COLLOQUIUM Department of Computer Science and Engineering University of South Carolina Computational Maps in the Neural Cortex James Bednar Department of Computer Science University of Texas Date: March 5, 2004 (Friday) Time: 3:30-4:30PM Place: Swearingen 1A03 (Faculty Lounge) Abstract For machine vision systems to truly approach human capabilities, it is crucial that we understand the specific computations and algorithms performed by biological visual systems. This talk will present results from a detailed, large-scale computational model that makes significant advances in understanding how biological visual systems develop and how they function. Simulations show how newborn processing for orientation, eye preference, and motion direction can be constructed using prenatal training patterns and learning algorithms, and how postnatal learning from natural scenes can ensure that the architecture is a good match to typical visual patterns. These results explain how newborns can have an orientation map at birth yet adapt to the visual environment, and they provide concrete and novel predictions about lateral connection patterns and visual illusions that can be tested in future experiments. They also suggest that generating training patterns artificially is an efficient way to develop a complex, adaptive device like a visual system. James Bednar is a postdoctoral researcher in the Department of Computer Science of the University of Texas at Austin. He completed his Ph.D. in Computer Science at UT in spring 2002, and also has an M.A. in Computer Science, a B.A. in Philosophy, and a B.S. in Electrical Engineering. His research focuses on computational modeling of cortical map development. Dr. Bednar is the lead author of the forthcoming Topographica modeling software package, under development through a Human Brain Project grant from the National Institutes of Health.