Neural networks update

Abstract

Neural networks have been intensively studied as a discipline in their own right in the last five years. Initial claims were extremely ambitious - by employing the brain's computing principles, networks would eliminate programming, revolutionize computer architecture and sensor interfacing, make analog VLSI a reality, and guide us to a new understanding of human cognition. Obviously, this was an exaggeration, but some real progress has been made. I'll describe work in two areas - statistical methods to deal with classification, prediction, and control in data-rich, intuition-poor problems, and VLSI solutions, both in digital and analog styles, to accommodate these architectures.