Categorization is the process of assigning objects and events to separate classes or categories. It is a vitally important skill that makes it possible, for example, to respond differently to nutrients and poisons, and to predators and prey. Human category learning is incredibly diverse, as are the categories that people must learn, and there is growing evidence that different neural circuits might mediate category learning in different special circumstances. The research proposed here has two major goals. The first it to test more fully the hypothesis that human category learning is mediated by multiple systems, and in so doing, to explore the properties of the putative component systems and the conditions under which they may contribute to normal category learning. The second major goal is to develop a biologically plausible computational model of one important possible subsystem - namely, one in which people use an explicit rule-based reasoning process to learn new categories. The components of this model will be models of single cells that are joined in simple circuits that have been implicated in rule-based categorization. To calibrate the model and to establish its biological plausibility, the component models will be fit to relevant available single-cell recording data. After calibrating the components in this way, the overall model will be tested against human behavioral category learning data. Thus, the model that will be developed in this project represents a new generation of computational models in cognitive psychology - its architecture will be patterned after real neural circuits that are known to exist, its components will be models of single neurons whose behavior is consistent with the firing properties of real cells, and it will attempt to account for human category' learning data as well as the best existing (cognitive) models.
Showing the most recent 10 out of 30 publications