Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

An Episodic Memory Model Can Account for Reinforcement Learning in Humans

Javier Alejandro Masís Obando¹, Younes Strittmatter², Jonathan D. Cohen¹; ¹Princeton University, ²Brown University

Presenter: Javier Alejandro Masís Obando

Multiple models have been proposed for how people solve reinforcement learning tasks, with the general conclusion that people may implement a mixture of "model-free" and "model-based" strategies. Here, we sought to investigate whether a process-based episodic memory model (called EGO) can account for human behavior in two inference tasks, the classic and revaluation 2-step tasks. The EGO model was able to mimic the behavior of humans and multiple RL models solely through its parametrization. While the parameters to mimic the same RL models across tasks differed, those to mimic human behavior were similar. These results suggest that rather than through a mixture of distinct processes, people may instead implement a single memory-based process to solve inference problems. The default parametrization of this process may reflect a global optimum for the average statistics of daily life, and may be adjusted through experience and control.

Topic Area: Memory, Learning & Knowledge Structures