Date of Graduation

12-2019

Document Type

Dissertation

Degree Name

Doctor of Philosophy in Psychology (PhD)

Degree Level

Graduate

Department

Psychological Science

Advisor

Connie Lamm

Committee Member

Jennifer Veilleux

Second Committee Member

Douglas Rhoads

Keywords

feedback-related negativity, prediction error, reinforcement learning

Abstract

Organisms encode rewarding and aversive experiences through reinforcement learning, capitalizing on prediction errors (PEs), which adapt action strategies over time. Computational theories are explicit that PE signals should update action weights continuously over the course of a behavioral task, an important time-dependent variation that is eschewed in traditional neuroscience studies that average over large numbers of trials. I examined variation in reaction times and feedback-locked cortical activity over time as a function of PE to critically examine theories indicating that PE signals drive time-dependent learning. We recorded EEG while participants completed a novel reinforcement task that varied prediction error on a trial-by-trial basis. I applied a computational framework that modeled reaction time changes over the task as a function of prediction error and time. In positive reinforcement conditions, reaction times improved over the course of the task regardless of the PE. For negative reinforcement, learning effects were moderated by PE. For better than expected outcomes, more positive prediction errors (further from expectation) drove faster reaction times over the course of the task, and for worse than expected outcomes, more negative prediction errors (further from expectation) drove faster reaction times over the course of the task. Behavioral analyses were supplemented by single-trial robust regression of feedback-locked EEG. The feedback-related negativity (FRN), a mediofrontal ERP component thought to convey a PE signal, showed robust changes in activation over time but did not respond to trial-by-trial magnitude of prediction errors. This time-dependent change was evident only for reward delivery and aversive stimulus delivery, which represent on average the most salient outcomes in the task. Mediofrontal brain activity during this same time window and at the same scalp location drove subsequent reaction time improvements over the course of the task following aversive stimulus delivery. I suggest that the standard approach of examining the ERP as an average across conditions obscures important adaptation effects of the FRN that reflect reinforcement learning as outcomes are learned.

Share

COinS