Cornell Cognitive Studies Symposium
Statistical Learning across Cognition
Creating Perceptual Representations that Recreate the World
Perceptual learning involves changes to an organism's perceptual system that improve its ability to respond to its environment and are caused by this environment. Given this influence of environmental patterns, perceptual learning depends upon statistical learning, defined here as the internal representation of external patterns of covariation among stimulus components. Two important mechanisms of perceptual learning are unitization and differentiation. Both mechanisms are influenced by the statistical structure of input patterns and also by how these patterns are used and categorized. By unitization, a single perceptual chunk is created for a complex assembly of stimulus components that reliably co-occur, particularly if the assembly is diagnostic for a useful categorization. By differentiation, perceptual dimensions that were originally fused become split apart if they vary independently within a set of stimuli, particularly if the dimensions are differentially diagnostic with respect to a useful categorization. These two mechanisms may seem to be in opposition to one another - one building large perceptual units out of smaller units, and the other breaking apart larger units into smaller units. However, a neural network model that develops perceptual detectors as part of concept learning will be described that reconciles these two mechanisms, showing that both can be explained in terms of creating appropriate units for necessary tasks.
Back to main page