Next talk: "Deep Learning for Poker: Inference From Patterns in an Adversarial Environment" by Nikolai Yakovenko, Dec 2nd
56 views
Skip to first unread message
Colin Raffel
Nov 24, 2015, 12:48:12 PM11/24/15
to cu-neu...@googlegroups.com
Hi all, slides and notes from Zhuo's talk last Wednesday are online. Our next talk will be on December 2nd; we're hosting Nikolai Yakovenko who has been working on learning-based systems for playing poker. Info follows, please forward to anyone you think would be interested. See you there!
Deep Learning for Poker: Inference From Patterns in an Adversarial Environment
Nikolai Yakovenko
4pm, December 2nd, 2015
CEPSR 750 (Interschool Lab)
With the seemingly effortless success of DeepMind (and others) at creating deep learning AIs that learn to master video games, it is not surprising that a similar deep learning system could learn to bet a poker hand, at the level of an experienced human player. However, unlike most video games, the poker AI has to learn a strategy that will not be easily exploitable by an adversarial agent, including an agent that perhaps might have access to 100,000 hands of that AI’s previous games. We quickly present our current system, which we demonstrated can learn different heads-up (one on one) limit poker games, going from zero knowledge to that of a competent human player. Then we would like to talk about adapting these techniques for the more complicated No-Limit Holdem game, in which a network must choose bet sizes, along with whether or not to bet, and must handle an opponent that can also bet against it, in more or less continuous amounts. We are preparing this system for the 2016 Annual Computer Poker Competition, where state of the art academic systems compete in heads-up No Limit Holdem. The winners have previously been based on solving for an approximate Nash equilibrium betting strategy for the poker game. Last year's winner (from Carnegie Mellon University) played well enough to earn a "statistical draw" against some of the best heads-up No Limit Holdem players in the world. Hopefully, our deep learning system can find holes in the equilibrium-seeking system’s game.
∿
Deep Learning for Poker: Inference From Patterns in an Adversarial Environment
Nikolai Yakovenko
4pm, December 2nd, 2015
CEPSR 750 (Interschool Lab)
With the seemingly effortless success of DeepMind (and others) at creating deep learning AIs that learn to master video games, it is not surprising that a similar deep learning system could learn to bet a poker hand, at the level of an experienced human player. However, unlike most video games, the poker AI has to learn a strategy that will not be easily exploitable by an adversarial agent, including an agent that perhaps might have access to 100,000 hands of that AI’s previous games. We quickly present our current system, which we demonstrated can learn different heads-up (one on one) limit poker games, going from zero knowledge to that of a competent human player. Then we would like to talk about adapting these techniques for the more complicated No-Limit Holdem game, in which a network must choose bet sizes, along with whether or not to bet, and must handle an opponent that can also bet against it, in more or less continuous amounts. We are preparing this system for the 2016 Annual Computer Poker Competition, where state of the art academic systems compete in heads-up No Limit Holdem. The winners have previously been based on solving for an approximate Nash equilibrium betting strategy for the poker game. Last year's winner (from Carnegie Mellon University) played well enough to earn a "statistical draw" against some of the best heads-up No Limit Holdem players in the world. Hopefully, our deep learning system can find holes in the equilibrium-seeking system’s game.
∿
Colin Raffel
Dec 1, 2015, 11:17:19 AM12/1/15
to cu-neu...@googlegroups.com
Hi all, friendly reminder that this talk is tomorrow. See you there!
∿
∿
Reply all
Reply to author
Forward
0 new messages