Here’s my first 6 hours into learning neural networks. (click to play)
You might recognize this racetrack/car system from before when I did just genetic algorithms. The difference here is I don’t directly map inputs to outputs anymore. Cars have a 2-layer system (Input → Layer1 → Layer2 → Output) and use genetic algorithms for a survival of the fittest sort of system. By generation 30 or so they can begin doing full laps!
This allows for pattern recognition and hopefully I’ll be able to make them do cool stuff in the future like drift.
Here they are 10 minutes later: