During the second half of our day we got to speak to Alex Cloud, a statistics Ph.D student who works for Riot Games. His research involves several topics including what’s called reinforcement learning. Reinforcement learning, or RL, is a type of machine learning that deals in autonomous agents interacting with the environment around them. They are trained to seek a specific goal and get “rewarded” for completing it, and “punished” for doing it incorrectly. Eventually the algorithm learns how to complete its goal in an effective way. The big question Alex has been trying to answer is “If I am able to simulate how the world works but don’t know the true state of the world, how do I plan ahead?” Several methods including RL, game theory, decision theory, and pure statistics have been used in his approach, and he recently submitted his findings to a computer science conference. By training a machine learning model to analyze previous scenarios in this “world” and checking for specific elements of the known world that predict the unknown parts, the entire world can be accurately predicted even if not all is known.
RL can go a long way towards solving many current challenges faced by the world including modeling conflict, autonomous vehicles, precision medicine, and business analytics. The example we were given was a machine that can reliably win at poker by accurately predicting its opponents’ hands. The possibilities are endless, and we’re excited to start experimenting with these technologies.