Schedule Nov 03, 2010
Physics-style Theory Can be Used to Discover a Universal Principle Governing the Design of Neural Network
Chuck Stevens, The Salk Institute

For evolution to work, neural circuits must have a scalable architecture -- that is, these circuits must be designed in a way that permits them to process more information by simply increasing the circuit size rather that redesigning it. I will describe one of the design principles that govern such scalable circuits in the vertebrate brain. I start with a consideration of the retina. Retinal ganglion cells (RGCs) sample information about the visual scene and encode this information for transmission to the brain. Because RGCs tile the retina, each RGC constitutes a "pixel". How large should this pixel be? Using a theoretical approach, I give an answer to this question. These same ideas can be extended to a give general design principle (supported by experimental evidence) that applies to circuits throughout the vertebrate brain.

Other video options

Author entry (protected)