Experimental and Computational Studies of Concept Learning

点击翻译按钮获取中文摘要

DESCRIPTION (provided by applicant): This research is aimed at developing better understanding of how people bring their prior knowledge to the table when learning about new concepts. Both experimental studies and computational models of these processes will be used to further understanding of this fundamental aspect of human cognition. The proposal focuses on effects and interactions that show that memorized exemplars of a problem are involved with concept learning, on processes involved in unsupervised sorting without feedback, and on how these two processes interact with pre-existing concepts and relational knowledge. New computational models will incorporate exemplars and unsupervised learning into an existing model of knowledge and supervised learning, accounting for a variety of previously observed and newly predicted effects. Experiments involving human participants will investigate interactions of prior knowledge with frequency, exposure, and concept structure. Experiments are paired with the modeling so that new empirical discoveries will go hand-in-hand with theoretical development. If successful, this model will be the only one in the field that accounts for this range of phenomena, encompassing both statistical learning and use of prior knowledge in concept acquisition. Relevance to Public Health: Categorization and category learning are fundamental aspects of cognition, allowing people to intelligently respond to the world. As categorization can be impaired by neurological disorders such as Parkinson's disease, dementia, and amnesia, a rigorous understanding of the processes involved in normal populations aides the research and treatment of disorders in patients. This project will provide a detailed computational model of concept learning, which can then serve as a model to investigate what has gone wrong when the process is disrupted in clinical populations.