pith. sign in

arxiv: 2605.17790 · v1 · pith:CJRKTG7Fnew · submitted 2026-05-18 · 💻 cs.AI

STRIDE: A Self-Reflective Agent Framework for Reliable Automatic Equation Discovery

Jiarui Su , Songjun Tu , Bei Sun , Xiaojun Liang This is my paper

Pith reviewed 2026-05-20 11:05 UTC · model grok-4.3

classification 💻 cs.AI
keywords equation discoverysymbolic regressionLLM agentsself-reflective frameworkclosed-loop discoveryautomatic symbolic laws
0
0 comments X

The pith

STRIDE turns fitted scores and candidate behavior into shared feedback for more reliable LLM equation discovery.

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

Current LLM-based systems for recovering symbolic equations from data often misjudge useful skeletons under unreliable fitting, discard near-correct equations that need repair, and accumulate redundant memories with little guidance. STRIDE addresses these issues by coordinating data-aware generation, mixed-fitting evaluation, critic-executor repair, and diversity-preserving semantic memory inside a closed loop. The framework converts fitted scores and observed candidate behavior into shared feedback that lets the system propose, assess, refine, and reuse equations more effectively. Experiments on standard symbolic-regression benchmarks and LSR-Synth suites report gains in accuracy, out-of-distribution robustness, and structural recovery across several LLM backbones, with ablations attributing the gains to the core components.

Core claim

STRIDE improves reliability by coordinating data-aware generation, mixed-fitting evaluation, critic-executor repair, and diversity-preserving semantic memory. By turning fitted scores and candidate behavior into shared feedback, STRIDE enables equations to be proposed, assessed, refined, and reused within a closed-loop discovery process.

What carries the argument

STRIDE self-reflective agent framework that coordinates data-aware generation, mixed-fitting evaluation, critic-executor repair, and diversity-preserving semantic memory using shared feedback from fitted scores and candidate behavior.

If this is right

  • STRIDE achieves higher accuracy than prior generation-centered loops on representative symbolic-regression benchmarks.
  • STRIDE exhibits improved out-of-distribution robustness when recovering symbolic laws from data.
  • STRIDE recovers equation structures more reliably across multiple LLM backbones.
  • Ablation studies confirm that each of the four coordinated components contributes to the observed performance gains.

Where Pith is reading between the lines

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

  • The closed-loop reflection pattern could be tested on related discovery tasks such as inferring differential equations or physical constraints from observational data.
  • If the semantic memory successfully preserves diversity, it may reduce the total number of LLM calls needed to reach a high-quality equation.
  • The critic-executor repair step might generalize to other agent systems where near-miss solutions can be iteratively corrected rather than discarded.

Load-bearing premise

That turning fitted scores and candidate behavior into shared feedback within the closed-loop process will reliably enable better proposal, assessment, refinement, and reuse without the LLM introducing new systematic errors or biases in the reflection steps.

What would settle it

Running STRIDE on the same benchmarks after removing the reflection and shared-feedback mechanisms and finding no measurable drop in accuracy, OOD robustness, or structural recovery would falsify the central claim.

Figures

Figures reproduced from arXiv: 2605.17790 by Bei Sun, Jiarui Su, Songjun Tu, Xiaojun Liang.

Figure 1
Figure 1. Figure 1: STRIDE versus LLM-SR. The upper part shows a representative LLM-SR iteration that samples a skeleton, [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overall framework of STRIDE: data-aware generation, mixed-fitting evaluation, critic–executor repair, [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: ID and OOD performance on the LSR-Synth suites. Bars report NMSE [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Iteration analysis of STRIDE and its ablated variants. The full system converges more consistently than [PITH_FULL_IMAGE:figures/full_fig_p006_4.png] view at source ↗
Figure 6
Figure 6. Figure 6: High-score semantic versus score-based mem [PITH_FULL_IMAGE:figures/full_fig_p007_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: Reflection gains from critic–executor repair. [PITH_FULL_IMAGE:figures/full_fig_p007_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Cross-domain LSR-Synth case analysis. Shaded regions denote OOD intervals. STRIDE more closely [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Examples of data hints extracted from training data. The hints summarize distribution statistics, symmetry [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Illustration of the proposed mixed parameter [PITH_FULL_IMAGE:figures/full_fig_p015_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Quantitative comparison of parameter fitting strategies across multiple benchmarks. In the equation [PITH_FULL_IMAGE:figures/full_fig_p016_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Illustration of critic–executor reflective repair. The critic agent analyzes a fitted candidate equation, [PITH_FULL_IMAGE:figures/full_fig_p020_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Case studies on the two oscillator benchmarks. We compare phase-space overlays and recovered [PITH_FULL_IMAGE:figures/full_fig_p021_12.png] view at source ↗
Figure 13
Figure 13. Figure 13: Case study on Stress-Strain. The left panel compares predictions across multiple temperatures, and the right panel zooms into the 20◦C condition. 45 48 51 54 57 60 time -0.0004 -0.0002 0 0.0002 Output CRK22 27 30 33 36 39 time -0.2 -0.1 0 0.1 0.2 PO14 532 536 540 544 548 temperature -90 -85 -80 -75 MatSci5 45 48 51 54 57 60 time -0.00045 -0.0003 -0.00015 BPG21 45 48 51 54 57 60 time -0.001 0.001 0.003 Out… view at source ↗
Figure 14
Figure 14. Figure 14: Representative LSR-Synth case studies. The shaded regions denote OOD intervals, and the curves [PITH_FULL_IMAGE:figures/full_fig_p022_14.png] view at source ↗
Figure 15
Figure 15. Figure 15: Prompt templates for the generator agent and critic–executor agents. [PITH_FULL_IMAGE:figures/full_fig_p023_15.png] view at source ↗
read the original abstract

LLM-based equation discovery offers a promising route to recovering symbolic laws from data, but many systems still rely on generation-centered loops that propose candidates, fit parameters, score results, and reuse selected examples. Such loops can misjudge useful skeletons under unreliable fitting, discard near-correct equations that require repair, and accumulate redundant memories that provide limited guidance. We propose STRIDE, a self-reflective agent framework that improves reliability by coordinating data-aware generation, mixed-fitting evaluation, critic--executor repair, and diversity-preserving semantic memory. By turning fitted scores and candidate behavior into shared feedback, STRIDE enables equations to be proposed, assessed, refined, and reused within a closed-loop discovery process. Experiments on representative symbolic-regression benchmarks and LSR-Synth suites show that STRIDE improves accuracy, OOD robustness, and structural recovery across multiple LLM backbones, with ablations and analyses confirming the contribution of its core components.

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 / 2 minor

Summary. The paper proposes STRIDE, a self-reflective agent framework for reliable automatic equation discovery with LLMs. It coordinates data-aware generation, mixed-fitting evaluation, critic-executor repair, and diversity-preserving semantic memory to address limitations in generation-centered loops such as misjudging skeletons under unreliable fitting, discarding near-correct equations, and accumulating redundant memories. By converting fitted scores and candidate behavior into shared feedback, the framework enables improved proposal, assessment, refinement, and reuse in a closed-loop process. Experiments on symbolic-regression benchmarks and LSR-Synth suites report gains in accuracy, OOD robustness, and structural recovery across multiple LLM backbones, with ablations confirming the contributions of core components.

Significance. If the empirical claims hold under rigorous verification, STRIDE could advance LLM-based symbolic regression by providing a more reliable closed-loop mechanism that mitigates common failure modes. The emphasis on self-reflection and semantic memory offers a structured way to improve generalization and structural recovery, with potential value for scientific discovery tasks where symbolic laws must be recovered from noisy or limited data.

major comments (2)
  1. [Abstract] Abstract: The central claim of improved accuracy, OOD robustness, and structural recovery (with ablations confirming component contributions) is asserted without any quantitative results, error bars, dataset details, or statistical tests; this makes it impossible to evaluate the magnitude or reliability of the reported gains and directly undermines assessment of the framework's effectiveness.
  2. [Methods (critic-executor and memory components)] Section describing critic-executor repair and semantic memory: The mechanism for turning fitted scores and candidate behavior into shared feedback is presented as enabling better refinement and reuse, but no analysis or audit is provided to show that LLM reflection steps avoid introducing systematic errors such as hallucinated repairs, inconsistent mathematical judgments, or confirmation bias toward prior skeletons; this is load-bearing for the closed-loop reliability claim.
minor comments (2)
  1. [Methods] Clarify the precise definition and implementation of 'mixed-fitting evaluation' with an example or pseudocode to aid reproducibility.
  2. [Abstract] The abstract mentions 'representative symbolic-regression benchmarks and LSR-Synth suites' without naming the specific datasets or providing references; add these details for context.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive and detailed comments. We address each major point below and outline the revisions we will incorporate to strengthen the manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of improved accuracy, OOD robustness, and structural recovery (with ablations confirming component contributions) is asserted without any quantitative results, error bars, dataset details, or statistical tests; this makes it impossible to evaluate the magnitude or reliability of the reported gains and directly undermines assessment of the framework's effectiveness.

    Authors: We agree that the abstract would be strengthened by including concrete quantitative indicators. In the revised manuscript we will update the abstract to report representative performance gains (e.g., accuracy and structural recovery improvements on the symbolic-regression benchmarks and LSR-Synth suites), reference the specific tables and figures that contain error bars and dataset details, and note that statistical comparisons were performed across LLM backbones. This change preserves abstract length while making the magnitude and reliability of the claims directly evaluable. revision: yes

  2. Referee: [Methods (critic-executor and memory components)] Section describing critic-executor repair and semantic memory: The mechanism for turning fitted scores and candidate behavior into shared feedback is presented as enabling better refinement and reuse, but no analysis or audit is provided to show that LLM reflection steps avoid introducing systematic errors such as hallucinated repairs, inconsistent mathematical judgments, or confirmation bias toward prior skeletons; this is load-bearing for the closed-loop reliability claim.

    Authors: We acknowledge that the absence of a targeted audit of the LLM reflection steps is a limitation for fully substantiating the reliability claim. While our ablations demonstrate net performance benefits, they do not isolate potential hallucination, inconsistency, or bias within individual critic-executor or memory operations. In the revision we will add a dedicated subsection (or supplementary analysis) that samples reflection outputs, compares proposed repairs against ground-truth equations where available, quantifies rates of hallucinated or inconsistent judgments, and discusses observed biases or mitigation strategies. This will directly address the load-bearing concern. revision: yes

Circularity Check

0 steps flagged

No circularity: framework evaluated on external benchmarks

full rationale

The paper describes an applied engineering framework (STRIDE) for LLM-driven symbolic regression. Its central claims concern empirical improvements in accuracy, OOD robustness, and structural recovery, measured directly against independent symbolic-regression benchmarks and LSR-Synth suites. Ablations and analyses are presented as confirming the contribution of core components such as critic-executor repair and semantic memory. No mathematical derivations, first-principles predictions, or fitted parameters are claimed that reduce by construction to the paper's own inputs or self-referential definitions. The evaluation remains externally falsifiable and does not rely on self-citation chains or renaming of known results as novel derivations.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available; no explicit free parameters, axioms, or invented entities are stated or can be inferred beyond the high-level description of the agent components.

pith-pipeline@v0.9.0 · 5691 in / 1205 out tokens · 42097 ms · 2026-05-20T11:05:28.059371+00:00 · methodology

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

50 extracted references · 50 canonical work pages · 4 internal anchors

  1. [1]

    , booktitle =

    Shojaee, Parshin and Meidani, Kazem and Gupta, Shashank and Barati Farimani, Amir and Reddy, Chandan K. , booktitle =. 2025 , url =

    work page 2025

  2. [2]

    International Conference on Machine Learning , year =

    LLM-SRBench: A New Benchmark for Scientific Equation Discovery with Large Language Models , author =. International Conference on Machine Learning , year =

    work page

  3. [3]

    2022 , howpublished =

    gplearn: Genetic Programming in Python, with a scikit-learn Inspired API , author =. 2022 , howpublished =

    work page 2022

  4. [4]

    Interpretable Machine Learning for Science with PySR and SymbolicRegression.jl

    Interpretable Machine Learning for Science with PySR and SymbolicRegression.jl , author =. arXiv preprint arXiv:2305.01582 , year =. 2305.01582 , archivePrefix =

    work page internal anchor Pith review Pith/arXiv arXiv

  5. [5]

    Deep symbolic regression: Recovering mathematical expressions from data via risk-seeking policy gradients

    Deep Symbolic Regression: Recovering Mathematical Expressions from Data via Risk-Seeking Policy Gradients , author =. arXiv preprint arXiv:1912.04871 , year =. 1912.04871 , archivePrefix =

    work page arXiv 1912

  6. [6]

    Advances in Neural Information Processing Systems , year =

    A Unified Framework for Deep Symbolic Regression , author =. Advances in Neural Information Processing Systems , year =

    work page

  7. [7]

    Advances in Neural Information Processing Systems , year =

    Symbolic Regression with a Learned Concept Library , author =. Advances in Neural Information Processing Systems , year =

    work page

  8. [8]

    Proceedings of the National Academy of Sciences , year =

    SR-LLM: An Incremental Symbolic Regression Framework Driven by LLM-Based Retrieval-Augmented Generation , author =. Proceedings of the National Academy of Sciences , year =. doi:10.1073/pnas.2516995122 , url =

    work page doi:10.1073/pnas.2516995122

  9. [9]

    International Conference on Learning Representations , year =

    SR-Scientist: Scientific Equation Discovery With Agentic AI , author =. International Conference on Learning Representations , year =

    work page

  10. [10]

    arXiv preprint arXiv:2506.04282 , year=

    DrSR: LLM based Scientific Equation Discovery with Dual Reasoning from Data and Experience , author =. arXiv preprint arXiv:2506.04282 , year =. doi:10.48550/arXiv.2506.04282 , url =

    work page doi:10.48550/arxiv.2506.04282

  11. [11]

    2025 , howpublished =

    GPT-5.1 Model , author =. 2025 , howpublished =

    work page 2025

  12. [12]

    ACM Computing Surveys , volume =

    Recent Advances in Symbolic Regression , author =. ACM Computing Surveys , volume =. 2025 , doi =

    work page 2025

  13. [13]

    Journal of Data-centric Machine Learning Research , year =

    Rethinking Symbolic Regression Datasets and Benchmarks for Scientific Discovery , author =. Journal of Data-centric Machine Learning Research , year =

    work page

  14. [14]

    Machine Learning and the Physical Sciences Workshop @ NeurIPS 2024 , year =

    Two-Stage Coefficient Estimation in Symbolic Regression for Scientific Discovery , author =. Machine Learning and the Physical Sciences Workshop @ NeurIPS 2024 , year =

    work page 2024

  15. [15]

    Applied Soft Computing , year =

    A Two-Stage Symbolic Regression Method for Discovering Mathematical Formulas , author =. Applied Soft Computing , year =

    work page

  16. [16]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics: Student Research Workshop , year =

    In-Context Symbolic Regression: Leveraging Large Language Models for Function Discovery , author =. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics: Student Research Workshop , year =. doi:10.18653/v1/2024.acl-srw.34 , url =

    work page doi:10.18653/v1/2024.acl-srw.34 2024

  17. [17]

    2006 , doi =

    Numerical Optimization , author =. 2006 , doi =

    work page 2006

  18. [18]

    1987 , publisher=

    Practical Methods of Optimization , author=. 1987 , publisher=

    work page 1987

  19. [19]

    Numerical Analysis: Proceedings of the Biennial Conference Held at Dundee, June 28--July 1, 1977 , pages =

    The Levenberg-Marquardt Algorithm: Implementation and Theory , author =. Numerical Analysis: Proceedings of the Biennial Conference Held at Dundee, June 28--July 1, 1977 , pages =

    work page 1977

  20. [20]

    SIAM Journal on Scientific Computing , volume=

    A subspace, interior, and conjugate gradient method for large-scale bound-constrained minimization problems , author=. SIAM Journal on Scientific Computing , volume=

    work page

  21. [21]

    Information Processing & Management , volume=

    Term-weighting approaches in automatic text retrieval , author=. Information Processing & Management , volume=. 1988 , doi=

    work page 1988

  22. [22]

    2008 , url =

    Introduction to Information Retrieval , author =. 2008 , url =

    work page 2008

  23. [23]

    Science , volume=

    Distilling free-form natural laws from experimental data , author=. Science , volume=. 2009 , doi=

    work page 2009

  24. [24]

    1992 , publisher =

    Genetic Programming: On the Programming of Computers by Means of Natural Selection , author =. 1992 , publisher =

    work page 1992

  25. [25]

    Artificial Intelligence Review , volume =

    Interpretable Scientific Discovery with Symbolic Regression: A Review , author =. Artificial Intelligence Review , volume =. 2024 , doi =

    work page 2024

  26. [26]

    Transactions on Machine Learning Research , year =

    Symbolic Regression is NP-hard , author =. Transactions on Machine Learning Research , year =

    work page

  27. [27]

    Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks , year =

    Contemporary Symbolic Regression Methods and their Relative Performance , author =. Proceedings of the Neural Information Processing Systems Track on Datasets and Benchmarks , year =

    work page

  28. [28]

    Science Advances , volume =

    AI Feynman: A Physics-Inspired Method for Symbolic Regression , author =. Science Advances , volume =. 2020 , doi =

    work page 2020

  29. [29]

    Nature , volume=

    Mathematical Discoveries from Program Search with Large Language Models , author=. Nature , volume=. 2024 , doi=

    work page 2024

  30. [30]

    Machine Learning: Science and Technology , volume =

    Rediscovering Orbital Mechanics with Machine Learning , author =. Machine Learning: Science and Technology , volume =. 2023 , doi =

    work page 2023

  31. [31]

    International Conference on Learning Representations , year =

    Symbolic Physics Learner: Discovering Governing Equations via Monte Carlo Tree Search , author =. International Conference on Learning Representations , year =

    work page

  32. [32]

    2024 , howpublished =

    Claude 3 Haiku: Our Fastest Model Yet , author =. 2024 , howpublished =

    work page 2024

  33. [33]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , month = nov, year =

    A Comprehensive Survey of Scientific Large Language Models and Their Applications in Scientific Discovery , author =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , month = nov, year =. doi:10.18653/v1/2024.emnlp-main.498 , url =

    work page doi:10.18653/v1/2024.emnlp-main.498 2024

  34. [34]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , year =

    Large Language Models Can Self-Improve , author =. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , year =. doi:10.18653/v1/2023.emnlp-main.67 , url =

    work page doi:10.18653/v1/2023.emnlp-main.67 2023

  35. [35]

    Findings of the Association for Computational Linguistics: EMNLP 2023 , year =

    Logic-LM: Empowering Large Language Models with Symbolic Solvers for Faithful Logical Reasoning , author =. Findings of the Association for Computational Linguistics: EMNLP 2023 , year =. doi:10.18653/v1/2023.findings-emnlp.248 , url =

    work page doi:10.18653/v1/2023.findings-emnlp.248 2023

  36. [36]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , month = nov, year =

    Ma, Yubo and Gou, Zhibin and Hao, Junheng and Xu, Ruochen and Wang, Shuohang and Pan, Liangming and Yang, Yujiu and Cao, Yixin and Sun, Aixin , editor =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , month = nov, year =. doi:10.18653/v1/2024.emnlp-main.880 , url =

    work page doi:10.18653/v1/2024.emnlp-main.880 2024

  37. [37]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , month = nov, year =

    Gupta, Priyanshu and Kirtania, Shashank and Singha, Ananya and Gulwani, Sumit and Radhakrishna, Arjun and Soares, Gustavo and Shi, Sherry , editor =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , month = nov, year =. doi:10.18653/v1/2024.emnlp-main.477 , url =

    work page doi:10.18653/v1/2024.emnlp-main.477 2024

  38. [38]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , month = nov, year =

    Xiang, Yufei and Shen, Yiqun and Zhang, Yeqin and Nguyen, Cam-Tu , editor =. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , month = nov, year =. doi:10.18653/v1/2024.emnlp-main.268 , url =

    work page doi:10.18653/v1/2024.emnlp-main.268 2024

  39. [39]

    Compilers Principles, Techniques , author =

    work page

  40. [40]

    Tu, Songjun and Ma, Yiwen and Lin, Jiahao and Zhang, Qichao and Lan, Xiangyuan and Li, Junfeng and Xu, Nan and Li, Linjing and Zhao, Dongbin , journal =

    work page

  41. [41]

    arXiv preprint arXiv:2603.28716 , year =

    Dynamic Dual-Granularity Skill Bank for Agentic RL , author =. arXiv preprint arXiv:2603.28716 , year =

    work page internal anchor Pith review arXiv

  42. [42]

    arXiv preprint arXiv:2602.12259 , year=

    Think like a Scientist: Physics-guided LLM Agent for Equation Discovery , author =. arXiv preprint arXiv:2602.12259 , year =

    work page arXiv

  43. [43]

    The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery

    The AI Scientist: Towards Fully Automated Open-Ended Scientific Discovery , author =. arXiv preprint arXiv:2408.06292 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  44. [44]

    and Cox, Sam and Schilter, Oliver and Baldassari, Carlo and White, Andrew D

    Bran, Andres M. and Cox, Sam and Schilter, Oliver and Baldassari, Carlo and White, Andrew D. and Schwaller, Philippe , journal =

    work page

  45. [45]

    Sun, Jingbo and Chong, Wenyue and Tu, Songjun and Zhang, Qichao and Zhang, Yaocheng and Chai, Jiajun and Wang, Xiaohan and Lin, Wei and Yin, Guojun and Zhao, Dongbin , journal =

    work page

  46. [46]

    Du, Mengge and Chen, Yuntian and Wang, Zhongzheng and Nie, Longfeng and Zhang, Dongxiao , journal =

    work page

  47. [47]

    arXiv preprint arXiv:2402.17879 , year =

    Automated Statistical Model Discovery with Language Models , author =. arXiv preprint arXiv:2402.17879 , year =

    work page arXiv

  48. [48]

    Agent Laboratory: Using LLM Agents as Research Assistants

    Agent Laboratory: Using LLM Agents as Research Assistants , author =. arXiv preprint arXiv:2501.04227 , year =

    work page internal anchor Pith review Pith/arXiv arXiv

  49. [49]

    arXiv preprint arXiv:2505.13259 , year =

    From Automation to Autonomy: A Survey on Large Language Models in Scientific Discovery , author =. arXiv preprint arXiv:2505.13259 , year =

    work page arXiv

  50. [50]

    Proceedings of the First Instructional Conference on Machine Learning , pages =

    Using TF-IDF to Determine Word Relevance in Document Queries , author =. Proceedings of the First Instructional Conference on Machine Learning , pages =

    work page