Characterizing the Relationship between Executive Function and Reinforcement Learning in Value Learning

Summer 2017

Nora Harhen : Cognitive Science

Donor: Pergo Fund
Mentor: Anne Collins

In everyday life, seldom are the choices we’ve made reinforced by objective reward like food or water. Rather, we tend to set goals for ourselves, and actions leading to those goals are what are reinforced, even in the absence of reward. Theoretical work has suggested that treating goal achievement as a pseudo-reward is an effective means to learn complex behavior, which may require going through many intermediary, value-neutral sub-goals before leading to reward. There has been indirect evidence for pseudo-rewards when reaching subgoals in EEG and fMRI, but as of yet, few studies have directly compared the reinforcing effects of goals vs. rewards on value learning. This project stems from our previous behavioral results which supported that pseudo-rewards can have reinforcing properties. To complement this, I seek to find EEG markers of value learning during trials in which feedback is internally-defined as rewarding rather than being objectively so. The results from this project could be important in characterizing the relationship between Executive Function (EF) and Reinforcement Learning (RL) as a cooperative one, in which EF influences RL to enable quick and efficient learning of complex behaviors.

I am incredibly grateful that the Pergo foundation enabled me to have a summer entirely dedicated to research. I felt like I grew a lot as a scientist as I troubleshooted problems with my protocol and read dozens of papers for novel ways to look at my data. It taught me how to adapt my way of thinking in response to the problem at hand. This experience only strengthened my conviction that I should go to grad school to pursue research in computational neuroscience.