Technical Program

Paper Detail

Paper: PS-1A.20
Session: Poster Session 1A
Location: H Lichthof
Session Time: Saturday, September 14, 16:30 - 19:30
Presentation Time:Saturday, September 14, 16:30 - 19:30
Presentation: Poster
Publication: 2019 Conference on Cognitive Computational Neuroscience, 13-16 September 2019, Berlin, Germany
Paper Title: Model abstraction for model-based reinforcement learning in the human orbitofrontal cortex
Manuscript:  Click here to view manuscript
License: Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 Unported License.
Authors: Yu Takagi, University of Oxford, United Kingdom; Wako Yoshida, Kyoto University, Japan; Saori Tanaka, Advanced Telecommunications Research Institute International, Japan
Abstract: How the human brain represents multiple models of the environment for decision-making (model-based reinforcement learning, MB-RL) is not well understood. We hypothesized that models are efficiently represented based on the similarity among them, to reduce redundancy, and this technique is called ‘model abstraction’ in the field of AI research. We designed a novel sequential learning task in which participants were required to simultaneously learn multiple models with a hidden latent structure, and studied corresponding brain activity using fMRI. By using an MVPA, we found that human OFC encodes the models reflecting the similarity among them. The degree of this ‘model abstraction’ ability was correlated with individual behavioral performance. Our results suggest that the human brains represent multiple models in a compact space, and this allows us to efficiently learn complex environments.