Towards Evaluating Computational Models of Intuitive Decision Making with fMRI Data

摘要

A vast array of everyday tasks require individuals to use intuition to make decisions and act effectively, including civilian and military professional tasks such as those undertaken by firefighters, police, search and rescue, small unit leaders, and information analysts. To better understand and train intuitive decision making (IDM), we envision future training systems will represent IDM through computational models and use these models to guide IDM learning. This paper presents the first steps to the problem of validating computational models of IDM. To test if these models correlate with human performance, we examine methods to analyze functional magnetic resonance imaging (fMRI) data of human participants performing intuitive tasks. In particular, we examine the use of a new deep learning representation called sum-product networks to perform model-based fMRI analysis. Sum-product networks have been shown to be simpler, faster, and more effective than previous deep learning approaches, making them ideal candidates for this computationally demanding analysis.