Modeling Delay Discounting using Gaussian Process with Active Learning
- Jorge Chang, Department of Psychology, The Ohio State University, Columbus, Ohio, United States
- Jiseob Kim, School of Computer Sicence and Engineering, Seoul National University, Seoul, Korea, Republic of
- Byoung-Tak Zhang, School of Computer Sicence and Engineering, Seoul National University, Seoul, Korea, Republic of
- Mark Pitt, Psychology, Ohio State University, Columbus, Ohio, United States
- Jay Myung, Department of Psychology, The Ohio State University, Columbus, Ohio, United States
AbstractWe explore a nonparametric approach to cognitive modeling. Traditionally, models in cognitive science have been parametric. As such, the model relies on the assumption that the data distribution can be defined by a finite set of parameters. However, there is no guarantee that such an assumption will hold, and it may introduce undesirable biases. For these reasons, a nonparametric approach to model building is appealing. We propose a novel framework that combines Gaussian Processes with active learning (GPAL), and evaluate it in the context of delay discounting (DD), a well-studied task in decision making. We evaluate GPAL in a simulation and a behavioral experiment, and compare it against a traditional parametric model. The results show that GPAL is a suitable modeling framework that is robust, reliable, and efficient, exhibiting high sensitivity to individual differences.