Papers
Topics
Authors
Recent
Detailed Answer
Quick Answer
Concise responses based on abstracts only
Detailed Answer
Well-researched responses based on abstracts and relevant paper content.
Custom Instructions Pro
Preferences or requirements that you'd like Emergent Mind to consider when generating responses
Gemini 2.5 Flash
Gemini 2.5 Flash 84 tok/s
Gemini 2.5 Pro 48 tok/s Pro
GPT-5 Medium 21 tok/s Pro
GPT-5 High 28 tok/s Pro
GPT-4o 96 tok/s Pro
GPT OSS 120B 462 tok/s Pro
Kimi K2 189 tok/s Pro
2000 character limit reached

Human-like Category Learning by Injecting Ecological Priors from Large Language Models into Neural Networks (2402.01821v2)

Published 2 Feb 2024 in cs.LG and cs.AI

Abstract: Ecological rationality refers to the notion that humans are rational agents adapted to their environment. However, testing this theory remains challenging due to two reasons: the difficulty in defining what tasks are ecologically valid and building rational models for these tasks. In this work, we demonstrate that LLMs can generate cognitive tasks, specifically category learning tasks, that match the statistics of real-world tasks, thereby addressing the first challenge. We tackle the second challenge by deriving rational agents adapted to these tasks using the framework of meta-learning, leading to a class of models called ecologically rational meta-learned inference (ERMI). ERMI quantitatively explains human data better than seven other cognitive models in two different experiments. It additionally matches human behavior on a qualitative level: (1) it finds the same tasks difficult that humans find difficult, (2) it becomes more reliant on an exemplar-based strategy for assigning categories with learning, and (3) it generalizes to unseen stimuli in a human-like way. Furthermore, we show that ERMI's ecologically valid priors allow it to achieve state-of-the-art performance on the OpenML-CC18 classification benchmark.

Definition Search Book Streamline Icon: https://streamlinehq.com
References (75)
  1. Gpt-4 technical report. arXiv preprint arXiv:2303.08774, 2023.
  2. Building artificial neural circuits for domain-general cognition: a primer on brain-inspired systems-level architecture. arXiv preprint arXiv:2303.13651, 2023.
  3. Anderson, J. R. The adaptive nature of human categorization. Psychol. Rev., 98(3):409–429, July 1991a. ISSN 0033-295X, 1939-1471. doi: 10.1037/0033-295X.98.3.409.
  4. Anderson, J. R. Is human cognition adaptive? Behavioral and brain sciences, 14(3):471–485, 1991b.
  5. Anthropic, P. B. C. Claude 2. https://www.anthropic.com/index/claude-2, July 2023. Accessed: 2024-1-15.
  6. Human Category Learning. Annual Review of Psychology, 56(1):149–178, February 2005. ISSN 0066-4308, 1545-2085. doi: 10.1146/annurev.psych.56.091103.070217. URL https://www.annualreviews.org/doi/10.1146/annurev.psych.56.091103.070217.
  7. Varieties of perceptual independence. Psychological Review, 93(2):154–179, 1986. ISSN 1939-1471. doi: 10.1037/0033-295X.93.2.154. Place: US Publisher: American Psychological Association.
  8. Deficits in category learning in older adults: Rule-based versus clustering accounts. Psychol. Aging, 32(5):473–488, August 2017. ISSN 0882-7974, 1939-1498. doi: 10.1037/pag0000183.
  9. Constitutional ai: Harmlessness from ai feedback. arXiv preprint arXiv:2212.08073, 2022.
  10. Barker, R. G. Ecological psychology: Concepts and Methods for Studying the Environment of Human Behavior. Stanford, CA: Stanford University Press, 1968.
  11. Modeling human exploration through resource-rational reinforcement learning. Advances in Neural Information Processing Systems, 35:31755–31768, 2022b.
  12. Heuristics from bounded meta-learned inference. Psychological review, 2022a.
  13. Meta-learned models of cognition. arXiv preprint arXiv:2304.06729, 2023.
  14. Openml benchmarking suites. arXiv:1708.03731v2 [stat.ML], 2019.
  15. Language models are realistic tabular data generators. arXiv preprint arXiv:2210.06280, 2022.
  16. Cognitive model priors for predicting human decisions. In International conference on machine learning, pp.  5133–5141. PMLR, 2019.
  17. Exploration beyond bandits. The drive for knowledge: The science of human information seeking, pp.  147–168, 2021.
  18. Brehmer, B. Hypotheses about relations between scaled variables in the learning of probabilistic inference tasks. Organizational Behavior and Human Performance, 11(1):1–27, 1974.
  19. Brunswik, E. Representative design and probabilistic theory in a functional psychology. Psychological review, 62(3):193, 1955.
  20. Data distributional properties drive emergent in-context learning in transformers. Advances in Neural Information Processing Systems, 35:18878–18891, 2022.
  21. Xgboost: A scalable tree boosting system. In Proceedings of the 22nd acm sigkdd international conference on knowledge discovery and data mining, pp.  785–794, 2016.
  22. Meta-in-context learning in large language models. arXiv preprint arXiv:2305.12907, 2023.
  23. Support-vector networks. Machine learning, 20:273–297, 1995.
  24. Vba: a probabilistic treatment of nonlinear models for neurobiological and behavioural data. PLoS computational biology, 10(1):e1003441, 2014.
  25. The dynamics of exemplar and prototype representations depend on environmental statistics. In Proceedings of the Annual Meeting of the Cognitive Science Society, volume 43, 2021.
  26. Heuristic decision making. Annual review of psychology, 62:451–482, 2011.
  27. Bayesian models of cognition. Carnegie Mellon University, 2008.
  28. Optimal predictions in everyday cognition. Psychological science, 17(9):767–773, 2006.
  29. Bayes in the age of intelligent machines. arXiv preprint arXiv:2311.10206, 2023.
  30. Textbooks are all you need. arXiv preprint arXiv:2306.11644, 2023.
  31. Hammond, K. R. Ecological validity: Then and now, 1998.
  32. Learning to learn using gradient descent. In Artificial Neural Networks—ICANN 2001: International Conference Vienna, Austria, August 21–25, 2001 Proceedings 11, pp.  87–94. Springer, 2001.
  33. TabPFN: A transformer that solves small tabular classification problems in a second. In The Eleventh International Conference on Learning Representations, 2023. URL https://openreview.net/forum?id=cp5PvcI6w8_.
  34. Role of feedback, category size, and stimulus distortion on the acquisition and utilization of ill-defined categories. Journal of Experimental Psychology: Learning, Memory, and Cognition, 10(1):83, 1984.
  35. Zero-shot compositional reinforcement learning in humans. PsyArXiv preprint PsyArXiv:ymve5, 2023.
  36. A recurrent network model of planning explains hippocampal replay and human behavior. bioRxiv, pp.  2023–01, 2023.
  37. Are there representational shifts during category learning? Cogn. Psychol., 45(4):482–553, December 2002. ISSN 0010-0285. doi: 10.1016/s0010-0285(02)00505-4.
  38. Adam: A method for stochastic optimization. arXiv preprint arXiv:1412.6980, 2014.
  39. Meta-learning of structured task distributions in humans and machines. In International Conference on Learning Representations, 2020.
  40. Using natural language and program abstractions to instill human inductive biases in machines. Advances in Neural Information Processing Systems, 35:167–180, 2022.
  41. Human-like systematic generalization through a meta-learning neural network. Nature, pp.  1–7, 2023.
  42. A rational model of function learning. Psychonomic bulletin & review, 22(5):1193–1215, 2015.
  43. Comparing decision bound and exemplar models of categorization. Perception & Psychophysics, 53(1):49–70, January 1993. ISSN 0031-5117, 1532-5962. doi: 10.3758/BF03211715. URL http://link.springer.com/10.3758/BF03211715.
  44. When do neural nets outperform boosted trees on tabular data? arXiv preprint arXiv:2305.02997, 2023.
  45. Context theory of classification learning. Psychol. Rev., 85(3):207–238, May 1978. ISSN 0033-295X, 1939-1471. doi: 10.1037/0033-295x.85.3.207.
  46. Orca-2: Teaching small language models how to reason. arXiv, November 2023.
  47. Transformers can do bayesian inference. arXiv preprint arXiv:2112.10510, 2021.
  48. Transformers can do bayesian inference. In International Conference on Learning Representations, 2022. URL https://openreview.net/forum?id=KSugKcbNf9.
  49. Neisser, U. Cognition and reality. W H Freeman/Times Books/ Henry Holt and Co, 1987.
  50. Chapter six - Systems of Category Learning: Fact or Fantasy? In Ross, B. H. (ed.), Psychology of Learning and Motivation, volume 54 of Advances in Research and Theory, pp.  167–215. Academic Press, January 2011. doi: 10.1016/B978-0-12-385527-5.00006-1. URL https://www.sciencedirect.com/science/article/pii/B9780123855275000061.
  51. Nosofsky, R. M. Attention, similarity, and the identification–categorization relationship. Journal of experimental psychology: General, 115(1):39, 1986.
  52. Exemplar and prototype models revisited: Response strategies, selective attention, and stimulus generalization. Journal of Experimental Psychology: Learning, Memory, and Cognition, 28(5):924, August 2002. ISSN 1939-1285. doi: 10.1037/0278-7393.28.5.924. URL https://psycnet.apa.org/fulltext/2002-15432-008.pdf. Publisher: US: American Psychological Association.
  53. Comparing models of rule-based classification learning: a replication and extension of shepard, hovland, and jenkins (1961). Mem. Cognit., 22(3):352–369, May 1994a. ISSN 0090-502X. doi: 10.3758/bf03200862.
  54. Rule-plus-exception model of classification learning. Psychol. Rev., 101(1):53–79, January 1994b. ISSN 0033-295X. doi: 10.1037/0033-295x.101.1.53.
  55. Meta-learning of sequential strategies. arXiv preprint arXiv:1905.03030, 2019.
  56. Pytorch: An imperative style, high-performance deep learning library. In Advances in Neural Information Processing Systems 32, pp.  8024–8035. Curran Associates, Inc., 2019.
  57. Using large-scale experiments and machine learning to discover theories of human decision-making. Science, 372(6547):1209–1214, 2021.
  58. Learning to learn. Kluwer Academic Publishers, 1998.
  59. Bayesian model selection for group studies—revisited. Neuroimage, 84:971–985, 2014.
  60. Meta-learning with memory-augmented neural networks. In International conference on machine learning, pp.  1842–1850. PMLR, 2016.
  61. Generating datasets with pretrained language models. arXiv preprint arXiv:2104.07540, 2021.
  62. A rational analysis of the optimism bias using meta-reinforcement learning. In Conference on Cognitive Computational Neuroscience (CCN 2023), 2023.
  63. Probing the compositionality of intuitive functions. Advances in neural information processing systems, 29, 2016.
  64. Compositional inductive biases in function learning. Cognitive psychology, 99:44–79, 2017.
  65. Learning and memorization of classifications. Psychological Monographs: General and Applied, 75(13):1–42, 1961. ISSN 0096-9753. doi: 10.1037/h0093825.
  66. Simon, H. A. Invariants of human behavior. Annual review of psychology, 41(1):1–20, 1990.
  67. Prototypes in the mist: The early epochs of category learning. Journal of Experimental Psychology: Learning, memory, and cognition, 24(6):1411, 1998.
  68. Stanford alpaca: An instruction-following llama model. https://github.com/tatsu-lab/stanford_alpaca, 2023.
  69. Generalization, similarity, and bayesian inference. Behavioral and brain sciences, 24(4):629–640, 2001.
  70. Ecological rationality: Intelligence in the world. Oxford University Press, Cary, NC, February 2012. ISBN 9786613594013.
  71. Llama: Open and efficient foundation language models. arXiv preprint arXiv:2302.13971, 2023.
  72. Attention is all you need. Advances in neural information processing systems, 30, 2017.
  73. SciPy 1.0: Fundamental Algorithms for Scientific Computing in Python. Nature Methods, 17:261–272, 2020. doi: 10.1038/s41592-019-0686-2.
  74. Learning to reinforcement learn. arXiv preprint arXiv:1611.05763, 2016.
  75. Improving text embeddings with large language models. arXiv preprint arXiv:2401.00368, 2023.
Citations (1)
View on Semantic Scholar
List To Do Tasks Checklist Streamline Icon: https://streamlinehq.com

Collections

Sign up for free to add this paper to one or more collections.

User Circle Single Streamline Icon: https://streamlinehq.com Sign Up

Summary

We haven't generated a summary for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Summarize Now
Ai Generate Text Spark Streamline Icon: https://streamlinehq.com

Paper Prompts

Sign up for free to create and run prompts on this paper using GPT-5.

List Streamline Icon: https://streamlinehq.com Explore 10 Community Prompts
Dice Question Streamline Icon: https://streamlinehq.com

Follow-up Questions

We haven't generated follow-up questions for this paper yet.

Ai Sparkles Streamline Icon: https://streamlinehq.com Generate Now
X Twitter Logo Streamline Icon: https://streamlinehq.com

Tweets