Program Highlights »
Demonstration
Wed Dec 5th 10:45 AM -- 07:30 PM @ Room 510 ABCD #D3
Play Imperfect Information Games against Neural Networks
Andy C Kitchen · Michela Benedetti · Hon Weng Chong

In this demonstration, attendees can try their skill (and their luck) against deep neural networks at Poker and other assorted imperfect information games. The deep neural network opponents are trained with cutting edge reinforcement learning algorithms to play near a Nash equilibrium. The neural network's value function e.g. how much the neural network expects to win, is visualised as you play — giving a live running estimate of how the neural network assigns value to changing game situations. We visualise internal activity and the NN's best prediction for the human players hidden information, given their past actions. (e.g. what cards are most likely to be in your hand, given your bets? What's the predicted probability you are bluffing?)