Poster E22, Monday, March 27, 2:30 – 4:30 pm, Pacific Concourse
On the Way to the Top: PINNACLE - A Theoretical Process-Model of Human Visual Category Learning
Ben Reuveni1, Paul J. Reber1; 1Northwestern University
We previously described a unique computational model of visual category learning that incorporates the roles of multiple memory systems to focus on how these systems interact during learning (Nomura, Maddox & Reber, 2007; Nomura & Reber, 2012). PINNACLE (Parallel Interactive Neural Networks Active in Category Learning) is a theoretically inspired model of category learning based on separate information processing streams for explicit rule discovery and implicit learning. PINNACLE produces behavior that matches human learning for both rule-based (RB) and information-integration (II) category learning. Trial-by-trial hypotheses about internal mental representations derived from these fits were used to guide fMRI analysis and identify regions of DLPFC active during mediation of competition between the two memory systems. To further explore the neural basis of cross-system strategy switching, we developed a new paradigm using stimuli order sequences designed to lure participants into initial use of an RB strategy followed by gradually revealing the structure to require an II strategy. Fitting PINNACLE to learning behavior with this paradigm revealed limitations in the original, simplified model. For example, over an extended set of trials performance showed periods of stable non-ceiling plateau performance that the simple, incremental learning mechanism within PINNACLE cannot fit. Here we report an update to the structure of PINNACLE that incorporates feedback based on reward-prediction error and uses perceptual noise to provide an improved account of category learning. We anticipate that this model will enable better predictions about strategy switching and the mental representation of category knowledge for use in future fMRI analyses.
Topic Area: THINKING: Decision making