Technical Program

Paper Detail

Paper: PS-2B.28
Session: Poster Session 2B
Location: Symphony/Overture
Session Time: Friday, September 7, 19:30 - 21:30
Presentation Time:Friday, September 7, 19:30 - 21:30
Presentation: Poster
Publication: 2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania
Paper Title: Computational mechanisms of human state-action-reward contingency learning under perceptual uncertainty
Manuscript:  Click here to view manuscript
Authors: Dirk Ostwald, Rasmus Bruckner, Hauke Heekeren, Freie Universit├Ąt Berlin, Germany
Abstract: To successfully interact with an everchanging world imbued with uncertainties, humans often have to learn probabilistic state-action-reward contingencies. Reinforcement learning algorithms have been able to provide a mechanistic picture of the neurocomputational principles that govern such learning and decision processes. However, standard reinforcement learning algorithms assume that the environmental state is fully observable. Humans, on the other hand, often have to learn the expected reward of choice options under considerable perceptual uncertainty. In this project we investigate the computational principles that govern probabilistic state-action-reward learning under perceptual uncertainty. To this end, we designed an integrated perceptual and economic decision making learning task and acquired behavioural data from 52 human participants. To interpret the participants' choice data, we developed a set of artificial agents which describe a range of cognitive-computational strategies. These strategies range from Bayes-optimal exploitative decision making that takes perceptual uncertainty parametrically into account to fully random choice policies. Our behavioural modelling initiative favoured an agent model that suggests that human participants integrate their subjective perceptual uncertainty when learning probabilistic state-action-reward contingencies. They tend, however, to underestimate the degree they should do so from a normative Bayes-optimal perspective.