Project Details When I was looking at the math for how neural networks worked, I got a little skeptical about the math used in them. Since the learning process used nothing but basic calculus principles, in theory there was nothing stopping me from using any function I wanted– and in practice, it could maybe even be more efficient in some situations. The finished code is essentially a general-purpose framework that I could use to put in any math function and get a neural network that uses it.

Project Details In this project, I created an algorithm to play weighted rock-paper-scissors. “Weighted” here means that depending on which choice you win with, you get different rewards. For example with weights [3, 2, 1], you get 3 points for winning with rock, 2 points for winning with paper, and 1 point for winning with scissors. I wanted the algorithm to converge toward optimal (least exploitable) play. In unweighted rock-paper-scissors, this looks like a 1/3 chance to play each choice.