Re:Neurosynaptic chips
by CurtHagenlocher on Wednesday September 18, 2013 @10:38PM
(#6409451)
I've programmed the Connection Machine and I took a few classes from Carver Mead -- arguably one of the fathers of the neurosynaptic approach -- and these are nothing alike. The Connection Machine was simply a highly parallel SIMD computer that looks a lot like a GPU but with relatively better cross-processor communications support. It was resolutely digital in nature.
By contrast, neuromorphic chips are based on spike trains and are a much more analog form of computing.
Re:Neurosynaptic chips
by etrautmann on Thursday September 19, 2013 @12:18AM
(#6409760)
I worked in the lab that Paul and John came out of at Stanford, they do some incredible work. This is largely the output of the DARPA SYNAPSE program if I'm not mistaken.
By contrast, neuromorphic chips are based on spike trains and are a much more analog form of computing.
So it's like the hardware version of the ANN? But with more powerful neurons (they seem to do a lot more than just decide a yes and no)
PS: I apologize for not having read the paper yet. At work right now and will delve in to it later