(Reinforcement?) Learning to forage optimally. (October 2017)

• Record Type: Journal Article
• Authors: Kolling, Nils; Akam, Thomas
• Is Part Of: Current opinion in neurobiology. Volume 46(2017)
• Page Start: 162
• Page End: 169
• Publication Date: 2017-10
• Subjects: Neurobiology -- Periodicals; 573.8
• DOI: 10.1016/j.conb.2017.08.008
• Languages: English
• ISSNs: 0959-4388

Abstract

Highlights: Decision neuroscience is increasingly studying ecologically inspired foraging tasks. Human behaviour in foraging tasks is not well explained by model-free reinforcement learning. Humans extrapolate trends in reward rate trajectories to guide foraging choices. Model-based average reward reinforcement learning may support patch foraging. Abstract : Foraging effectively is critical to the survival of all animals and this imperative is thought to have profoundly shaped brain evolution. Decisions made by foraging animals often approximate optimal strategies, but the learning and decision mechanisms generating these choices remain poorly understood. Recent work with laboratory foraging tasks in humans suggest their behaviour is poorly explained by model-free reinforcement learning, with simple heuristic strategies better describing behaviour in some tasks, and in others evidence of prospective prediction of the future state of the environment. We suggest that model-based average reward reinforcement learning may provide a common framework for understanding these apparently divergent foraging strategies.