As I'm sure some of you know, when I started my Ph.D. study at OSU, I started out in AI. But last spring quarter I switched my specialization to Computer Architecture. Not surprisingly, some of my recent research has involved using machine learning to help solve architecture problems. For instance, to optimize the voltages mentioned in the paper linked below, we need to learn individual chip characteristics on-line because we can't predict them in advance. To do the learning, we want to use neural nets. Because of all the experimentation and neural net training we needed to do, I needed fast neural net code, vectorized SSE3. I've started on a little project that some people might be interested in, but I thought I'd discuss it with people before I create a sourceforge project or whatever. (And BTW, I am aware of things like QuickNet and linear algebra libraries like BLAS.)
Basically, it's Ruby code that spits out C code with SSE intrinsics that implements the neural net you want. You can find the code for that here: http://www.cse.ohio-state.edu/~millerti/gen_sse.rb And here's my paper I presented at WEED (Workshop on Energy Efficient Design), which was part of ISCA (International Symposium on Computer Architecture). http://www.cse.ohio-state.edu/~millerti/WEED09-miller.pdf -- Timothy Normand Miller http://www.cse.ohio-state.edu/~millerti Open Graphics Project _______________________________________________ Open-graphics mailing list [email protected] http://lists.duskglow.com/mailman/listinfo/open-graphics List service provided by Duskglow Consulting, LLC (www.duskglow.com)
