Technical Program

Paper Detail

Paper: RS-2A.2
Session: Late Breaking Research 2A
Location: Late-Breaking Research Area
Session Time: Friday, September 7, 17:15 - 19:15
Presentation Time:Friday, September 7, 17:15 - 19:15
Presentation: Poster
Publication: 2018 Conference on Cognitive Computational Neuroscience, 5-8 September 2018, Philadelphia, Pennsylvania
Paper Title: Active sampling decreases the perceived volatility of uncertain environments and associated neural representations
Authors: Aurelien Weiss, Valentin Wyart, Valerian Chambon, Ecole Normale Superieure, France; Jan Drugowitsch, Harvard Medical School, United States
Abstract: Beside obvious differences in input, perceptual and reward-guided decisions diverge in the degree of control conferred to the decision-maker in the sampling of evidence. We thus asked whether humans learn differently from the same uncertain evidence when the evidence in question corresponds either to cues sampled passively as in perceptual decisions, or to outcomes sampled actively as in reward-guided decisions. We designed a probabilistic reversal learning paradigm with two conditions, in which subjects have to track a latent state of their environment: either the category (A or B) from which presented stimuli are drawn (cue-based condition), or the action (left or right) which draws stimuli from a target category (outcome-based condition). Computational modeling of human behavior showed a lower perceived rate of reversals (volatility) in the outcome-based condition. Multivariate analyses of magnetoencephalographic (MEG) signals revealed that stimuli were coded in relation to their assigned latent state more strongly in the outcome-based condition, an effect originating from the right medial temporal lobe (MTL). Together, these findings demonstrate that humans rely more strongly on their internal model of the environment when they learn about its current state through active sampling rather than by observation.