November Meetup: Win any game with these 3 weird tricks. Number 3 will shock you

Galois will be providing food and drinks at 6:30pm. Talk will start at 7:00pm.

Daniel Wagner will give a talk titled "Win any game with these 3 weird tricks. Number three will shock you".

In the world of machine learning, there's been a fair chunk of hubbub recently around the α0 algorithm. Pioneered by the folks at Deepmind, it is a triumph in the space of abstract game-playing AIs. One thing that makes it worth a hubbub is that it doesn't require any expert knowledge of the game you want it to play; just tell it the rules, and it figures out the strategy on its own. The algorithm uses Monte Carlo tree search and deep neural nets at its core. It works pretty well: Deepmind were the first to achieve superhuman go play, and they have now matched or improved on the state of the art in a handful of other games, including chess and backgammon.

For some people, machine learning techniques trigger an automatic disdain reflex. I myself have sometimes said "it's what you use when you don't know what to use". In this talk, I'd like to challenge that reaction, and convince you that Monte Carlo tree search is a grounded, principled algorithm; that it's the obviously correct choice; and that each engineering decision is backed by solid theory with real LaTeX proofs and everything.

Of course, to do that, I'll have to tell you what problem Monte Carlo tree search is solving, how it works, and how to turn it into an MLG pro gamer (batteries and salty trash talk not included). If there's a bit of time left after the tree search tutorial material, I'll also give a short walkthrough of my home-brew Haskell implementation and demo it doing a bang-up job of dropping pills and clearing viruses in Dr. Mario.