Neither AI, Artificial Neural Networks, Fuzzy Logic, nor Bayesian Networks have led to powerful, scalable ISP. So we turn once more to biological systems for inspiration. Cortex is the ultimate cognitive processor. And although we are a long ways from understanding the details of how it does what it does, some of the basic computations are beginning to take shape. Nature has, so it appears, produced a general purpose computational device that is a fundamental component of higher level intelligence.

A number of groups are looking at hierarchical Bayesian structures as an approximation to cortex. I will briefly present a generic, cortical like model based on distributed data representation and Bayesian Belief Propagation that we are using as a rough approximation to cortex to drive our hardware explorations.

I then will summarize our work on the implementation of this model using a range of architectures from digital to mixed signal in both CMOS and the CMOL (Cmos / MOLecular) nanogrid technology proposed by Likharev.

Host: Garrett Kenyon, P-21