Pith. sign in

REVIEW 2 major objections 5 minor 101 references

Evolutionary intelligence turns isolated scientific search into cumulative discovery by retaining experience across cycles.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-13 00:54 UTC pith:AM7KSVIG

load-bearing objection Clean, useful taxonomy that renames the EC-to-agentic-science transition around experience retention; no new results, but the five-question checklist is worth keeping. the 2 major comments →

arxiv 2607.09025 v1 pith:AM7KSVIG submitted 2026-07-10 cs.NE cs.AIcs.CE

Evolutionary Intelligence for Scientific Discovery: From Evolutionary Computation to Cumulative Discovery Systems

Chao Wang , Lingling Li , Fang Liu , Licheng Jiao This is my paper
classification cs.NE cs.AIcs.CE
keywords Evolutionary computationAI for Sciencescientific discoveryautonomous experimentationfoundation modelsself-evolving agentsexperience retentionquality-diversity
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

This review argues that evolutionary computation already gives scientific AI the population-based search it needs for open-ended, feedback-driven exploration, yet traditional practice mostly refines candidates for fixed problems. Cumulative discovery instead requires systems that archive failures, candidate lineages, and experimental records so later cycles reuse that evidence. The authors name this broader practice evolutionary intelligence and supply a five-question framework—what evolves, how candidates change, why they are selected, where feedback originates, and when evolution runs—to show how isolated trajectories become reusable scientific insight. They illustrate the idea from molecules and models through symbolic programs to full automated research workflows, then identify evaluation, process traceability, and shared infrastructure as the bottlenecks that must be solved next. A sympathetic reader cares because the same organizational logic could convert autonomous labs and foundation-model agents from one-off optimizers into systems that actually accumulate knowledge.

Core claim

Evolutionary intelligence characterizes scientific AI systems that sustain exploration by linking candidate refinement with experience retention across evolutionary cycles. A five-dimensional framework—what evolves, how candidates change, why candidates are selected, where feedback originates, and when evolution occurs—clarifies how this linkage transforms isolated search trajectories into cumulative scientific insight.

What carries the argument

The five-dimensional analytical framework (what evolves; how candidates change; why candidates are selected; where feedback originates; when evolution occurs). It is the organizing device that maps diverse scientific AI systems onto a common cycle of experience retention, representation, utilization, and transfer.

Load-bearing premise

The decisive difference between ordinary evolutionary search and cumulative scientific discovery is systematic retention and reuse of experience across cycles.

What would settle it

Run two otherwise identical autonomous discovery loops under the same experimental budget—one that discards non-elite candidates and feedback, one that archives failures, lineages, and experimental logs—and test whether the archiving system yields more transferable design rules or higher success on later, distinct tasks.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • Evaluation must track knowledge transfer and interpretable insight across cycles, not only final candidate quality.
  • Autonomous laboratories and foundation-model agents should treat failures, lineages, and experimental logs as first-class scientific evidence.
  • Shared databases of failed trials and modification histories become necessary infrastructure for cross-domain cumulative discovery.
  • Human roles shift toward setting goals and validating process records rather than hand-tuning every candidate.
  • Systems that evolve research components and full workflows, not just final scientific targets, become the default architecture.

Where Pith is reading between the lines

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

  • If experience retention is the missing piece, quality-diversity archives and multi-fidelity logs may matter more than new mutation operators.
  • The five dimensions can serve as a review checklist that separates genuine cumulative systems from single-cycle optimizers in agentic science claims.
  • Simulation-to-experiment gaps may themselves be treated as evolving objects under the same framework, with the discrepancy archived and corrected over cycles.
  • Longitudinal benchmarks that score reuse of past failures would give the first quantitative test of the evolutionary-intelligence thesis.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

2 major / 5 minor

Summary. This perspective/review argues that evolutionary computation (EC), while well suited to population-based search in open-ended scientific candidate spaces, has largely been used for candidate refinement on predefined problems, and that cumulative scientific discovery additionally requires systematic experience retention across cycles. The authors introduce evolutionary intelligence (EI) as the class of scientific AI systems that couple candidate refinement with structured retention and reuse of full search trajectories (including failures, lineages, and experimental logs). They supply a five-dimensional analytical framework (what evolves; how candidates change; why they are selected; where feedback originates; when evolution occurs), describe an EI cycle from experience retention through knowledge representation, utilization, and transfer to scientific insight, illustrate the paradigm across four discovery modes (concrete entities, computational models/representations, symbolic reasoning, and automated research workflows), and outline bottlenecks in evaluation, process traceability, simulation-to-experiment gaps, and shared infrastructure.

Significance. If adopted, the paper would give the AI-for-science and evolutionary-computation communities a shared vocabulary for systems that already mix foundation models, autonomous labs, and evolutionary search (FunSearch, AlphaEvolve, self-driving labs, agentic science). The five-question checklist and the EC-to-EI comparison in Table 1 are concrete organizing tools rather than pure slogans, and the roadmap in Section 6 correctly elevates longitudinal evaluation, lineage traceability, and shared negative-result infrastructure as first-class research problems. Strengths include broad, up-to-date literature coverage spanning classical EC, quality-diversity, neuroevolution, foundation-model program search, and closed-loop experimentation, plus an explicit framing of failed trials and candidate lineages as scientific evidence. As a definitional perspective rather than an empirical claim, its value is taxonomic and agenda-setting; that value is real for a field currently fragmenting across agentic science, self-evolving agents, and autonomous discovery.

major comments (2)
  1. [Section 2, Table 1] Section 2 and Table 1: the load-bearing demarcation between EC archives (elites, non-dominated fronts, diverse behaviours, search states) and EI archives (failures, lineages, experimental records, annotations) is under-specified relative to quality-diversity / MAP-Elites and evolutionary transfer optimization, which already retain diverse behaviours, failed niches, and cross-task experience. The manuscript should state an operational criterion for when an existing QD or transfer-EC system counts as EI (e.g., explicit reuse of failure lineages to reshape proposal distributions or experimental priorities across distinct scientific tasks), otherwise the EC-to-EI transition risks being a relabeling of practices already present in the cited literature (Mouret & Clune 2015; Cully et al. 2015; Tan et al. 2021).
  2. [Section 6] Section 6 (Evaluating cumulative discovery): the call for longitudinal metrics that track knowledge utilization, transfer, and interpretable insight is central to the paper's claim that EI is more than candidate refinement, yet no concrete, falsifiable metric or protocol is proposed (e.g., transfer gain after N cycles with vs. without failure archives; reproducibility of derived design rules; reduction in repeated constraint violations). Without at least one worked example of such a metric applied to a cited system (FunSearch, a self-driving lab, or Promptbreeder), the evaluation roadmap remains aspirational and does not yet allow readers to test whether a system is accumulating insight or merely refining candidates.
minor comments (5)
  1. [Fig. 1, Fig. 2] Figure 1 and Figure 2: several labels are misspelled or broken across lines ("Scientific Targrts", "Intera ction", "Feed back", "To" dangling). These should be corrected for production quality.
  2. [Abstract, Section 1] Abstract and Section 1: "experience retention" is introduced as the decisive gap between EC and cumulative discovery; a short sentence acknowledging alternative candidates (richer multi-objective selection, higher-fidelity surrogates, tighter human oversight) and stating why the authors privilege retention would reduce the appearance of stipulation without changing the thesis.
  3. [Section 5] Section 5 modes: the four modes are clear, but a compact table mapping 2-3 representative systems per mode onto the five dimensions would make the taxonomy immediately usable for readers and reviewers.
  4. [Section 2] Terminology consistency: "evolutionary intelligence (EI)" is sometimes used as a system class and sometimes as a cycle/process; a single definitional sentence early in Section 2 would help.
  5. [References] A few references carry 2025-2026 dates and arXiv IDs that may shift before publication; verify final bibliographic details and, where possible, prefer peer-reviewed versions.

Circularity Check

0 steps flagged

Definitional reframing of EC as EI via experience retention; no fitted predictions, forced uniqueness, or load-bearing self-citation reductions.

full rationale

This is a perspective/review that stipulatively introduces evolutionary intelligence (EI) as the class of systems linking candidate refinement to experience retention, then organizes existing literature (FunSearch, AlphaEvolve, autonomous labs, quality-diversity methods, etc.) under a five-question checklist. No equations, parameters, or quantitative predictions appear; nothing is fitted to data and then re-presented as a forecast, and no uniqueness theorem or ansatz is imported from overlapping-author prior work to forbid alternatives. Self-citations (e.g., Wang et al. on LLM-EC intersections or multitask transfer) are ordinary background and do not underwrite a central formal claim. The only mild definitional character is that EI is defined to do what the paper then says it does; that is normal for a conceptual review and does not reduce any independent result by construction. Score remains 1 solely to register the stipulative premise; the derivation chain is otherwise empty and self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 3 axioms · 2 invented entities

As a conceptual review the paper rests on standard background facts about evolutionary computation and on the authors’ definitional choice to elevate experience retention as the distinguishing feature of EI. No numerical free parameters appear. The principal invented entities are the EI label itself and the five-dimensional analytical frame.

axioms (3)
  • domain assumption Population-based evolutionary search is well-suited to open-ended scientific candidate spaces that lack reliable gradients and contain noisy or costly feedback.
    Stated in Section 1 and used throughout as the computational basis for EI; drawn from the classical EC literature the authors cite.
  • ad hoc to paper Cumulative scientific discovery requires systematic retention and reuse of the full search trajectory (including failures and lineages), not merely elite candidates.
    Core definitional premise introduced in the abstract and Sections 1–2 that converts ordinary EC into the authors’ notion of EI.
  • domain assumption Scientific feedback comprises computational proxies, physical experimental outcomes, and human expert judgement, all of which can reshape subsequent selection and variation.
    Assumed in Sections 2–3 when expanding the classical scalar-fitness model of EC.
invented entities (2)
  • Evolutionary intelligence (EI) no independent evidence
    purpose: Label for scientific AI systems that couple candidate refinement with structured experience retention across evolutionary cycles.
    Newly coined framing that organizes the review; independent evidence is limited to the descriptive power of the subsequent literature survey.
  • Five-dimensional analytical framework (what/how/why/where/when) no independent evidence
    purpose: Checklist for dissecting any scientific AI system and for distinguishing isolated search from cumulative discovery.
    Original organizational device of the paper; its utility is demonstrated by re-describing existing systems rather than by external validation.

pith-pipeline@v1.1.0-grok45 · 18067 in / 2261 out tokens · 29325 ms · 2026-07-13T00:54:09.465162+00:00 · methodology

0 comments
read the original abstract

Artificial intelligence (AI) is shifting scientific discovery from task-specific workflows towards autonomous systems that organize exploration with experimental and human feedback in open-ended candidate spaces. Evolutionary computation (EC) provides a computational basis for feedback-driven discovery because population-based search can maintain diverse scientific candidates while steering exploration through accumulated evidence. However, EC predominantly focuses on candidate refinement for predefined problems, whereas cumulative discovery requires experience retention. To bridge this gap, this review introduces evolutionary intelligence (EI) for scientific discovery. EI characterizes scientific AI systems that sustain exploration by linking candidate refinement with experience retention across evolutionary cycles. We introduce a five-dimensional analytical framework that asks what evolves, how candidates change, why candidates are selected, where feedback originates, and when evolution occurs. This framework clarifies how EI transforms isolated search trajectories into cumulative scientific insight. We further demonstrate this paradigm across diverse discovery modes, from evolving concrete scientific entities to orchestrating automated research workflows. Finally, we identify critical bottlenecks regarding evaluation, process traceability, and shared infrastructure, providing a concrete roadmap for advancing the transition from EC to EI in scientific discovery.

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

101 extracted references · 22 linked inside Pith

  1. [1]

    Wang, H.et al.Scientific discovery in the age of artificial intelligence.Nature 620, 47–60 (2023)

    2023

  2. [2]

    J., Tansley, S., Tolle, K

    Hey, A. J., Tansley, S., Tolle, K. M.et al. The fourth paradigm: data-intensive scientific discoveryVol. 1 (Microsoft research Redmond, WA, 2009)

    2009

  3. [3]

    T., Davies, D

    Butler, K. T., Davies, D. W., Cartwright, H., Isayev, O. & Walsh, A. Machine learning for molecular and materials science.Nature559, 547–555 (2018)

    2018

  4. [4]

    Merchant, A.et al.Scaling deep learning for materials discovery.Nature624, 80–85 (2023)

    2023

  5. [5]

    nature596, 583–589 (2021)

    Jumper, J.et al.Highly accurate protein structure prediction with alphafold. nature596, 583–589 (2021)

    2021

  6. [6]

    D.et al.The automation of science.Science324, 85–89 (2009)

    King, R. D.et al.The automation of science.Science324, 85–89 (2009)

    2009

  7. [7]

    A., MacKnight, R., Kline, B

    Boiko, D. A., MacKnight, R., Kline, B. & Gomes, G. Autonomous chemical research with large language models.Nature624, 570–578 (2023)

    2023

  8. [8]

    Lehman, J.et al.Evolution through large models.arXiv preprint arXiv:2206.08896(2022)

    Pith/arXiv arXiv 2022

  9. [9]

    P.et al.Self-driving laboratory for accelerated discovery of thin- film materials.Science Advances6, eaaz8867 (2020)

    MacLeod, B. P.et al.Self-driving laboratory for accelerated discovery of thin- film materials.Science Advances6, eaaz8867 (2020)

    2020

  10. [10]

    M., Donina, L., Dragone, V., Long, D.-L

    Granda, J. M., Donina, L., Dragone, V., Long, D.-L. & Cronin, L. Controlling an organic synthesis robot with machine learning to search for new reactivity. Nature559, 377–381 (2018)

    2018

  11. [11]

    & Crockett, M

    Messeri, L. & Crockett, M. J. Artificial intelligence and illusions of understanding in scientific research.Nature627, 49–58 (2024)

    2024

  12. [12]

    Eiben, A. E. & Smith, J. From evolutionary computation to the evolution of things.Nature521, 476–482 (2015)

    2015

  13. [13]

    Fogel, D. B. An introduction to simulated evolutionary optimization.IEEE transactions on neural networks5, 3–14 (1994)

    1994

  14. [14]

    & Sutskever, I

    Salimans, T., Ho, J., Chen, X., Sidor, S. & Sutskever, I. Evolution strategies as a scalable alternative to reinforcement learning.arXiv preprint arXiv:1703.03864 (2017)

    Pith/arXiv arXiv 2017

  15. [15]

    & Miettinen, K

    Jin, Y., Wang, H., Chugh, T., Guo, D. & Miettinen, K. Data-driven evolutionary optimization: An overview and case studies.IEEE Transactions on Evolutionary Computation23, 442–458 (2018). 13

    2018

  16. [16]

    A critical problem in benchmarking and analysis of evolutionary computation methods.Nature Machine Intelligence4, 1238–1245 (2022)

    Kudela, J. A critical problem in benchmarking and analysis of evolutionary computation methods.Nature Machine Intelligence4, 1238–1245 (2022)

    2022

  17. [17]

    H.Adaptation in natural and artificial systems: an introductory analysis with applications to biology, control, and artificial intelligence(MIT press, 1992)

    Holland, J. H.Adaptation in natural and artificial systems: an introductory analysis with applications to biology, control, and artificial intelligence(MIT press, 1992)

    1992

  18. [18]

    B., Goldberg, D

    Booker, L. B., Goldberg, D. E. & Holland, J. H. Classifier systems and genetic algorithms.Artificial intelligence40, 235–282 (1989)

    1989

  19. [19]

    Koza, J. R. Genetic programming as a means for programming computers by natural selection.Statistics and computing4, 87–112 (1994)

    1994

  20. [20]

    & Schwefel, H.-P

    Beyer, H.-G. & Schwefel, H.-P. Evolution strategies–a comprehensive introduc- tion.Natural computing1, 3–52 (2002)

    2002

  21. [21]

    & Ostermeier, A

    Hansen, N. & Ostermeier, A. Completely derandomized self-adaptation in evolution strategies.Evolutionary computation9, 159–195 (2001)

    2001

  22. [22]

    Wierstra, D.et al.Natural evolution strategies.The Journal of Machine Learning Research15, 949–980 (2014)

    2014

  23. [23]

    & Stanley, K

    Lehman, J. & Stanley, K. O. Abandoning objectives: Evolution through the search for novelty alone.Evolutionary computation19, 189–223 (2011)

    2011

  24. [24]

    & Clune, J

    Mouret, J.-B. & Clune, J. Illuminating search spaces by mapping elites.arXiv preprint arXiv:1504.04909(2015)

    Pith/arXiv arXiv 2015

  25. [25]

    K., Soros, L

    Pugh, J. K., Soros, L. B. & Stanley, K. O. Quality diversity: A new frontier for evolutionary computation.Frontiers in Robotics and AI3, 40 (2016)

    2016

  26. [26]

    Lehman, J., Meyerson, E., El-Gaaly, T., Stanley, K. O. & Ziyaee, T. Evo- lution and the knightian blindspot of machine learning.arXiv preprint arXiv:2501.13075(2025)

    Pith/arXiv arXiv 2025

  27. [27]

    & Mouret, J.-B

    Cully, A., Clune, J., Tarapore, D. & Mouret, J.-B. Robots that can adapt like animals.Nature521, 503–507 (2015)

    2015

  28. [28]

    C., Feng, L

    Tan, K. C., Feng, L. & Jiang, M. Evolutionary transfer optimization-a new frontier in evolutionary computation research.IEEE Computational Intelligence Magazine16, 22–33 (2021)

    2021

  29. [29]

    Wang, C.et al.When large language models meet evolutionary algorithms: Potential enhancements and challenges.Research8, 0646 (2025)

    2025

  30. [30]

    Gao, H.-a.et al.A survey of self-evolving agents: What, when, how, and where to evolve on the path to artificial super intelligence.arXiv preprint arXiv:2507.21046(2025). 14

    Pith/arXiv arXiv 2025

  31. [31]

    Frazier, P. I. A tutorial on bayesian optimization.arXiv preprint arXiv:1807.02811(2018)

    Pith/arXiv arXiv 2018

  32. [32]

    Active learning literature survey

    Settles, B. Active learning literature survey. Tech. Rep., University of Wisconsin–Madison (2009)

    2009

  33. [33]

    Romera-Paredes, B.et al.Mathematical discoveries from program search with large language models.Nature625, 468–475 (2024)

    2024

  34. [34]

    Novikov, A.et al.Alphaevolve: A coding agent for scientific and algorithmic discovery.arXiv preprint arXiv:2506.13131(2025)

    Pith/arXiv arXiv 2025

  35. [35]

    & Rockt¨ aschel, T

    Fernando, C., Banarse, D., Michalewski, H., Osindero, S. & Rockt¨ aschel, T. Promptbreeder: Self-referential self-improvement via prompt evolution.arXiv preprint arXiv:2309.16797(2023)

    Pith/arXiv arXiv 2023

  36. [36]

    A.et al.Gepa: Reflective prompt evolution can outperform reinforcement learning.arXiv preprint arXiv:2507.19457(2025)

    Agrawal, L. A.et al.Gepa: Reflective prompt evolution can outperform reinforcement learning.arXiv preprint arXiv:2507.19457(2025)

    Pith/arXiv arXiv 2025

  37. [37]

    Nature648, 312–319 (2025)

    Oh, J.et al.Discovering state-of-the-art reinforcement learning algorithms. Nature648, 312–319 (2025)

    2025

  38. [38]

    Zhang, B., Wang, C. & Wu, K. Evom: Agentic meta-evolution of actor- critic architectures for reinforcement learning.arXiv preprint arXiv:2606.26327 (2026)

    Pith/arXiv arXiv 2026

  39. [39]

    Xia, P.et al.Skillrl: Evolving agents via recursive skill-augmented reinforcement learning.arXiv preprint arXiv:2602.08234(2026)

    Pith/arXiv arXiv 2026

  40. [40]

    Burger, B.et al.A mobile robotic chemist.Nature583, 237–241 (2020)

    2020

  41. [41]

    Xi, Z.et al.Agentgym: Evolving large language model-based agents across diverse environments.arXiv preprint arXiv:2406.04151(2024)

    Pith/arXiv arXiv 2024

  42. [42]

    & Clune, J

    Zhang, J., Hu, S., Lu, C., Lange, R. & Clune, J. Darwin godel machine: Open-ended evolution of self-improving agents.arXiv preprint arXiv:2505.22954 (2025)

    Pith/arXiv arXiv 2025

  43. [43]

    E.et al.A multi-agent system for automating scientific discovery

    Ghareeb, A. E.et al.A multi-agent system for automating scientific discovery. Nature1–3 (2026)

    2026

  44. [44]

    Gottweis, J.et al.Accelerating scientific discovery with co-scientist.Nature1–3 (2026)

    2026

  45. [45]

    Wei, J.et al.From ai for science to agentic science: A survey on autonomous scientific discovery.arXiv preprint arXiv:2508.14111(2025). 15

    arXiv 2025

  46. [46]

    Ren, S.et al.Towards scientific intelligence: A survey of llm-based scientific agents.arXiv preprint arXiv:2503.24047(2025)

    arXiv 2025

  47. [47]

    Chen, Q.et al.Ai4research: A survey of artificial intelligence for scientific research.arXiv preprint arXiv:2507.01903(2025)

    Pith/arXiv arXiv 2025

  48. [48]

    Chen, Z.et al.Scienceagentbench: Toward rigorous assessment of language agents for data-driven scientific discovery.arXiv preprint arXiv:2410.05080 (2024)

    Pith/arXiv arXiv 2024

  49. [49]

    & Murai, F

    Assump¸ c˜ ao, H., Ferreira, D., Campos, L. & Murai, F. Codeevolve: An open source evolutionary coding agent for algorithm discovery and optimization. arXiv preprint arXiv:2510.14150(2025)

    Pith/arXiv arXiv 2025

  50. [50]

    Akiba, T., Shing, M., Tang, Y., Sun, Q. & Ha, D. Evolutionary optimization of model merging recipes.Nature Machine Intelligence7, 195–204 (2025)

    2025

  51. [51]

    Xia, P.et al.Agent0: Unleashing self-evolving agents from zero data via tool- integrated reasoning.arXiv preprint arXiv:2511.16043(2025)

    arXiv 2025

  52. [52]

    Lu, C.et al.Towards end-to-end automation of ai research.Nature651, 914–919 (2026)

    2026

  53. [53]

    & Forrest, S

    Miikkulainen, R. & Forrest, S. A biological perspective on evolutionary computation.Nature Machine Intelligence3, 9–15 (2021)

    2021

  54. [54]

    Sarkar, B.et al.Evolution strategies at the hyperscale.arXiv preprint arXiv:2511.16652(2025)

    arXiv 2025

  55. [55]

    & Bielza, C

    Larra˜ naga, P. & Bielza, C. Estimation of distribution algorithms in machine learning: A survey.IEEE Transactions on Evolutionary Computation28, 1301– 1321 (2024)

    2024

  56. [56]

    The cross-entropy method for combinatorial and continuous optimization.Methodology and computing in applied probability1, 127–190 (1999)

    Rubinstein, R. The cross-entropy method for combinatorial and continuous optimization.Methodology and computing in applied probability1, 127–190 (1999)

    1999

  57. [57]

    The application of bayesian methods for seeking the extremum

    Mockus, J. The application of bayesian methods for seeking the extremum. Towards global optimization2, 117 (1998)

    1998

  58. [58]

    O., Clune, J., Lehman, J

    Stanley, K. O., Clune, J., Lehman, J. & Miikkulainen, R. Designing neural networks through neuroevolution.Nature Machine Intelligence1, 24–35 (2019)

    2019

  59. [59]

    Jaderberg, M.et al.Population based training of neural networks.arXiv preprint arXiv:1711.09846(2017)

    Pith/arXiv arXiv 2017

  60. [60]

    Wang, G.et al.Voyager: An open-ended embodied agent with large language models.arXiv preprint arXiv:2305.16291(2023). 16

    Pith/arXiv arXiv 2023

  61. [61]

    Chu, Y.et al.Programmable rna translation through deep learning-driven ires discovery and de novo generation.Nature Machine Intelligence8, 559–574 (2026)

    2026

  62. [62]

    B¨ ack, T.Evolutionary algorithms in theory and practice: evolution strategies, evolutionary programming, genetic algorithms(Oxford university press, 1996)

    1996

  63. [63]

    & Zumer, V

    Brest, J., Greiner, S., Boskovic, B., Mernik, M. & Zumer, V. Self-adapting control parameters in differential evolution: A comparative study on numerical benchmark problems.IEEE transactions on evolutionary computation10, 646– 657 (2006)

    2006

  64. [64]

    & Feng, L

    Gupta, A., Ong, Y.-S. & Feng, L. Multifactorial evolution: Toward evolutionary multitasking.IEEE Transactions on Evolutionary Computation20, 343–357 (2015)

    2015

  65. [65]

    Wang, C., Liu, J., Wu, K. & Wu, Z. Solving multitask optimization problems with adaptive knowledge transfer via anomaly detection.IEEE Transactions on Evolutionary Computation26, 304–318 (2021)

    2021

  66. [66]

    Pan, S. J. & Yang, Q. A survey on transfer learning.IEEE Transactions on knowledge and data engineering22, 1345–1359 (2009)

    2009

  67. [67]

    & Levine, S

    Finn, C., Abbeel, P. & Levine, S. Model-agnostic meta-learning for fast adaptation of deep networks.arXiv preprint arXiv:1703.03400(2017)

    Pith/arXiv arXiv 2017

  68. [68]

    & Lipson, H

    Yosinski, J., Clune, J., Bengio, Y. & Lipson, H. How transferable are features in deep neural networks?Advances in neural information processing systems27 (2014)

    2014

  69. [69]

    Ayg¨ un, E.et al.An ai system to help scientists write expert-level empirical software.Nature1–3 (2026)

    2026

  70. [70]

    G¨ odel machines: self-referential universal problem solvers making provably optimal self-improvements.arXiv preprint cs/0309048(2003)

    Schmidhuber, J. G¨ odel machines: self-referential universal problem solvers making provably optimal self-improvements.arXiv preprint cs/0309048(2003)

    Pith/arXiv arXiv 2003

  71. [71]

    Stanley, K. O. & Miikkulainen, R. Evolving neural networks through augmenting topologies.Evolutionary computation10, 99–127 (2002)

    2002

  72. [72]

    L.et al.De novo design of protein structure and function with rfdiffusion.Nature620, 1089–1100 (2023)

    Watson, J. L.et al.De novo design of protein structure and function with rfdiffusion.Nature620, 1089–1100 (2023)

    2023

  73. [73]

    Dauparas, J.et al.Robust deep learning–based protein sequence design using proteinmpnn.Science378, 49–56 (2022)

    2022

  74. [74]

    S., Riley, P

    Gilmer, J., Schoenholz, S. S., Riley, P. F., Vinyals, O. & Dahl, G. E. Neural message passing for quantum chemistry.International Conference on Machine Learning1263–1272 (2017). 17

    2017

  75. [75]

    T., Sauceda, H

    Sch¨ utt, K. T., Sauceda, H. E., Kindermans, P.-J., Tkatchenko, A. & M¨ uller, K.- R. Schnet–a deep learning architecture for molecules and materials.The Journal of chemical physics148(2018)

    2018

  76. [76]

    P., Simm, G., Ortner, C

    Batatia, I., Kovacs, D. P., Simm, G., Ortner, C. & Cs´ anyi, G. Mace: Higher order equivariant message passing neural networks for fast and accurate force fields. Advances in neural information processing systems35, 11423–11436 (2022)

    2022

  77. [77]

    Yao, S.et al.Tree of thoughts: Deliberate problem solving with large language models.Advances in neural information processing systems36, 11809–11822 (2023)

    2023

  78. [78]

    Advances in neural information processing systems36, 68539–68551 (2023)

    Schick, T.et al.Toolformer: Language models can teach themselves to use tools. Advances in neural information processing systems36, 68539–68551 (2023)

    2023

  79. [79]

    Madaan, A.et al.Self-refine: Iterative refinement with self-feedback.Advances in neural information processing systems36, 46534–46594 (2023)

    2023

  80. [80]

    & Meyarivan, T

    Deb, K., Pratap, A., Agarwal, S. & Meyarivan, T. A fast and elitist multiobjec- tive genetic algorithm: Nsga-ii.IEEE transactions on evolutionary computation 6, 182–197 (2002)

    2002

Showing first 80 references.