The reinforcement metalearner as a biologically plausible meta-learning framework.

Autor:	Vriens T; Institute of Cognitive Sciences and Technologies, CNR, Rome, Italy Tim.Vriens@unicampus.it, massimo.silvetti@istc.cnr.ithttps://ctnlab.it/index.php/massimo-silvetti/, https://www.istc.cnr.it/en/people/massimo-silvetti., Horan M; Sainsbury Wellcome Centre, University College London, London, UKmattias.horan.19@ucl.ac.uk., Gottlieb J; Department of Neuroscience, Columbia University, New York, NY, USAjg2141@columbia.edu, https://zuckermaninstitute.columbia.edu/jacqueline-gottlieb-phd.; Zuckerman Mind Brain Behavior Institute, Columbia University, New York, NY, USA., Silvetti M; Institute of Cognitive Sciences and Technologies, CNR, Rome, Italy Tim.Vriens@unicampus.it, massimo.silvetti@istc.cnr.ithttps://ctnlab.it/index.php/massimo-silvetti/, https://www.istc.cnr.it/en/people/massimo-silvetti.
Jazyk:	angličtina
Zdroj:	The Behavioral and brain sciences [Behav Brain Sci] 2024 Sep 23; Vol. 47, pp. e168. Date of Electronic Publication: 2024 Sep 23.
DOI:	10.1017/S0140525X24000219
Abstrakt:	We argue that the type of meta-learning proposed by Binz et al. generates models with low interpretability and falsifiability that have limited usefulness for neuroscience research. An alternative approach to meta-learning based on hyperparameter optimization obviates these concerns and can generate empirically testable hypotheses of biological computations.
Databáze:	MEDLINE
Externí odkaz:	Zobrazit plný text záznamu