Computational modeling of human sequential decision making in high-dimensional environments

Authors

Seoyoon Choi, Dartmouth CollegeFollow
Chong Wang, Dartmouth CollegeFollow
Alireza Soltani, Dartmouth CollegeFollow

Files

Student Class

2027

Student Affiliation

WISP Intern

First Advisor

Alireza Soltani

First Advisor Department

Department of Psychological and Brain Sciences

Description

Reinforcement learning is the process through which one learns the values of their actions and make adaptive decisions to maximize their reward. In naturalistic environments, this process can be challenging, as (1) choices can have long-term consequences and (2) choice options can have many features.

Previous studies showed that humans used successor representations to learn cognitive maps for planning. Other studies demonstrated that humans used feature-based learning and selective attention to learn the values of multi-featured stimuli. However, the interaction between these processes is largely unexplored.

To investigate this interaction, we designed a novel multidimensional two-step task, as well as six computational models representing different learning strategies. Using these models, we made concrete predictions about human behavior in our task paradigm, which can be tested in future experiments.

Publication Date

Spring 5-22-2024

Keywords

Reinforcement learning, computational modeling, decision-making, successor representation, feature-based learning, selective attention

Disciplines

Computational Neuroscience

Dartmouth Digital Commons Citation

Choi, Seoyoon; Wang, Chong; and Soltani, Alireza, "Computational modeling of human sequential decision making in high-dimensional environments" (2024). Wetterhahn Science Symposium Posters 2024. 13.
https://digitalcommons.dartmouth.edu/wetterhahn_2024/13