EUVIMEDEUVIMED entdecken
Open Access · CC BY 4.0

Action subsampling supports policy compression in large action spaces

Shuze Liu iD, Samuel Joseph Gershman

PLOS Computational Biology · 05.09.2025 · 21 (9) · e1013444

DOI: 10.1371/journal.pcbi.1013444 ↗

Kurz gefasst

Fachartikel in der Zeitschrift PLOS Computational Biology, online erschienen am 05.09.2025. Verfasst von Shuze Liu und 1 weiteren Autor:innen. Beteiligte Einrichtung: PhD Program in Neuroscience, Harvard University, Cambridge, Massachusetts, United States of America. Hauptthemen laut MEDLINE: Decision Making, Cognition. Die Arbeit wurde bisher 2-mal zitiert (Crossref) und nennt 71 Literaturangaben. Der Volltext ist frei zugänglich (Open Access, CC BY 4.0). Gefördert von: Schmidt Sciences Polymath Program, Kempner Institute for the Study of Natural and Artificial Intelligence.

AbstractSchlagwörterBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Real-world decision-making often involves navigating large action spaces with state-dependent action values, taxing the limited cognitive resources at our disposal. While previous studies have explored cognitive constraints on generating action consideration sets or refining state-action mappings (policy complexity), their interplay remains underexplored. In this work, we present a resource-rational framework for policy compression that integrates both constraints, offering a unified perspective on decision-making under cognitive limitations. Through simulations, we characterize the suboptimality arising from reduced action consideration sets and reveal the complex interaction between policy complexity and action consideration set size in mitigating this suboptimality. We then use such normative insight to explain empirically observed phenomena in option generation, including the preferential sampling of generally valuable options and increased correlation in responses across contexts under cognitive load. We further validate the framework’s predictions through a contextual multi-armed bandit experiment, showing how humans flexibly adapt their action consideration sets and policy complexity to maintain near-optimality in a task-dependent manner. Our study demonstrates the importance of accounting for fine-grained resource constraints in understanding human cognition, and highlights the presence of adaptive metacognitive strategies even in simple tasks.

Schlagwörter

MeSH-Schlagwörter (MEDLINE)

Decision MakingCognitionHumansComputational BiologyComputer SimulationMaleFemaleAdultYoung Adult

Autor:innen und Institutionen

  • PhD Program in Neuroscience, Harvard University, Cambridge, Massachusetts, United States of America.
  • Department of Psychology and Center for Brain Science, Harvard University, Cambridge, Massachusetts, United States of America.
  • Center for Brains, Minds, and Machines, Massachusetts Institute of Technology, Cambridge, Massachusetts, United States of America.

ORCID-Profile: Shuze Liu

Förderung

  • Schmidt Sciences Polymath Program
  • Kempner Institute for the Study of Natural and Artificial Intelligence

Zitieren

APA

Liu, S., & Gershman, S. J. (2025). Action subsampling supports policy compression in large action spaces. PLOS Computational Biology, 21(9), e1013444. https://doi.org/10.1371/journal.pcbi.1013444

Vancouver

Liu S, Gershman SJ. Action subsampling supports policy compression in large action spaces. PLOS Computational Biology. 2025;21(9):e1013444. doi:10.1371/journal.pcbi.1013444

BibTeX
@article{Liu2025,
  author = {Liu, Shuze and Gershman, Samuel Joseph},
  title = {Action subsampling supports policy compression in large action spaces},
  journal = {PLOS Computational Biology},
  year = {2025},
  volume = {21},
  number = {9},
  pages = {e1013444},
  doi = {10.1371/journal.pcbi.1013444}
}
Kostenlos starten →