pith. sign in

arxiv: 2606.26722 · v1 · pith:U46VOV2Inew · submitted 2026-06-25 · 💻 cs.AI · physics.optics

Socratic agents for autonomous scientific discovery in high-dimensional physical systems

Xianrui Zeng , Pengfei Liu , Yirui Zang , Yang Shen , Fei Yu , Chunlei Yu , Minghao Liu , Yang Du This is my paper

Pith reviewed 2026-06-26 04:42 UTC · model grok-4.3

classification 💻 cs.AI physics.optics
keywords autonomous scientific discoverymulti-agent systemsSocratic interrogationoptical imaginghypothesis generationclosed-loop experimentationmultimode fiberphysical consistency
0
0 comments X

The pith

A multi-agent AI system uses Socratic interrogation to autonomously discover and validate physical hypotheses in complex optical systems.

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

The paper presents AHOIS, a multi-agent system that incorporates Socratic questioning to achieve epistemic autonomy in scientific discovery. It tests this on a multimode-fibre optical platform with complex wave behavior, where the agents propose a random-interference encoding, validate it experimentally, identify failure modes such as instability and noise, and achieve classification performance on standard datasets. This approach allows the system to build and revise explanations without relying on pre-encoded models or human-designed workflows. The results indicate that embedding critical interrogation improves the consistency and validity of discovered hypotheses and experimental plans.

Core claim

AHOIS embeds Socratic midwifery into closed-loop experimentation through a physics-critic agent that interrogates hypotheses via causal questioning, constraint checking, counterexample generation, and falsification-criteria formulation. Evaluated on a real multimode-fibre optical platform without prior encoding schemes, classifiers or speckle models, the system proposed and validated a random-interference encoding hypothesis, discovered task-adaptive sparse-measurement strategies, diagnosed failure modes including encoding instability, fluorescence contamination and detector noise, and translated a published imaging protocol into an executable workflow. The encoding produced 16x16 measuremen

What carries the argument

The physics-critic agent that interrogates hypotheses through causal questioning, constraint checking, counterexample generation and falsification-criteria formulation to improve physical consistency.

If this is right

  • The discovered random-interference encoding enables 16x16 measurements with effective rank 56.9 for image classification tasks.
  • Task-adaptive sparse-measurement strategies emerge autonomously in high-dimensional optical systems.
  • Distinct failure modes such as encoding instability, fluorescence contamination and detector noise are diagnosed without prior models.
  • Published imaging protocols translate into executable workflows on non-original hardware configurations.
  • Socratic interrogation produces measurable gains in physical consistency, hypothesis completeness and experimental-plan validity.

Where Pith is reading between the lines

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

  • This method could extend to other high-dimensional physical systems where wave or field transformations are complex and indirect.
  • Explicit mechanisms for generating counterexamples may prove essential for AI systems to achieve reliable self-correction in experimental settings.
  • The translation of protocols indicates potential for adapting known methods to new hardware without manual redesign.

Load-bearing premise

The physics-critic agent can reliably generate causal questions, counterexamples and falsification criteria that improve physical consistency and experimental validity in systems with unmodeled wave transformations.

What would settle it

A direct comparison showing that hypotheses and plans generated with the physics-critic agent exhibit measurably higher physical consistency and validity than those from the system without it, on repeated trials of the optical platform experiment.

read the original abstract

The automation of scientific discovery has reached an inflection point. While AI systems now operate instruments, optimize parameters and generate hypotheses, most remain procedural: they execute workflows fixed by human designers. True autonomous science demands epistemic autonomy--the capacity to construct, challenge and revise physical explanations in response to evidence. Here we introduce AHOIS, a multi-agent AI scientist that embeds Socratic midwifery into closed-loop experimentation. A physics-critic agent interrogates hypotheses through causal questioning, constraint checking, counterexample generation and falsification-criteria formulation. We evaluate AHOIS on a real multimode-fibre optical platform, a high-dimensional system with complex wave transformations, indirect detection, environmental drift and multi-modal acquisition. Without prior encoding schemes, classifiers or speckle models, the system autonomously proposed and validated a random-interference encoding hypothesis, discovered task-adaptive sparse-measurement strategies, diagnosed distinct failure modes (encoding instability, fluorescence contamination and detector noise) and translated a published imaging protocol into an executable workflow on a non-original configuration. The discovered encoding yielded 16x16 measurements with effective rank 56.9 and classification accuracies of 76.97% on MNIST and 83.17% on Fashion-MNIST. Ablations show that Socratic interrogation improves physical consistency, hypothesis completeness, uncertainty calibration and experimental-plan validity. These results establish a route from workflow automation towards evidence-grounded, self-correcting autonomous discovery in complex physical environments.

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

3 major / 2 minor

Summary. The paper introduces AHOIS, a multi-agent AI system embedding Socratic interrogation via a physics-critic agent for closed-loop autonomous scientific discovery. Evaluated on a real multimode-fibre optical platform without prior encoding schemes, classifiers or speckle models, the system is claimed to have autonomously proposed and validated a random-interference encoding hypothesis, discovered task-adaptive sparse-measurement strategies, diagnosed failure modes (encoding instability, fluorescence contamination, detector noise), translated a published imaging protocol, and achieved 76.97% MNIST and 83.17% Fashion-MNIST classification accuracy using 16x16 measurements of effective rank 56.9. Ablations are reported to demonstrate that Socratic interrogation improves physical consistency, hypothesis completeness, uncertainty calibration and experimental-plan validity.

Significance. If the central claims hold, the work would mark a notable step toward epistemic autonomy in AI-driven physical experimentation, showing that multi-agent Socratic methods can generate and self-correct hypotheses in high-dimensional unmodeled optical systems rather than merely executing human-specified workflows. The concrete performance numbers on real hardware and the reported discovery of an encoding with effective rank 56.9 provide a falsifiable benchmark for future autonomous-discovery systems.

major comments (3)
  1. [Abstract / Ablations] Abstract and ablation results: the claim that the physics-critic produces causal questions, counterexamples and falsification criteria that measurably improve physical consistency rests on reported ablation gains, yet no concrete examples of critic-generated outputs (e.g., specific questions referencing the rank-56.9 encoding, failure-mode diagnoses, or data traces) are supplied; without such traces it is impossible to verify that improvements arise from physically grounded interrogation rather than generic prompting or statistical effects.
  2. [Methods (physics-critic agent)] Methods description of the physics-critic agent: the paper states that the critic formulates falsification criteria and constraint checks for an unmodeled multimode-fibre system whose wave transformations and speckle statistics are not explicitly represented; however, no mechanism is shown by which the critic accesses or reasons over raw measurement data to generate causal counterexamples, leaving the grounding of its contributions unverified.
  3. [Experimental protocol] Experimental protocol and post-hoc handling: the abstract reports concrete accuracies and the effective rank of the discovered encoding, but the precise experimental protocol, data-acquisition sequence, and any post-experiment filtering or selection steps are not detailed; this gap prevents independent assessment of whether the reported performance follows from the autonomous workflow or from unstated human choices.
minor comments (2)
  1. [Results] Notation for 'effective rank 56.9' should be defined explicitly (e.g., via singular-value threshold or participation ratio) and tied to a specific equation or table.
  2. [Ablations] The manuscript would benefit from a table listing the exact critic prompts or output templates used in the ablations to allow reproducibility.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for their constructive comments, which highlight important areas for improving the clarity and verifiability of our claims regarding the physics-critic agent's contributions and the experimental details. We address each major comment below and commit to revisions where the manuscript requires additional evidence or elaboration.

read point-by-point responses

  1. Referee: [Abstract / Ablations] Abstract and ablation results: the claim that the physics-critic produces causal questions, counterexamples and falsification criteria that measurably improve physical consistency rests on reported ablation gains, yet no concrete examples of critic-generated outputs (e.g., specific questions referencing the rank-56.9 encoding, failure-mode diagnoses, or data traces) are supplied; without such traces it is impossible to verify that improvements arise from physically grounded interrogation rather than generic prompting or statistical effects.

    Authors: We agree that the lack of explicit examples of the critic's outputs (such as specific causal questions or counterexamples tied to the discovered encoding and failure modes) makes it difficult to confirm the physical grounding of the improvements. The current manuscript relies on aggregate ablation metrics without providing these traces. We will add a new subsection or appendix with verbatim examples of critic-generated outputs from the experimental runs, including questions referencing the rank-56.9 encoding and diagnoses of encoding instability. revision: yes

  2. Referee: [Methods (physics-critic agent)] Methods description of the physics-critic agent: the paper states that the critic formulates falsification criteria and constraint checks for an unmodeled multimode-fibre system whose wave transformations and speckle statistics are not explicitly represented; however, no mechanism is shown by which the critic accesses or reasons over raw measurement data to generate causal counterexamples, leaving the grounding of its contributions unverified.

    Authors: The manuscript describes the critic's high-level functions (formulating falsification criteria and constraint checks) but does not detail the interface or reasoning process by which it ingests and operates on raw measurement data traces. This is a valid observation, as the grounding mechanism is not explicitly shown. We will revise the Methods section to include a precise description of the data-access pipeline and how the agent derives causal counterexamples from experimental outputs. revision: yes

  3. Referee: [Experimental protocol] Experimental protocol and post-hoc handling: the abstract reports concrete accuracies and the effective rank of the discovered encoding, but the precise experimental protocol, data-acquisition sequence, and any post-experiment filtering or selection steps are not detailed; this gap prevents independent assessment of whether the reported performance follows from the autonomous workflow or from unstated human choices.

    Authors: We acknowledge that the experimental protocol, including the exact data-acquisition sequence and any post-experiment filtering or selection, is not described in sufficient detail to allow independent verification. The abstract presents the final performance metrics without the supporting protocol steps. We will expand the Experimental Setup and Results sections to provide a complete, step-by-step account of the protocol and handling procedures. revision: yes

Circularity Check

0 steps flagged

No significant circularity; results rest on external physical measurements

full rationale

The paper reports outcomes from closed-loop experiments on a real multimode-fibre optical platform, including measured classification accuracies (76.97% MNIST, 83.17% Fashion-MNIST) and effective rank (56.9) obtained from physical data acquisition. No derivation chain, equations, or fitted parameters are presented that reduce the reported results to quantities defined inside the paper by construction. The physics-critic component is evaluated via ablations on experimental validity, but these are empirical comparisons against external benchmarks rather than self-referential definitions or self-citation load-bearing premises. Minor internal references to the AHOIS framework itself do not substitute for the physical measurements. The central claims therefore remain self-contained against external hardware validation.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 2 invented entities

The paper contributes a new agent architecture and its experimental application; it does not introduce new physical constants or laws but relies on standard assumptions about agent reasoning and experimental validity.

axioms (1)
  • domain assumption Socratic-style causal questioning and counterexample generation by an AI agent can systematically improve hypothesis physical consistency and experimental plan validity in complex physical systems.
    Invoked in the design and evaluation of the physics-critic agent and the ablation studies.
invented entities (2)
  • AHOIS multi-agent system no independent evidence
    purpose: To embed epistemic autonomy via Socratic interrogation into closed-loop physical experimentation.
    New system introduced in this work; no independent prior validation cited.
  • physics-critic agent no independent evidence
    purpose: To interrogate hypotheses through causal questioning, constraint checking, counterexample generation and falsification-criteria formulation.
    Core component defined for this paper; no external evidence of its standalone performance provided.

pith-pipeline@v0.9.1-grok · 5806 in / 1576 out tokens · 32396 ms · 2026-06-26T04:42:14.035613+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

62 extracted references · 10 canonical work pages

  1. [1]

    Nature427(6971), 247–252 (2004)

    King, R.D., Whelan, K.E., Jones, F.M., Reiser, P.G., Bryant, C.H., Muggleton, S.H., Kell, D.B., Oliver, S.G.: Functional genomic hypothesis generation and experimentation by a robot scientist. Nature427(6971), 247–252 (2004)

    2004

  2. [2]

    Science324(5923), 85–89 (2009)

    King, R.D., Rowland, J., Oliver, S.G., Young, M., Aubrey, W., Byrne, E., Liakata, M., Markham, M., Pir, P., Soldatova, L.N.,et al.: The automation of science. Science324(5923), 85–89 (2009)

    2009

  3. [3]

    science324(5923), 81–85 (2009)

    Schmidt, M., Lipson, H.: Distilling free-form natural laws from experimental data. science324(5923), 81–85 (2009)

    2009

  4. [4]

    Nature 583(7815), 237–241 (2020)

    Burger, B., Maffettone, P.M., Gusev, V.V., Aitchison, C.M., Bai, Y., Wang, X., Li, X., Alston, B.M., Li, B., Clowes, R.,et al.: A mobile robotic chemist. Nature 583(7815), 237–241 (2020)

    2020

  5. [5]

    Nature624(7990), 86 (2023)

    Szymanski, N.J., Rendy, B., Fei, Y., Kumar, R.E., He, T., Milsted, D., McDer- mott, M.J., Gallant, M., Cubuk, E.D., Merchant, A.,et al.: An autonomous 22 laboratory for the accelerated synthesis of inorganic materials. Nature624(7990), 86 (2023)

    2023

  6. [6]

    Nature, 1–3 (2026)

    Gottweis, J., Weng, W.-H., Daryin, A., Tu, T., Sirkovic, P., Myaskovsky, A., Glowaty, G., Weissenberger, F., Orlandi, A., Popovici, D., et al.: Accelerating scientific discovery with co-scientist. Nature, 1–3 (2026)

    2026

  7. [7]

    Nature, 1–3 (2026)

    Ghareeb, A.E., Chang, B., Mitchener, L., Yiu, A., Szostkiewicz, C.J., Shved, D., Gyimesi, G.J., Laurent, J.M., Wright, S.M., Razzak, M.T., et al.: A multi-agent system for automating scientific discovery. Nature, 1–3 (2026)

    2026

  8. [8]

    arXiv preprint (2026) arXiv:2604.27092

    Yang, S., Chen, F., Zhao, R., Wu, J., Wang, Y., Luo, H., Han, N., Chen, Q., Hu, Y., Li, W., Li, M., Chen, H., Yang, Y.: End-to-end autonomous scientific discovery on a real optical platform. arXiv preprint (2026) arXiv:2604.27092

    Pith/arXiv arXiv 2026

  9. [9]

    Nature (2026) https://doi.org/10

    Ghareeb, A.E., Chang, B., Mitchener, L., Yiu, A., Szostkiewicz, C.J., Laurent, J.M., Razzak, M.T., White, A.D., Hinks, M.M., Rodriques, S.G.: A multi-agent system for automating scientific discovery. Nature (2026) https://doi.org/10. 1038/s41586-026-10652-y

    2026

  10. [10]

    Wang, L., Ma, C., Feng, X., Zhang, Z., Yang, H., Zhang, J., Chen, Z., Tang, J., Chen, X., Lin, Y.,et al.: A survey on large language model based autonomous agents. Front. Comput. Sci.18(6), 186345 (2024) https://doi.org/ 10.1007/s11704-024-40231-1

    work page doi:10.1007/s11704-024-40231-1 2024

  11. [11]

    Guo, T., Chen, X., Wang, Y., Chang, R., Pei, S., Chawla, N.V., Wiest, O.: Large language model based multi-agents: A survey of progress and challenges (2024) https://doi.org/10.13140/RG.2.2.36311.85928

    work page doi:10.13140/rg.2.2.36311.85928 2024

  12. [12]

    arXiv preprint (2025) arXiv:2502.18864

    Gottweis, J., Weng, W.-H., Daryin, A., Tu, T., Palepu, A., Sirkovic, P., Myaskovsky, A., Weissenberger, F., Rong, K., Tanno, R., et al.: Towards an AI co-scientist. arXiv preprint (2025) arXiv:2502.18864

    Pith/arXiv arXiv 2025

  13. [13]

    URL http://dx.doi.org/10.1038/s41586-023-067 92-0

    Boiko, D.A., MacKnight, R., Kline, B., Gomes, G.: Autonomous chemical research with large language models. Nature624(7992), 570–578 (2023) https://doi.org/ 10.1038/s41586-023-06792-0

    work page doi:10.1038/s41586-023-06792-0 2023

  14. [14]

    Nature624(7990), 86–91 (2023) https://doi.org/10.1038/ s41586-023-06734-w

    Szymanski, N.J., Rendy, B., Fei, Y., Kumar, R.E., Milsted, A., McDermott, M., Gallant, M., Cubuk, E.D., Merchant, A., Kim, H., Jain, A., Bartel, C.J., Persson, K., Zeng, Y., Ceder, G.: An autonomous laboratory for the accelerated synthesis of inorganic materials. Nature624(7990), 86–91 (2023) https://doi.org/10.1038/ s41586-023-06734-w

    2023

  15. [15]

    arXiv preprint (2026) arXiv:2603.08127

    Lyu, Y., Zhang, X., Yi, X.: Evoscientist: towards multi-agent evolving AI scien- tists for end-to-end scientific discovery. arXiv preprint (2026) arXiv:2603.08127

    arXiv 2026

  16. [16]

    arXiv preprint (2026) arXiv:2602.00169

    Zhang, H., Li, Y., Huang, W., Hou, Z., Song, Y., Liu, X., Effaty, F., Jiang, J., 23 Wu, S., Ding, Q., et al.: Towards agentic intelligence for materials science. arXiv preprint (2026) arXiv:2602.00169

    arXiv 2026

  17. [17]

    arXiv preprint (2026) arXiv:2604.17406

    Zhu, X., Cai, Y., Liu, Z., Wang, C., Li, F., Jin, W., Liu, W., Bing, Z., Zheng, B., Chai, J., et al.: Evomaster: a foundational evolving agent framework for agentic science at scale. arXiv preprint (2026) arXiv:2604.17406

    Pith/arXiv arXiv 2026

  18. [18]

    arXiv preprint (2026) arXiv:2606.01316

    Zhao, Z., Wen, H., Wu, Y., Ma, J., Wen, Y., Jian, J., Ge, J., Tang, X., An, B., Yin, M., et al.: Science earth: towards a planet-scale operating system for AI-native scientific discovery. arXiv preprint (2026) arXiv:2606.01316

    Pith/arXiv arXiv 2026

  19. [19]

    arXiv preprint (2026) arXiv:2606.10402

    Bianchi, F., Kwon, Y., Pappu, A., Zou, J.: Harnessing the collective intelligence of AI agents in the wild for new discoveries. arXiv preprint (2026) arXiv:2606.10402

    Pith/arXiv arXiv 2026

  20. [20]

    arXiv preprint (2026) arXiv:2605.24018

    Xiong, X., Ren, Y., Xiong, D.: Evosci: a bio-inspired multi-agent framework for the evolution of scientific discovery. arXiv preprint (2026) arXiv:2605.24018

    Pith/arXiv arXiv 2026

  21. [21]

    arXiv preprint (2026) arXiv:2602.07040

    Bicker, E.: Aster: autonomous scientific discovery over 20x faster than existing methods. arXiv preprint (2026) arXiv:2602.07040

    arXiv 2026

  22. [22]

    arXiv preprint (2026) arXiv:2602.13769

    Liu, Q., Hao, R., Li, C., Ma, W.: Or-agent: bridging evolutionary search and structured research for automated algorithm discovery. arXiv preprint (2026) arXiv:2602.13769

    arXiv 2026

  23. [23]

    arXiv preprint (2026) arXiv:2604.02688

    Zhang, C.: Matclaw: an autonomous code-first LLM agent for end-to-end materials exploration. arXiv preprint (2026) arXiv:2604.02688

    Pith/arXiv arXiv 2026

  24. [24]

    arXiv preprint (2026) arXiv:2603.20986

    Manna, S., Chan, H., Sankaranarayanan, S.K.R.S.: Automoose: an agentic AI for autonomous phase-field simulation. arXiv preprint (2026) arXiv:2603.20986

    arXiv 2026

  25. [25]

    arXiv preprint (2026) arXiv:2605.08956

    Bisht, H., Kumar, V., Jablonka, K.M., Mausam, Krishnan, N.M.A.: Agentic AI scientists are not built for autonomous scientific discovery. arXiv preprint (2026) arXiv:2605.08956

    Pith/arXiv arXiv 2026

  26. [26]

    Ho, S.T., Liu, M., Nghiem, H., Huang, F.: Soundnessbench: can your AI sci- entist really tell good research ideas from bad ones? arXiv preprint (2026) arXiv:2605.30329

    Pith/arXiv arXiv 2026

  27. [27]

    arXiv preprint (2026) arXiv:2604.25256

    Lei, X., et al.: Autoresearchbench: benchmarking AI agents on complex scientific literature discovery. arXiv preprint (2026) arXiv:2604.25256

    Pith/arXiv arXiv 2026

  28. [28]

    arXiv preprint (2026) arXiv:2603.09756

    Wua, Y., Su, T., Hu, R., Zhao, M., Hu, S., Pan, D., Huang, J.: Epistemic closure: autonomous mechanism completion for physically consistent simulation. arXiv preprint (2026) arXiv:2603.09756

    arXiv 2026

  29. [29]

    arXiv preprint (2026) arXiv:2606.11851 24

    Chen, J., Liu, S., Yang, L.: Statefuldiscovery: evidence-calibrated claim formation in open-ended scientific discovery. arXiv preprint (2026) arXiv:2606.11851 24

    Pith/arXiv arXiv 2026

  30. [30]

    arXiv preprint (2026) arXiv:2606.08234

    Swaminathan, T., Jiang, R., Zhang, L., Xu, M.: Scitrace: trajectory-aware safety reasoning for scientific discovery agents. arXiv preprint (2026) arXiv:2606.08234

    Pith/arXiv arXiv 2026

  31. [31]

    Shinn, N., Cassano, F., Gopinath, A., Narasimhan, K., Yao, S.: Reflexion: lan- guage agents with verbal reinforcement learning. Adv. Neural Inf. Process. Syst. (NeurIPS)36, 8634–8652 (2023)

    2023

  32. [32]

    Yao, S., Zhao, J., Yu, D., Du, N., Shafran, I., Narasimhan, K., Cao, Y.: React: synergizing reasoning and acting in language models. Int. Conf. Learn. Represent. (ICLR) (2023)

    2023

  33. [33]

    arXiv preprint (2026) arXiv:2603.27584

    Jia, J., Chen, H., Sun, R., Song, Y., Wang, H., Bu, J., Wu, L.: Sci-mind: cognitively-inspired adversarial debate for autonomous mathematical modeling. arXiv preprint (2026) arXiv:2603.27584

    arXiv 2026

  34. [34]

    arXiv preprint (2026) arXiv:2605.23917

    Oh, J., Kim, B., Li, J., Park, Y.J., Park, J.S.: Multi-persona debate sys- tem for automated scientific hypothesis generation. arXiv preprint (2026) arXiv:2605.23917

    Pith/arXiv arXiv 2026

  35. [35]

    arXiv preprint (2026) arXiv:2606.10607

    Li, X., Wang, Y., Li, H., Zhou, C., Gao, E., Han, B., Liu, T., Zhang, K., Bondell, H., Gong, M.: Causal ensemble agent: hierarchical causal discovery with LLM- guided expert reweighting. arXiv preprint (2026) arXiv:2606.10607

    Pith/arXiv arXiv 2026

  36. [36]

    arXiv preprint (2026) arXiv:2605.26029

    Yang, J., Zhang, D., Song, X., Dai, Q., Liu, X., Chen, Y., Vashishtha, A., Shi, J., Tan, C., Peng, H.: Causalab: a scalable environment for interactive causal discovery toward AI scientists. arXiv preprint (2026) arXiv:2605.26029

    Pith/arXiv arXiv 2026

  37. [37]

    Vellekoop, I.M., Mosk, A.P.: Focusing coherent light through opaque strongly scattering media. Opt. Lett.32(16), 2309–2311 (2007) https://doi.org/10.1364/ OL.32.002309

    2007

  38. [38]

    Physical review letters 104(10), 100601 (2010)

    Popoff, S.M., Lerosey, G., Carminati, R., Fink, M., Boccara, A.C., Gigan, S.: Measuring the transmission matrix in optics: An approach to the study and con- trol¡? format?¿ of light propagation in disordered media. Physical review letters 104(10), 100601 (2010)

    2010

  39. [40]

    Reviews of Modern Physics89(1), 015005 (2017)

    Rotter, S., Gigan, S.: Light fields in complex media: Mesoscopic scattering meets wave control. Reviews of Modern Physics89(1), 015005 (2017)

    2017

  40. [41]

    Nature491(7423), 232–234 (2012) https://doi.org/10.1038/nature11578 25

    Bertolotti, J., Putten, E.G., Blum, C., Akbulut, D., Vos, W.L., Lagendijk, A., Vos, W.L.: Non-invasive imaging through opaque scattering layers. Nature491(7423), 232–234 (2012) https://doi.org/10.1038/nature11578 25

    work page doi:10.1038/nature11578 2012

  41. [42]

    Katz, O., Heidmann, P., Fink, M., Gigan, S.: Non-invasive single-shot imag- ing through scattering layers and around corners via speckle correlations. Nat. Photonics8(10), 784–790 (2014) https://doi.org/10.1038/nphoton.2014.189

    work page doi:10.1038/nphoton.2014.189 2014

  42. [43]

    ˇCiˇ zm´ ar, T., Dholakia, K.: Exploiting multimode waveguides for pure fibre-based imaging. Nat. Commun.3, 1027 (2012) https://doi.org/10.1038/ncomms2024

    work page doi:10.1038/ncomms2024 2012

  43. [44]

    Advances in Optics and Photonics15(2), 524–612 (2023)

    Cao, H., ˇCiˇ zm´ ar, T., Turtaev, S., Tyc, T., Rotter, S.: Controlling light propagation in multimode fibers for imaging, spectroscopy, and beyond. Advances in Optics and Photonics15(2), 524–612 (2023)

    2023

  44. [45]

    Nature photonics9(8), 529–535 (2015)

    Pl¨ oschner, M., Tyc, T., ˇCiˇ zm´ ar, T.: Seeing through chaos in multimode fibres. Nature photonics9(8), 529–535 (2015)

    2015

  45. [46]

    Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C.L., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A.,et al.: Training language models to follow instructions with human feedback. Adv. Neural Inf. Process. Syst. (NeurIPS)35, 27730–27744 (2022)

    2022

  46. [47]

    Wei, J., Wang, X., Schuurmans, D., Bosma, M., Ichter, B., Xia, F., Chi, E.H., Le, Q.V., Zhou, D.: Chain-of-thought prompting elicits reasoning in large language models. Adv. Neural Inf. Process. Syst. (NeurIPS)35, 24824–24837 (2022)

    2022

  47. [48]

    Schick, T., Dwivedi-Yu, J., Dess` ı, R., Raileanu, R., Lomeli, M., Zettlemoyer, L., Cancedda, N., Scialom, T.: Toolformer: language models can teach themselves to use tools. Adv. Neural Inf. Process. Syst. (NeurIPS)36, 68539–68551 (2023)

    2023

  48. [49]

    arXiv preprint (2023) arXiv:2309.02427

    Sumers, T.R., Yao, S., Narasimhan, K., Griffiths, T.L.: Cognitive architectures for language agents. arXiv preprint (2023) arXiv:2309.02427

    Pith/arXiv arXiv 2023

  49. [50]

    arXiv preprint (2026) arXiv:2606.03755

    Zhu, L., Gao, L., Chen, Y., Zhu, D., Huang, J.: LAP: an agent-to-instrument protocol for autonomous science. arXiv preprint (2026) arXiv:2606.03755

    Pith/arXiv arXiv 2026

  50. [51]

    Nature Communications15(1), 6572 (2024) https://doi.org/10.1038/ s41467-024-50835-7

    Yang, X.,et al.: Curriculum learning for ab initio deep learned refractive optics. Nature Communications15(1), 6572 (2024) https://doi.org/10.1038/ s41467-024-50835-7

    2024

  51. [52]

    arXiv preprint arXiv:2601.06448 (2026)

    Zeng, X., Zang, Y., Liu, P., Yu, F., Yang, Y., ˇCiˇ zm´ ar, T., Du, Y.: Physics- guided foundation model for universal speckle removal in ultrathin multimode fiber imaging. arXiv preprint arXiv:2601.06448 (2026)

    arXiv 2026

  52. [53]

    Optics & Laser Technology191, 113301 (2025) https://doi.org/https://doi.org

    Xu, J.,et al.: Dual holographic and polarization encoding for high fidelity image transmission through multimode fibers. Optics & Laser Technology191, 113301 (2025) https://doi.org/https://doi.org

    2025

  53. [54]

    Ronneberger, O., Fischer, P., Brox, T.: U-net: convolutional networks for biomedi- cal image segmentation. Med. Image Comput. Comput.-Assist. Interv. (MICCAI), 26 234–241 (2015) https://doi.org/10.1007/978-3-319-24574-4 28

    work page doi:10.1007/978-3-319-24574-4 2015

  54. [55]

    arXiv preprint (2026) arXiv:2606.09550

    Cui, S.: Inquitree: evaluating AI agents in the scientific inquiry loop with paper- derived research trees. arXiv preprint (2026) arXiv:2606.09550

    Pith/arXiv arXiv 2026

  55. [56]

    Matter4(9), 2702–2726 (2021) https://doi.org/ 10.1016/j.matt.2021.06.036

    Stach, E.,et al.: Autonomous experimentation systems for materials develop- ment: A community perspective. Matter4(9), 2702–2726 (2021) https://doi.org/ 10.1016/j.matt.2021.06.036

    work page doi:10.1016/j.matt.2021.06.036 2021

  56. [57]

    arXiv preprint (2026) arXiv:2606.05050

    Song, Z., Zhang, Z., Cheng, L.: Autonomous heterogeneous catalyst dis- covery with a self-evolving multi-agent digital twin. arXiv preprint (2026) arXiv:2606.05050

    Pith/arXiv arXiv 2026

  57. [58]

    arXiv preprint (2026) arXiv:2602.02919

    Jiang, J., Ding, T., Zhu, Z.: Deltaevolve: accelerating scientific discovery through momentum-driven evolution. arXiv preprint (2026) arXiv:2602.02919

    arXiv 2026

  58. [59]

    Krenn, M., Pollice, R., Guo, S.Y., Aldeghi, M., Cervera-Lierta, A., Friederich, P., Passos Gomes, G., H¨ ase, F., Jinich, A., Nigam, A.,et al.: On scientific understanding with artificial intelligence. Nat. Rev. Phys.4(12), 761–769 (2022) https://doi.org/10.1038/s42254-022-00518-3

    work page doi:10.1038/s42254-022-00518-3 2022

  59. [60]

    arXiv preprint (2026) arXiv:2606.14266

    Lin, H., Liu, C., Yan, G.: Large language model based agent for automated discovery in computational physics. arXiv preprint (2026) arXiv:2606.14266

    arXiv 2026

  60. [61]

    arXiv preprint (2026) arXiv:2605.06607

    Somasekharan, N., Pathak, R., Dhanakoti, M., Zhang, T., Yue, L., Zhu, A., Pan, S.: AI CFD scientist: toward open-ended computational fluid dynamics discovery with physics-aware AI agents. arXiv preprint (2026) arXiv:2605.06607

    Pith/arXiv arXiv 2026

  61. [62]

    He, K., Zhang, X., Ren, S., Sun, J.: Deep residual learning for image recognition. Proc. IEEE Conf. Comput. Vis. Pattern Recognit. (CVPR), 770–778 (2016) https: //doi.org/10.1109/CVPR.2016.90

    work page doi:10.1109/cvpr.2016.90 2016

  62. [63]

    Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: Attention is all you need. Adv. Neural Inf. Process. Syst. (NeurIPS)30, 5998–6008 (2017) 27

    2017