA, November 28 - December 9, 2022, 2022

    2022

  40. [40]

    Xiong, T

    W. Xiong, T. Hoang, and W. Y . Wang. Deeppath: A reinforcement learning method for knowledge graph reasoning. InProceedings of the 2017 Conference on Empirical Methods in Natural Language Processing, EMNLP 2017, Copenhagen, Denmark, September 9-11, 2017, pages 564–573. Association for Computational Linguistics, 2017. 12

    2017

  41. [41]

    J. Xu, H. Fei, H. Zhou, X. Quan, Q. Huang, S. Wu, W. Y . Wang, M.-L. Lee, and W. Hsu. Logicreward: Incentivizing llm reasoning via step-wise logical supervision.The International Conference on Learning Representations, 2026

    2026

  42. [42]

    S. Yao, D. Yu, J. Zhao, I. Shafran, T. Griffiths, Y . Cao, and K. Narasimhan. Tree of thoughts: Deliberate problem solving with large language models. InAdvances in Neural Information Processing Systems 36: Annual Conference on Neural Information Processing Systems 2023, NeurIPS 2023, New Orleans, LA, USA, December 10 - 16, 2023, 2023

    2023

  43. [43]

    E. Yee, A. Li, C. Tang, Y . H. Jung, R. Paturi, and L. Bergen. Dissociation of faithful and unfaithful reasoning in llms.CoRR, abs/2405.15092, 2024

    work page arXiv 2024

  44. [44]

    F. Yu, A. Gao, and B. Wang. Ovm, outcome-supervised value models for planning in mathe- matical reasoning. InFindings of the Association for Computational Linguistics: NAACL 2024, Mexico City, Mexico, June 16-21, 2024, pages 858–875, 2024

    2024

  45. [45]

    J. Yun, J. Sohn, J. Park, H. Kim, X. Tang, Y . Shao, Y . Koo, M. Ko, Q. Chen, M. Gerstein, M. Moor, and J. Kang. Med-prm: Medical reasoning models with stepwise, guideline-verified process rewards. InProceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, pages 16565–16582, 2025

    2025

  46. [46]

    Zhang, S

    D. Zhang, S. Zhoubian, Z. Hu, Y . Yue, Y . Dong, and J. Tang. Rest-mcts*: LLM self-training via process reward guided tree search. InAdvances in Neural Information Processing Systems 38: Annual Conference on Neural Information Processing Systems 2024, NeurIPS 2024, Vancouver, BC, Canada, December 10 - 15, 2024, 2024

    2024

  47. [47]

    Linking process to outcome: Conditional reward modeling for llm reasoning.arXiv preprint arXiv:2509.26578, 2025

    Z. Zhang, Z. Shan, K. Song, Y . Li, and K. Ren. Linking process to outcome: Conditional reward modeling for LLM reasoning.CoRR, abs/2509.26578, 2025

    work page arXiv 2025

  48. [48]

    A Survey of Process Reward Models: From Outcome Signals to Process Supervisions for Large Language Models

    C. Zheng, J. Zhu, Z. Ou, Y . Chen, K. Zhang, R. Shan, Z. Zheng, M. Yang, J. Lin, Y . Yu, and W. Zhang. A survey of process reward models: From outcome signals to process supervisions for large language models.CoRR, abs/2510.08049, 2025. 13

    work page internal anchor Pith review Pith/arXiv arXiv 2025