arxiv: 2605.14033 · v1 · submitted 2026-05-13 · 💻 cs.AI · cs.LG

Recognition: 1 theorem link

· Lean Theorem

Sheaf-Theoretic Transport and Obstruction for Detecting Scientific Theory Shift in AI Agents

David N. Olivieri , Roque J. Hern\'andez

Authors on Pith no claims yet

Pith reviewed 2026-05-15 05:41 UTC · model grok-4.3

classification 💻 cs.AI cs.LG

keywords sheaf theoryobstructiontheory shiftAI agentsrepresentational transportdeformationextensiongluing

0 comments

The pith

Sheaf obstruction measures rank the intended theory deformation or extension as the lowest-failure candidate in AI agent transitions.

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

This paper develops a finite sheaf-theoretic framework that organizes an AI agent's representational contexts into source, overlap, target, and validation charts. It quantifies failure of transport through five obstruction components: residual fit, overlap incompatibility, constraint violation, limiting-relation failure, and representational cost. On a controlled transition-card benchmark, direct ranking by total obstruction identifies the intended change and separates deformation within the existing language from extension of that language. A sympathetic reader cares because the approach isolates a diagnostic subproblem: detecting when current representations stop gluing coherently without requiring full autonomous theory invention.

Core claim

The central claim is that obstruction measures computed from residual fit, overlap incompatibility, constraint violation, limiting-relation failure, and representational cost allow direct ranking of transition candidates, with the intended deformation or extension emerging as the lowest-obstruction choice and transition type separated in the benchmark.

What carries the argument

Finite sheaf structure on contexts with source, overlap, target, and validation charts fitted, restricted, and tested for gluing; obstruction aggregates the five failure modes to rank transition candidates.

If this is right

The intended deformation or extension ranks as the lowest-obstruction candidate on the controlled benchmark.
Obstruction profiles distinguish deformation within a source language from extension of that language.
A constellation kernel over the same signatures supplies a secondary representational-similarity probe.

Where Pith is reading between the lines

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

The same obstruction-ranking procedure could monitor live AI learning to flag when representations require branching.
Obstruction ideas might transfer to detecting incoherence across multiple agents sharing overlapping knowledge bases.
Extending the charts dynamically could turn the diagnostic into an incremental update rule for ongoing agent operation.

Load-bearing premise

The five obstruction components can be defined and combined in a way that reliably separates deformation from extension without post-hoc tuning on the benchmark itself.

What would settle it

A new benchmark of transitions where the lowest-obstruction candidate is not the intended deformation or extension, or where the obstruction profiles fail to separate the two transition types.

Figures

Figures reproduced from arXiv: 2605.14033 by David N. Olivieri, Roque J. Hern\'andez.

**Figure 1.** Figure 1: Geometric intuition for restriction, gluing, and obstruction. The context landscape is covered by overlapping local [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗

**Figure 2.** Figure 2: Finite local-to-global obstruction computation. For each candidate constellation, source and target fits are [PITH_FULL_IMAGE:figures/full_fig_p015_2.png] view at source ↗

**Figure 3.** Figure 3: Workflow from transition cards to obstruction ranking and constellation kernels. Each candidate move [PITH_FULL_IMAGE:figures/full_fig_p020_3.png] view at source ↗

**Figure 4.** Figure 4: Obstruction-margin ledger across transition families. Each row shows one representative transition card. [PITH_FULL_IMAGE:figures/full_fig_p028_4.png] view at source ↗

**Figure 5.** Figure 5: Candidate landscapes for one extension-required and one deformation-su [PITH_FULL_IMAGE:figures/full_fig_p029_5.png] view at source ↗

**Figure 6.** Figure 6: Weight sensitivity and selection stability. Panel (a) shows top-1 accuracy as each obstruction block is multiplied [PITH_FULL_IMAGE:figures/full_fig_p031_6.png] view at source ↗

**Figure 7.** Figure 7: Stress-test margins across expanded transition cards. Panel (a) shows, for each transition family, the margin [PITH_FULL_IMAGE:figures/full_fig_p032_7.png] view at source ↗

**Figure 8.** Figure 8: Robustness under observation noise and reduced record availability. Panel (a) reports mean top-1 accuracy [PITH_FULL_IMAGE:figures/full_fig_p033_8.png] view at source ↗

**Figure 9.** Figure 9: Secondary constellation-kernel probe. Panel (a) ablates the kernel blocks in Eq. [PITH_FULL_IMAGE:figures/full_fig_p034_9.png] view at source ↗

read the original abstract

Scientific theory shift in AI agents requires more than fitting equations to data. An artificial scientific agent must detect whether an existing representational framework remains transportable into a new regime, or whether its language has become locally-to-globally obstructed and must be extended. This paper develops a finite sheaf-theoretic framework for detecting theory-shift candidates through transport and obstruction. Contexts are organized as a local-to-global structure in which source, overlap, target, and validation charts are fitted, restricted, and tested for gluing. Obstruction measures failure of coherence through residual fit, overlap incompatibility, constraint violation, limiting-relation failure, and representational cost. We evaluate the framework on a controlled transition-card benchmark designed to separate deformation within a source language from extension of that language. The main result is direct obstruction ranking: the intended deformation or extension is usually the lowest-obstruction candidate, and transition type is separated in the benchmark. A constellation kernel over the same signatures is included only as a secondary representational-similarity probe. The aim is not to reconstruct historical paradigm shifts or solve open-ended autonomous theory invention, but to isolate a finite diagnostic subproblem for AI agents: detecting when representational transport fails and extension becomes the coherent next move.

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.

Desk Editor's Note private letter to a colleague

The paper sketches a sheaf obstruction ranking to let AI agents detect when to extend rather than deform their representations, but without explicit formulas the benchmark separation could be tuned rather than derived.

read the letter

The core claim is that finite sheaf theory supplies a practical diagnostic for theory shift: organize contexts into source, overlap, target and validation charts, then rank candidates by obstruction to transport. On a controlled transition-card benchmark the lowest-obstruction option matches the intended deformation or extension and the two types separate cleanly. That framing isolates a finite subproblem for autonomous discovery agents without pretending to reconstruct historical shifts or invent new theories from scratch. The five obstruction pieces—residual fit, overlap incompatibility, constraint violation, limiting-relation failure, and representational cost—give a structured way to quantify local-to-global gluing failure, and the benchmark design looks controlled enough to test the distinction. A secondary constellation kernel is included only as a similarity probe, which keeps the focus narrow. The soft spot is the missing definitions. The abstract lists the components but supplies no equations showing how each is extracted from the charts or how they are combined into a scalar score. Without those derivations it is impossible to tell whether the ranking follows from the sheaf structure or whether the measures and weights were adjusted until the benchmark cases worked. The circularity concern in the stress-test note therefore lands: if any component is normalized on the same data used to demonstrate separation, the result is suggestive but not yet diagnostic of the underlying framework. No machine-checked proofs or parameter-free derivations are mentioned, so the empirical separation stands alone. This work is for researchers building AI systems that must manage their own representational changes. A reader looking for concrete diagnostics rather than broad claims about paradigm shifts will find the setup useful. It deserves a serious referee to examine the full derivations, check reproducibility of the ranking, and confirm whether the obstruction is canonical or heuristic.

Referee Report

3 major / 2 minor

Summary. The paper develops a finite sheaf-theoretic framework for detecting scientific theory shifts in AI agents. Contexts are structured as source, overlap, target, and validation charts that are fitted and tested for gluing; obstruction is measured via five components (residual fit, overlap incompatibility, constraint violation, limiting-relation failure, representational cost). On a controlled transition-card benchmark the central claim is that direct obstruction ranking identifies the intended deformation or extension as the lowest-obstruction candidate and separates transition types.

Significance. If the obstruction scores can be shown to be canonically extracted from the sheaf gluing data without benchmark-derived weights or post-hoc selection, the approach would supply a concrete, finite diagnostic for when an AI agent's representational language must be extended rather than merely deformed, addressing a recognized gap between fitting and genuine theory change.

major comments (3)

[Abstract] Abstract: the five obstruction components are enumerated but no equations define how residual fit, overlap incompatibility, constraint violation, limiting-relation failure, or representational cost are computed from the source/overlap/target charts or how they are aggregated into a scalar ranking.
[§4] §4 (benchmark evaluation): the reported separation result lacks error bars, statistical tests, or an explicit statement that the component weights and selection were fixed independently of the benchmark labels; without this the ranking may reduce to a fitted diagnostic.
[§3] §3 (framework definition): the claim that obstruction is sheaf-theoretic requires an explicit derivation showing each component arises from local-to-global gluing failure rather than an ad-hoc linear combination; the current description leaves open the possibility that the measures presuppose the very transition labels they are meant to predict.

minor comments (2)

[Abstract] The constellation kernel is introduced as a secondary probe; its precise relationship to the primary obstruction ranking and whether it uses the same charts should be stated explicitly.
Pseudocode or an algorithmic description of the full obstruction computation pipeline would improve reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive and detailed comments. We address each major point below and have revised the manuscript to improve clarity, rigor, and statistical presentation where the comments identify gaps.

read point-by-point responses

Referee: [Abstract] Abstract: the five obstruction components are enumerated but no equations define how residual fit, overlap incompatibility, constraint violation, limiting-relation failure, or representational cost are computed from the source/overlap/target charts or how they are aggregated into a scalar ranking.

Authors: We agree that the abstract would benefit from greater precision. In the revised manuscript we will insert concise defining expressions for each of the five components (residual fit as the L2 norm of the restriction mismatch, overlap incompatibility as the failure of the cocycle condition on pairwise restrictions, etc.) together with the weighted sum that produces the scalar obstruction score, with pointers to the full derivations in §3. revision: yes
Referee: [§4] §4 (benchmark evaluation): the reported separation result lacks error bars, statistical tests, or an explicit statement that the component weights and selection were fixed independently of the benchmark labels; without this the ranking may reduce to a fitted diagnostic.

Authors: We accept this criticism. The revised §4 will report standard errors or bootstrap confidence intervals for all separation metrics, include a non-parametric statistical test (Wilcoxon rank-sum) comparing obstruction ranks across transition types, and add an explicit paragraph stating that the five component weights were fixed a priori from the sheaf axioms and never tuned on the benchmark labels. revision: yes
Referee: [§3] §3 (framework definition): the claim that obstruction is sheaf-theoretic requires an explicit derivation showing each component arises from local-to-global gluing failure rather than an ad-hoc linear combination; the current description leaves open the possibility that the measures presuppose the very transition labels they are meant to predict.

Authors: We will expand §3 with a dedicated subsection that derives each obstruction term directly from the sheaf gluing axioms. Starting from the equalizer diagram for the source-overlap-target triple, we show that residual fit measures the failure of the restriction maps to equalize, overlap incompatibility quantifies the non-vanishing of the Čech 1-cocycle, and so on; the linear combination is therefore the canonical obstruction class in the first cohomology of the sheaf. The benchmark labels appear only in the validation step and play no role in the computation of the measures themselves. revision: yes

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The paper introduces a sheaf-theoretic framework for obstruction measures (residual fit, overlap incompatibility, constraint violation, limiting-relation failure, representational cost) and reports empirical ranking results on a controlled benchmark. No equations, self-citations, or derivations are quoted in the available text that reduce the obstruction ranking or transition separation to a fitted parameter or input by construction. The benchmark is presented as an external test of the framework rather than the source of its definitions, leaving the central claim independent of the evaluation data.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

The framework rests on standard sheaf axioms for gluing local sections plus domain assumptions that the five obstruction components can be computed from chart restrictions and that the benchmark transitions are representative of real theory-shift candidates.

free parameters (1)

obstruction component weights
The relative importance of residual fit, overlap incompatibility, constraint violation, limiting-relation failure, and representational cost must be chosen or fitted to produce the reported ranking.

axioms (1)

standard math Sheaf gluing condition: local sections on overlapping charts must agree on overlaps to define a global section.
Invoked when testing whether source, overlap, and target charts can be glued without obstruction.

pith-pipeline@v0.9.0 · 5515 in / 1382 out tokens · 35186 ms · 2026-05-15T05:41:38.313087+00:00 · methodology

discussion (0)

Reference graph

Works this paper leans on

43 extracted references · 43 canonical work pages · 2 internal anchors

[1]

, & author Brandenburger, A

author Abramsky, S. , & author Brandenburger, A. ( year 2011 ). title The sheaf-theoretic structure of non-locality and contextuality . journal New Journal of Physics \/ , volume 13 \/ , pages 113036 . :10.1088/1367-2630/13/11/113036

work page doi:10.1088/1367-2630/13/11/113036 2011
[2]

, author Gebhart, T

author Ayzenberg, A. , author Gebhart, T. , author Magai, G. , & author Solomadin, G. ( year 2025 ). title Sheaf theory: from deep geometry to deep learning . arXiv:2502.15476 http://arxiv.org/abs/2502.15476

work page arXiv 2025
[3]

, author Di Giovanni, F

author Bodnar, C. , author Di Giovanni, F. , author Chamberlain, B. P. , author Li \`o , P. , & author Bronstein, M. M. ( year 2022 ). title Neural sheaf diffusion: A topological perspective on heterophily and oversmoothing in GNN s . In booktitle Advances in Neural Information Processing Systems \/ . volume volume 35

work page 2022
[4]

author Borgwardt, K. M. , & author Kriegel, H.-P. ( year 2005 ). title Shortest-path kernels on graphs . In booktitle Proceedings of the Fifth IEEE International Conference on Data Mining \/ (pp. pages 74--81 ). publisher IEEE . :10.1109/ICDM.2005.132

work page doi:10.1109/icdm.2005.132 2005
[5]

author Brunton, S. L. , author Proctor, J. L. , & author Kutz, J. N. ( year 2016 ). title Discovering governing equations from data by sparse identification of nonlinear dynamical systems . journal Proceedings of the National Academy of Sciences \/ , volume 113 \/ , pages 3932--3937 . :10.1073/pnas.1517384113

work page doi:10.1073/pnas.1517384113 2016
[6]

( year 2018 )

author Caramello, O. ( year 2018 ). title Theories, Sites, Toposes: Relating and Studying Mathematical Theories through Topos-Theoretic Bridges \/ . address Oxford : publisher Oxford University Press

work page 2018
[7]

, author Chen, S

author Chen, Z. , author Chen, S. , author Ning, Y. , author Zhang, Q. , author Wang, B. , author Yu, B. , author Li, Y. , author Liao, Z. , author Wei, C. , author Lu, Z. , author Dey, V. , author Xue, M. , author Baker, F. N. , author Burns, B. , author Adu-Ampratwum, D. , author Huang, X. , author Ning, X. , author Gao, S. , author Su, Y. , & author Su...

work page arXiv 2025
[8]

, author Sanchez Gonzalez, A

author Cranmer, M. , author Sanchez Gonzalez, A. , author Battaglia, P. , author Xu, R. , author Cranmer, K. , author Spergel, D. , & author Ho, S. ( year 2020 ). title Discovering symbolic models from deep learning with inductive biases . In booktitle Advances in Neural Information Processing Systems \/ (pp. pages 17429--17442 ). volume volume 33

work page 2020
[9]

author Curry, J. M. ( year 2014 ). title Sheaves, Cosheaves and Applications \/ . Ph.D. thesis University of Pennsylvania. note Available as arXiv:1303.3255

work page internal anchor Pith review Pith/arXiv arXiv 2014
[10]

, & editor Ippoliti, E

editor Danks, D. , & editor Ippoliti, E. (Eds.) ( year 2018 ). title Building Theories: Heuristics and Hypotheses in Sciences \/ volume volume 41 of series Studies in Applied Philosophy, Epistemology and Rational Ethics \/ . address Cham : publisher Springer . :10.1007/978-3-319-72787-5

work page doi:10.1007/978-3-319-72787-5 2018
[11]

, author Hummes Flores, B

author Felber, S. , author Hummes Flores, B. , & author Rincon Galeana, H. ( year 2025 ). title A sheaf-theoretic characterization of tasks in distributed systems . arXiv:2503.02556 http://arxiv.org/abs/2503.02556

work page arXiv 2025
[12]

author Feynman, R. P. , author Leighton, R. B. , & author Sands, M. ( year 2011 ). title The Feynman Lectures on Physics \/ . ( edition The new millennium edition ed.). address New York : publisher Basic Books

work page 2011
[13]

( year 2000 )

author G \"a rdenfors, P. ( year 2000 ). title Conceptual Spaces: The Geometry of Thought \/ . address Cambridge, MA : publisher MIT Press

work page 2000
[14]

, author Flach, P

author G \"a rtner, T. , author Flach, P. , & author Wrobel, S. ( year 2003 ). title On graph kernels: Hardness results and efficient alternatives . In booktitle Learning Theory and Kernel Machines \/ (pp. pages 129--143 ). publisher Springer volume volume 2777 of series Lecture Notes in Computer Science \/ . :10.1007/978-3-540-45167-9_11

work page doi:10.1007/978-3-540-45167-9_11 2003
[15]

, author Hansen, J

author Gebhart, T. , author Hansen, J. , & author Schrater, P. ( year 2023 ). title Knowledge sheaves: A sheaf-theoretic framework for knowledge graph embedding . In booktitle Proceedings of The 26th International Conference on Artificial Intelligence and Statistics \/ . publisher PMLR volume volume 206 of series Proceedings of Machine Learning Research \/

work page 2023
[16]

, & author Ghrist, R

author Hansen, J. , & author Ghrist, R. ( year 2019 ). title Toward a spectral theory of cellular sheaves . journal Journal of Applied and Computational Topology \/ , volume 3 \/ , pages 315--358 . :10.1007/s41468-019-00038-7

work page doi:10.1007/s41468-019-00038-7 2019
[17]

, author Tibshirani, R

author Hastie, T. , author Tibshirani, R. , & author Friedman, J. ( year 2009 ). title The Elements of Statistical Learning: Data Mining, Inference, and Prediction \/ . ( edition 2nd ed.). address New York : publisher Springer

work page 2009
[18]

( year 1999 )

author Haussler, D. ( year 1999 ). title Convolution Kernels on Discrete Structures \/ . type Technical Report number UCSC-CRL-99-10 University of California, Santa Cruz

work page 1999
[19]

, author Sch \"o lkopf, B

author Hofmann, T. , author Sch \"o lkopf, B. , & author Smola, A. J. ( year 2008 ). title Kernel methods in machine learning . journal The Annals of Statistics \/ , volume 36 \/ , pages 1171--1220 . :10.1214/009053607000000677

work page doi:10.1214/009053607000000677 2008
[20]

, author C \^o t \'e , M.-A

author Jansen, P. , author C \^o t \'e , M.-A. , author Khot, T. , author Bransom, E. , author Dalvi Mishra, B. , author Majumder, B. P. , author Tafjord, O. , & author Clark, P. ( year 2024 ). title DiscoveryWorld : A virtual environment for developing and evaluating automated scientific discovery agents . In booktitle Advances in Neural Information Proc...

work page 2024
[21]

author Johnstone, P. T. ( year 2002 ). title Sketches of an Elephant: A Topos Theory Compendium \/ . address Oxford : publisher Oxford University Press

work page 2002
[22]

author Kuhn, T. S. ( year 1962 ). title The Structure of Scientific Revolutions \/ . address Chicago : publisher University of Chicago Press

work page 1962
[23]

, author Simon, H

author Langley, P. , author Simon, H. A. , author Bradshaw, G. L. , & author Zytkow, J. M. ( year 1987 ). title Scientific Discovery: Computational Explorations of the Creative Processes \/ . address Cambridge, MA : publisher MIT Press

work page 1987
[24]

, author Perrone, F

author Lieto, A. , author Perrone, F. , author Pozzato, G. L. , & author Chiodino, E. ( year 2019 ). title Beyond subgoaling: A dynamic knowledge generation framework for creative problem solving in cognitive architectures . journal Cognitive Systems Research \/ , volume 58 \/ , pages 305--316 . :10.1016/j.cogsys.2019.08.005

work page doi:10.1016/j.cogsys.2019.08.005 2019
[25]

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

author Lu, C. , author Lu, C. , author Lange, R. T. , author Foerster, J. , author Clune, J. , & author Ha, D. ( year 2024 ). title The AI scientist: Towards fully automated open-ended scientific discovery . journal arXiv preprint arXiv:2408.06292 \/ ,

work page internal anchor Pith review Pith/arXiv arXiv 2024
[26]

, & author Moerdijk, I

author Mac Lane, S. , & author Moerdijk, I. ( year 1992 ). title Sheaves in Geometry and Logic: A First Introduction to Topos Theory \/ . address New York : publisher Springer

work page 1992
[27]

author Majumder, B. P. , author Surana, H. , author Agarwal, D. , author Dalvi Mishra, B. , author Meena, A. , author Prakhar, A. , author Vora, T. , author Khot, T. , author Sabharwal, A. , & author Clark, P. ( year 2025 ). title DiscoveryBench : Towards data-driven discovery with large language models . In booktitle International Conference on Learning ...

work page arXiv 2025
[28]

, author Chiba, N

author Matsubara, Y. , author Chiba, N. , author Igarashi, R. , & author Ushiku, Y. ( year 2022 ). title Rethinking symbolic regression datasets and benchmarks for scientific discovery . journal arXiv preprint arXiv:2206.10540 \/ , . arXiv:2206.10540 http://arxiv.org/abs/2206.10540

work page arXiv 2022
[29]

editor Morgan, M. S. , & editor Morrison, M. (Eds.) ( year 1999 ). title Models as Mediators: Perspectives on Natural and Social Science \/ . address Cambridge : publisher Cambridge University Press

work page 1999
[30]

author Nersessian, N. J. ( year 2008 ). title Creating Scientific Concepts \/ . address Cambridge, MA : publisher MIT Press

work page 2008
[31]

author Popper, K. R. ( year 1959 ). title The Logic of Scientific Discovery \/ . address London : publisher Hutchinson

work page 1959
[32]

( year 2017 )

author Robinson, M. ( year 2017 ). title Sheaves are the canonical data structure for sensor integration . journal Information Fusion \/ , volume 36 \/ , pages 208--224 . :10.1016/j.inffus.2016.12.002

work page doi:10.1016/j.inffus.2016.12.002 2017
[33]

, author Barekatain, M

author Romera-Paredes, B. , author Barekatain, M. , author Novikov, A. , author Balog, M. , author Kumar, M. P. , author Dupont, E. , author Ruiz, F. J. R. , author Ellenberg, J. S. , author Wang, P. , author Fawzi, O. , author Kohli, P. , & author Fawzi, A. ( year 2024 ). title Mathematical discoveries from program search with large language models . jou...

work page doi:10.1038/s41586-023-06924-6 2024
[34]

Science324(5923), 81–85 (2009).https://doi.org/10.1126/science.1165893

author Schmidt, M. , & author Lipson, H. ( year 2009 ). title Distilling free-form natural laws from experimental data . journal Science \/ , volume 324 \/ , pages 81--85 . :10.1126/science.1165893

work page doi:10.1126/science.1165893 2009
[35]

, & author Smola, A

author Sch \"o lkopf, B. , & author Smola, A. J. ( year 2002 ). title Learning with Kernels: Support Vector Machines, Regularization, Optimization, and Beyond \/ . address Cambridge, MA : publisher MIT Press

work page 2002
[36]

, & author Cristianini, N

author Shawe-Taylor, J. , & author Cristianini, N. ( year 2004 ). title Kernel Methods for Pattern Analysis \/ . address Cambridge : publisher Cambridge University Press

work page 2004
[37]

, author Schweitzer, P

author Shervashidze, N. , author Schweitzer, P. , author van Leeuwen, E. J. , author Mehlhorn, K. , & author Borgwardt, K. M. ( year 2011 ). title Weisfeiler--lehman graph kernels . journal Journal of Machine Learning Research \/ , volume 12 \/ , pages 2539--2561

work page 2011
[38]

( year 1974 )

author Stone, M. ( year 1974 ). title Cross-validatory choice and assessment of statistical predictions . journal Journal of the Royal Statistical Society: Series B \/ , volume 36 \/ , pages 111--147

work page 1974
[39]

( year 2009 )

author Sun, R. ( year 2009 ). title Theoretical status of computational cognitive modeling . journal Cognitive Systems Research \/ , volume 10 \/ , pages 124--140 . :10.1016/j.cogsys.2008.07.002

work page doi:10.1016/j.cogsys.2008.07.002 2009
[40]

( year 2012 )

author Thagard, P. ( year 2012 ). title The Cognitive Science of Science: Explanation, Discovery, and Conceptual Change \/ . address Cambridge, MA : publisher MIT Press

work page 2012
[41]

Science Advances6(16), eaay2631 (2020).https: //doi.org/10.1126/sciadv.aay2631

author Udrescu, S.-M. , & author Tegmark, M. ( year 2020 ). title AI Feynman : A physics-inspired method for symbolic regression . journal Science Advances \/ , volume 6 \/ , pages eaay2631 . :10.1126/sciadv.aay2631

work page doi:10.1126/sciadv.aay2631 2020
[42]

author Vishwanathan, S. V. N. , author Schraudolph, N. N. , author Kondor, R. , & author Borgwardt, K. M. ( year 2010 ). title Graph kernels . journal Journal of Machine Learning Research \/ , volume 11 \/ , pages 1201--1242

work page 2010
[43]

, author Jansen, P

author Wang, R. , author Jansen, P. , author C \^o t \'e , M.-A. , & author Ammanabrolu, P. ( year 2022 ). title ScienceWorld : Is your agent smarter than a 5th grader? In booktitle Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing \/ (pp. pages 11279--11298 ). address Abu Dhabi, United Arab Emirates : publisher Associ...

work page doi:10.18653/v1/2022.emnlp-main.775 2022