Odelia Schwartz
Computational Neurobiology Lab, Salk Institute
Tuesday, February 28
1065 Kemper Hall
3 :10-4:00 p.m. - Refreshments to follow in 1131 Kemper
Neural Representation of Visual Information
Linear models have been a dominant paradigm in understanding the brain for the past half-century. In visual processing, the standard model assumes that the response of a neuron is determined by a single linear filter, for example tuned to a localized spatial region and orientation of the visual scene. However, it is becoming increasingly clear through experiment that visual neurons exhibit striking nonlinear behaviors. I suggest that these nonlinearities are not an accident of biological development, but arise because the system has evolved so as to optimally represent images in the world. I describe a particular model, in which the response of a neuron is determined by multiple linear filters whose outputs combine nonlinearly. I show that the model, with parameters optimized for an ensemble of photographic images, can account for some nonlinear response properties of typical neurons in vision. Related models have also been effective in the field of image processing. Finally, through experimental collaboration, we estimate the parameters in this class of model from data, thereby challenging the standard linear model of neural representation.