Poker AI

The poker AI that I develop today use a family of algorithms based on CFR (counterfactual regret). The technique allows for rapidly finding approximate solutions to very large extensive-form games like poker. The most notable usage of CFR is Cepheus, an “essentially perfect” solution to heads-up limit hold’em constructed by the University of Alberta. For a human, it is impossible to defeat.

As a software engineer, I adapt CFR for solving different poker games, like heads-up hold’em, omaha, and other popular variants. Current work also includes formats with many players. While not an academic, I do enjoy running experiments and searching for new improvements. I’ve been lucky to get some feedback from other researchers in the field too.

Many of us participate in the most competitive computer poker tournament in the world: the AAAI Computer Poker Competition. I submitted an entry for the last two editions. Both were for the no limit hold’em instant runoff events, which measures the agent that is closest to an unbeatable Nash equilibrium. You can read more about the entries here:

Act1 (2016)
Prelude (2014)