Friday, December 17, 2010

Computational neuroscience in 2010...and in 1956

The last paper issue of IEEE Spectrum featured an interesting article on memristor-based artificial neuroscience, specifically the MoNETA (Modular Neural Exploring Traveling Agent).  Sorry to tease, but the on-line version seems to be MIA and the subject is way to deep for me to summarize well in one paragraph.  But, I'll try. Although there are undoubtedly many contributors, the core of MoNETA is a 'brain-inspired' processor being developed at HP Labs and a software system from Boston University.  I guess MoNETA is technically just the software, but the two might as well be lumped together here.  If you remember back, the memristor is widely considered to be the 4th basic circuit element (along side resistors, capacitors, and inductors) and was discovered (invented?) by HP a few years ago.  Because it naturally remembers its current state, this component can readily be used to build simulated neurons. The HP neural chips will have hundreds of core processors each with a pile of ultra-dense memristor lattices. (While 'pile' isn't too technical a term, it is fairly descriptive.) Utilizing this artificial brain will requires new software approaches.  While HP is providing a special-purpose 'OS", Boston U. will provide the applications package. Their goal is to create "a general purpose mammalian-type intelligence."  This is fascinating and well above my old VonNeuman-limited brain.

That covers the '2010' from the title, and now for the '1956'.  It apears Time Magazine reported on the state of neural computing back then. :)

1956 ... missile brains!