Dnn ops - Githubissues

This is a dirty and buggy snapshot of the progress on integrating all the needed ops for DNN.

Also contains fixes for some of the bugs :)

CUDA version of DNN ops mostly implemented (some backward passes missing)
CUDA convolution keeps crashing :-/
Script level alternatives for the operations are there to replace the CUDA ops
Pooling script level operation is crashing :-/
Contains some implementations for shape/type inference but this is not in a solid state
In some locations of the scripts there needed to be a type indicator to make it work. So all these ops rely on atm, which might not be ideal.
Contains an example with a LeNet implementation for MNIST character classification (ported from SystemDS)

daphne-eu / daphne