Technical Program

Paper Detail

Paper: PS-1B.24
Session: Poster Session 1B
Location: H Fl├Ąche 1.OG
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: Learning what is relevant for rewards via serial hypothesis testing
Manuscript:  Click here to view manuscript
License: Creative Commons License
This work is licensed under a Creative Commons Attribution 3.0 Unported License.
Authors: Mingyu Song, Ming Bo Cai, Yael Niv, Princeton University, United States
Abstract: Living in a world where any object bears features in many dimensions, it is crucial but also challenging for humans to figure out what dimensions are relevant for rewards. How do humans learn from trial and error to obtain rewards when multiple (or an unknown number of) dimensions need to be taken into account, and feedback is probabilistic? In this work, we designed a paradigm tailored to study such complex but naturalistic scenarios. In the experiment, participants configured three-dimensional stimuli by selecting features for each dimension and received probabilistic feedbacks. Participants demonstrated learning, selecting more rewarding features over the course of a game. To investigate their learning process, we compared three classes of learning models: a Bayesian model, reinforcement learning models and serial hypothesis testing models, and found evidence supporting the latter. This suggests that when facing complex learning scenarios with a great number of possible rules, people tend to actively test one hypothesis at a time, as opposed to evaluating all the possibilities or learning values of all features incrementally.