I'd love to see something similar to Torch7 for Julia. It's something that I (also given infinite time) was thinking about writing. Lua is close enough to Julia that I don't think it would be that hard to keep most of the design.
What makes Torch7 nice is that modules are easy to define and then all you have to worry about is finding the gradient for that module. I also found it easier to work with than Theano -- especially if you're working with changing structures (e.g. BPTS or BPTT). It might still be a good idea to use AD on a module basis. On Saturday, June 7, 2014 4:18:47 AM UTC-4, [email protected] wrote: > > > I am just wondering if there is code for multi layer perceptron anywhere > in julia with parallel processing and iRprop+ for gradient decent? >
