Faculty • Affiliated Faculty • Research Associates • Graduate Students

 

James Bednar

Office:

Taylor 4.138

Email:

jbednar@cs.utexas.edu

Homepage:

cs.utexas.edu/users/jbednar

Faculty Advisor:

Risto Miikkulainen

Research Interests

My research concentrates on biologically realistic modeling of cognitive processes using massive artificial neural networks. The dramatic advances in computing technology over the past few decades are beginning to enable us to make equally dramatic advances in our understanding of the human mind. With the computing power now available and soon to be available, realistic simulations of cortical processing are becoming practical. This enables us to make and test detailed hypotheses about brain function. The overall goal is to make cognitive research into an empirical science, rather than the purely philosophical domain it has been for centuries.

Projects

Self Organization of the Visual Cortex Using the RF-LISSOM Model: RF-LISSOM (Sirosh and Miikkulainen, 1994) is a self-organizing computational model that attempts to replicate the detailed development of the primary visual cortex of humans. The model is based on the SOM model (Kohonen 1982) widely used for data visualization. LISSOM extends SOM to be more powerful and more biologically realistic by using only Hebbian learning and by including lateral connections between neurons. The fundamental thesis driving the RF-LISSOM work is that the cortex organizes itself to capture correlations found in visual inputs and in internally-generated sources of activation.

Tilt Aftereffects in RF-LISSOM: Assuming that the self organizing processes continue operating on the adult RF-LISSOm structure, tilt aftereffects in the adult can be modeled as a result of adapting lateral inhibition and re-normalization. Unlike previous models, the model can account for both the direct (small angle) and indirect (large angle) effects, and it shows the functional relevance of this behavior. This work shows that the same fundamental learning processes which drive the initial development of the cortex may also be operating in the adult over short time scales.

Self Organization Driven by Pattern Generators: Work with the RF-LISSOM model has shown that it can develop realistic cortical structures when presented with approximations of the visual environment. However, the brain already has significant structure at birth, so environmental inputs cannot account for all of this self-organization. This ongoing research project explores a surprisingly simple but very effective way that an organism's genome can specify detailed cortical structures, by generating training patterns internally. The end result is that genetic information is expressed through the same robust learning mechanisms which also incorporate information from the environment.

Neural Networks Research Group

Selected Publications

Developmental Psychology

Psychophysics

Computational Neuroscience