This was initially a game I worked on sophomore year of college as a side project. Called Fuzzy Bear, you controlled a skull and had to dodge many explosive teddy bears that flew in from around the borders. I hadn’t really programmed much before, so it was a nice introduction to game logic and basic programming flow.
Later on, I revisited the project in an attempt to add some neural networks to teach the skull to dodge on its own accord. It was my first time working with neural nets, and took me some time to understand. In the end, the skull was able to learn and maneuvered back and forth to dodge the bears. The system was not perfect though as the skull tended to get stuck on the boundaries.
In the demo, every time the skull dies, it attempts to learn from its mistake to do better next time around. One thing to note is that the training data is lost once the browser is refreshed or the page changes.
Below are some links.