So you’ve all seen this, right?
It seems that AI is everywhere these days, from self driving cars to thermostats. Unfortunately, while amazing progress is being made by some, my recent results look more like this:
It’s a chicken and an egg problem… To generate quality learning data, you need quality players. To train quality players, you first need quality training data.
Yes, I know, Google’s team managed to train a neural network to play Go from scratch. They also developed custom hardware just to process neural networks. My computer isn’t that fast (even with a 1060 from NVidia). It’ll take me longer.
My current plan is to dig in to the recent API updates for CNTK and redo my whole testing suite. Due to other projects it’s going to be a few weeks until I can get to that though. I’ll update this site again when I’ve got something to show for that.