Temporal Credit Assignment and Strategic Behavior: Neural and Computational Perspectives on Reinforcement Learning in Primates

Date of Award

Fall 1-1-2025

Document Type

Dissertation

Degree Name

Doctor of Philosophy (PhD)

Department

Interdepartmental Neuroscience Program

First Advisor

Seo, Hyojung

Abstract

The credit assignment problem concerns how the reward is accredited to relevant past actions. In this study, we trained monkeys to play a variable delay task and recorded neural activity in prefrontal cortex and striatum to search for neural correlate involved in credit assignment. Behaviorally, long delay selectively reduced win - stay but not lose - switch tendencies, suggesting asymmetric effects of delay on positive versus negative outcomes. Neural recordings revealed that choice memory decayed during delays but reactivated at feedback, particularly in the dorsal striatum, where positive outcome signals were attenuated by long delays. These findings highlight a versatile, energy-efficient credit assignment mechanism based on reactivation of action memories.Strategic behavior in competitive games provides a window into the cognitive mechanisms of decision-making. To uncover the strategies underlying monkey's behavior in an iterated matching pennies (MP) game, we applied disentangled recurrent neural network (disRNN) to monkey behavioral data. Interpreting the trained models revealed a small set of cognitive components that jointly explain individual behavior, demonstrating that monkeys flexibly combine multiple strategies beyond model-free reinforcement learning. These results illustrate how deep learning–based model discovery can generate algorithmic-level hypotheses about the cognitive basis of strategic decision-making.

This document is currently not available here.

Share

COinS