Bridging Levels: Using a Cognitive Model to Connect Brain and Behavior in Category Learning

Mental localization efforts tend to stress the where more than the what. We argue that the proper targets for localization are well-specified cognitive models. We make this case by relating an existing cognitive model of category learning to a learning circuit involving the hippocampus, perirhinal, and prefrontal cortex. Results from groups varying in function along this circuit (e.g., infants, amnesics, older adults) are successfully simulated by reducing the model’s ability to form new clusters in response to surprising events, such as an error in supervised learning or an unfamiliar stimulus in unsupervised learning. Reported task dissociations (e.g., categorization vs. recognition) are explained in terms of cluster recruitment demands. A major goal of cognitive psychology has been to develop an understanding of behavior in terms of computational principles. However, we are often left with the question of what these models tell us about the brain. The answer is certainly not clear. The growing area of cognitive neuroscience offers an endless source of new embers for this debate, as more and more cognitive function is localized and described in terms of specific brain processes. However, by focusing on the localization of mental function (i.e., where is processes X in the brain?), we run the risk of amassing a list of brain areas associated with certain tasks in the absence of useful linking theories reflecting how those regions interact to control behavior in our daily lives. In this paper, we argue that well-specified, process models of cognitive functions are the appropriate targets for localization. Successful process models offer a number advantages over folk psychological, ad hoc, or traditional psychological theories. For example, model developed in cognitive psychology make predictions, have mechanisms and dynamics which can be related to brain measures, and offer a simple and clear starting point for developing theories of brain function. To support our conjecture, we focus on relating a process model of human category learning to a learning circuit involving the hippocampus, perirhinal cortext, and prefrontal cortex (PFC). The model we consider, Supervised and Unsupervised STratified Adaptive Incremental Network (SUSTAIN), is applied to human data from a number of populations (infants, amnesics, and older adults) who differ in their category learning ability. Armed with its computational principles and the proposed mapping, SUSTAIN is able to predict how degraded function along this circuit affects category learning performance for these (and other) groups. In particular, SUSTAIN relates the degree of preserved function to how readily members of a group can individuate events, as opposed to collapsing experiences together into a common gestalt (see Figure 1). After introducing the model, we explain the close correspondence between aspects of the model and the currently understood function of a learning circuit involving PFC, the hippocampus, and perirhinal cortex. We then review a number of simulations which support our theory. In doing so, we provide a novel framework for understanding the role this circuit plays in category learning ability. In addition, our analysis suggests a recasting of several dichotomies popular in the field, such as the distinction between categorization and recognition, recollective and familiarity-driven responding, and episodic and semantic memory. SUSTAIN and the Proposed Mapping We begin by introducing the computational theory (SUSTAIN) and the bridge theory linking SUSTAIN to functional components in the brain. Due to limited space, readers interested in the mathematical details of the model are directed elsewhere (Love, Medin, & Gureckis, 2004).