Decomposing dynamical subprocesses for compositional generalization
Proceedings of the National Academy of Sciences · 12.11.2024 · 121 (46)
DOI: 10.1073/pnas.2408134121 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the National Academy of Sciences, online erschienen am 08.11.2024. Verfasst von Lennart Luettgau und 6 weiteren Autor:innen. Beteiligte Einrichtung: Imaging Neuroscience, Max Planck UCL Centre for Computational Psychiatry and Ageing Research, University College London, London WC1B 5EH, United King…. Hauptthemen laut MEDLINE: Learning, Cognition. Die Arbeit wurde bisher 13-mal zitiert (Crossref) und nennt 58 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Wellcome Trust.
Abstract (Originalsprache)
A striking feature of human cognition is an exceptional ability to rapidly adapt to novel situations. It is proposed this relies on abstracting and generalizing past experiences. While previous research has explored how humans detect and generalize single sequential processes, we have a limited understanding of how humans adapt to more naturalistic scenarios, for example, complex, multisubprocess environments. Here, we propose a candidate computational mechanism that posits compositional generalization of knowledge about subprocess dynamics. In two samples ( N = 238 and N = 137), we combined a novel sequence learning task and computational modeling to ask whether humans extract and generalize subprocesses compositionally to solve new problems. In prior learning, participants experienced sequences of compound images formed from two graphs’ product spaces (group 1: G1 and G2, group 2: G3 and G4). In transfer learning, both groups encountered compound images from the product of G1 and G3, composed entirely of new images. We show that subprocess knowledge transferred between task phases, such that in a new task environment each group had enhanced accuracy in predicting subprocess dynamics they had experienced during prior learning. Computational models utilizing predictive representations, based solely on the temporal contiguity of experienced task states, without an ability to transfer knowledge, failed to explain these data. Instead, behavior was consistent with a predictive representation model that maps task states between prior and transfer learning. These results help advance a mechanistic understanding of how humans discover and abstract subprocesses composing their experiences and compositionally reuse prior knowledge as a scaffolding for new experiences.
Schlagwörter
MeSH-Schlagwörter (MEDLINE)
Stichwörter der Autor:innen
Autor:innen und Institutionen
- Imaging Neuroscience, Max Planck UCL Centre for Computational Psychiatry and Ageing Research, University College London, London WC1B 5EH, United Kingdom.
- Imaging Neuroscience, Wellcome Centre for Human Neuroimaging, University College London, London WC1N 3AR, United Kingdom.
- Division of Psychiatry, Faculty of Brain Sciences, University College London, London W1T 7NF, United Kingdom.
- Clinical and Movement Neurosciences, Department of Motor Neuroscience, University College London, London WC1N 3BG, United Kingdom.
- Google DeepMind, London N1 C4AG, United Kingdom.
- Science and Engineering Department, School of Biological and Behavioural Sciences, Queen Mary University of London, London E1 4DQ, United Kingdom.
ORCID-Profile: Lennart Luettgau, Zeb Kurth-Nelson, Rani Moran, Raymond J Dolan
Förderung
- Wellcome Trust
Zitieren
APA
Luettgau, L., Erdmann, T., Veselic, S., Stachenfeld, K. L., Kurth-Nelson, Z., Moran, R., & Dolan, R. J. (2024). Decomposing dynamical subprocesses for compositional generalization. Proceedings of the National Academy of Sciences, 121(46). https://doi.org/10.1073/pnas.2408134121
Vancouver
Luettgau L, Erdmann T, Veselic S, Stachenfeld KL, Kurth-Nelson Z, Moran R, et al. Decomposing dynamical subprocesses for compositional generalization. Proceedings of the National Academy of Sciences. 2024;121(46). doi:10.1073/pnas.2408134121
BibTeX
@article{Luettgau2024,
author = {Luettgau, Lennart and Erdmann, Tore and Veselic, Sebastijan and Stachenfeld, Kimberly L and Kurth-Nelson, Zeb and Moran, Rani and Dolan, Raymond J},
title = {Decomposing dynamical subprocesses for compositional generalization},
journal = {Proceedings of the National Academy of Sciences},
year = {2024},
volume = {121},
number = {46},
doi = {10.1073/pnas.2408134121}
}
EUVIMED