The more I use Deep Learning, the more I am amazed by it. Some things which would be hard to do programmatically are easy with the right Neural Network. It feels like we are just starting to scratch the possibilities.
Today I was at a Computation meets Data Science Conference, organised by Wolfram Research and the CQF. There were some interesting talks. The ones I enjoyed the most used Mathematica to analyse data in real time in interesting ways. It looks like Mathematica has good support for building neural networks now. I was impressed at how quickly Jon Macloone from Wolfram was able to get some quite useful neural network models up and running. Jon made the point that for some problems you are able to get results really quickly with neural nets, and others it’s really hard to get good results, and it’s not obvious which problems are which.
I’ve been playing more with Neural-Style Transfer. It’s fun to play with, but with the code I’m using, I’m struggling to get results I can use. I was trying to merge the style of a Platon photo with a photo taken of me (when I was growing a beard). It was weird what information the neural net decided to take from the style photo.
I just started experimenting with Image-Style Transfer. I’ve been excited about it for a long time, but reading this code on Nvidia’s latest paper prompted me to start playing with it in earnest. Of course, in the Coursera Deep Learning courses we studied this as well. As I don’t have an Nvidia card installed on my notebook, I started off with this Torch Implementation.