…ugh, sorry that title…
I spent this weekend training a network for an online MIT class in autonomous driving. In the traffic navigation challenge they ask the students to score at least a 65mph average. There is also a class on squeal general AI.
That base challenge wasn’t too tough. I found my way to 69mph by following the steps described. I made some common sense choices about the size of the learning parameters. I did some fine tuning, but hitting the leaderboard, would not be so easy. The top ten are pushing nearly 76mph.
I’ve heard a few deep learning lecturers talk about the process as having an art to it. I have been poking at this traffic contest and an image classification project. I think it’s something I have a knack for.
I maxed out at 73.96mph, not bad. I can’t tell if what I’m learning will translate to a real world problem, but super fun regardless.