ALCOVE: An exemplar-based connectionist model of category learning
John K. Kruschke · Psychology Press eBooks · 2020
This chapter describes a connectionist model of category learning called attention learning covering map (ALCOVE). It discusses variations, extensions, and limitations of ALCOVE. ALCOVE is a connectionist model of category learning that incorporates an exemplar-based representation with error-driven learning. ALCOVE is also closely related to standard back-propagation networks. Although ALCOVE is a feedforward network that learns by gradient descent on error, it is unlike standard back propagation in its architecture, its behavior, and its goals. ALCOVE is a feed-forward connectionist network with three layers of nodes. On presentation of a training exemplar to ALCOVE, the association strengths and dimensional attention strengths are changed by a small amount so that the error decreases. ALCOVE was applied to the six category types by using three input nodes, eight hidden nodes, and two output nodes. Several reasonable extensions of ALCOVE that might allow it to fit a wider array of category learning phenomena, without violating the motivating principles of the model, are possible.