Viral, how far away are we from a deep learning framework in Julia ? It looks like there are many streams, from GSoC mentoring proposal for what seemed like a CuDNN layer, to a LLVM backend for CUDA/SPIR-V. I figure it'll take atleast 2 years for things to stabilize.
I assume you are not speaking about MXNet.jl and TensorFlow.jl, which are really good, and in some ways, even better than the Python wrappers for those libraries.
There are several pure Julia DL frameworks - Mocha.jl, KNet.jl and Merlin.jl. What will make things really exciting is native GPU codegen, which has recently landed in Julia, but will take some time to stabilize and make its way into these packages.
We do have a deep learning workshop and a GPU workshop in this year's JuliaCon covering the state of the art and the way forward.
