[svcrunch]

I generally don't waste time with patents. I think most patents in deep learning can be overturned by prior art.

My current approach to IP is trade secrets. If we publish, we are careful to avoid details that would make the techniques easy to productionize.