Feb. 17, 11:45, Wei Ji Ma. Optimality and Probabilistic Computation in Visual Categorization

Patrick Garrigan

St. Joseph's University

Confirming efficient coding principles and behavioral methods to understand shape representation

At early stages of processing, the human visual system extracts and  
encodes information in a manner that most efficiently represents
visual scenes as they are projected onto the retina. That is, within
biophysical constraints, early visual representations faithfully
encode images in a compact format so that the image information can
be communicated to higher-level visual areas that are more
specialized. Higher-level visual representations should be efficient
as well, but they must also be designed with specific behaviors in
mind. These representations must consider not efficient
representation of information, but rather efficient representation of
the information that supports specific behaviors. One important
behavior is visual shape recognition. I will present a theoretical
framework for studying shape representation that considers both
efficient coding principles and specific behaviors. I will then use
this framework to demonstrate why some shapes are easier to learn to
recognize than other shapes of equivalent complexity.