There are matrix multiplication +/ .* inside RNN. In j806, there are some optimization code taking advantage of AVX and FMA instruction, speed improvement should be significant, run benchmark to see how much. YMMV.
On Tue, Dec 5, 2017 at 12:50 PM, 'Skip Cave' via Programming <[email protected]> wrote: > I have been working on translating some machine learning algorithms into J > I don't know Python, so Raul helped me translate a Recurrent Neural Network > originally written in Python, into J. > > Here's a link to the start of our discussion thread on the J programming > forum. > A Neural Network in 11 lines of Python (Part 1) > <http://jsoftware.com/pipermail/programming/2017-November/049711.html> > > This algorithm is is the basis of the photo classification process. > > Skip > > > Skip Cave > Cave Consulting LLC > > On Mon, Dec 4, 2017 at 10:10 PM, Brian Schott <[email protected]> > wrote: > >> Devon, >> >> I recently noticed convolutional neural networks for image recognition and >> developed a very limited J version for my own purposes. I am not sure the >> result is ready for prime time, but it was really enlightening to see what >> could be done especially with J's array orientation. I got a lot from the >> following link. >> >> http://cs231n.github.io/convolutional-networks/ >> >> >> >> On Mon, Dec 4, 2017 at 10:37 PM, Devon McCormick <[email protected]> >> wrote: >> >> > I'm learning about machine-learning in general and both courses I've >> taken >> > are basically intros to Python's sci-kit which is good for a high-level >> > look but gives no clue to the underlying algos. >> > >> > I searched the J wiki for "quadratic" and found a mention from a 2005 >> > NYCJUG meeting where we were lamenting that other environments simply >> have >> > quadratic solvers available but J does not. Looking into it more, I was >> > unsure "quadratic" means in "quadratic solver"; from "Numerical >> Recipes", I >> > gather that it may refer to quadratic convergence. On further >> reflection, >> > I'm not exactly sure what this means either but suspect it means >> something >> > like O(sqrt(n)) solution time but am not sure. I'm also unclear if this >> > non-linear solver in J - http://code.jsoftware.com/wiki/Scripts/nlls - >> is >> > an example of a quadratic solver (with this follow-up page I put >> together - >> > http://code.jsoftware.com/wiki/NYCJUG/2010-11-09/ >> > Levenberg-MarquardtAlgorithm >> > ). >> > >> > Perhaps someone more acquainted with this domain can shed some light. >> > >> > My ultimate goal for these sort of tools is to do something with the >> > million or so photos I have but I'm seeing from what I've learned so far >> > that the real work is in feature-extraction, so that at least is >> something >> > where I can use J to good effect. >> > >> > I'm less interested in a wrapper such as the one Scott Locklin mentioned >> > and more interested in expressing the algorithm in an understandable, >> > compact form but may end up surrendering to the popularity contest and >> > simply using the Python library. >> > >> > >> ---------------------------------------------------------------------- >> For information about J forums see http://www.jsoftware.com/forums.htm >> > ---------------------------------------------------------------------- > For information about J forums see http://www.jsoftware.com/forums.htm ---------------------------------------------------------------------- For information about J forums see http://www.jsoftware.com/forums.htm
