Here's the implementation that I found:

https://github.com/aleju/mario-ai

This hopefully won't take a ridiculous amount of time to train ("maybe" 10 hours according to the author). This one is a bit "cheating" as Dr. Abbott said (it kinda enforces a certain behavior - favor moving right) but I think it's interesting to study all the same, and it doesn't take days or months of training time to accomplish.