Humans employ a wide array of strategies when learning to perform a task. Connectionist learning models, however, have typically focused on only a single learning strategy - on the slow induction of regularities from direct experience with a task domain. While these models succeed at explaining many phenomena, they fail to capture a rapid learning which results when explicit methods are used, such as hypothesis testing or relying upon verbal instruction. The research proposed here embarks on a journey towards a unified computation model of human learning - a model which will explain both healthy behavior and the use of compensatory strategies in the face of difficulty or disability. Specifically, a standard connectionist implicit learning mechanism will be augmented with an attractor network model of frontal working memory and a """"""""fast weight"""""""" model of medial temporal declarative memory, forming an integrated system capable of both inductive generalization and learning from explicit verbal instruction. This complex model will be tested against human behavior in a number of category learning domains.
