[AIRG] Hanabi AI Challenge

["Kuusisto, Finn" <FKuusisto@xxxxxxxxxxxxx>]

Hey AIRGers,
We may not have a meeting today, but next week I will be presenting on a new AI challenge proposed by DeepMind: The Hanabi Challenge.

There has been a lot of recent progress in competitive AI game playing, but there has been very little effort put toward cooperative gameplay with imperfect information. Hanabi is a cooperative card game with imperfect information. It's essentially a set collection game where everyone at the table knows everyone's hand except their own, and must coordinate very minimal information sharing while attempting to play cards to achieve a common goal (sets on the table). It's hard.

I'll try to remember to bring my copy of the game to demonstrate, and we'll discuss the challenge as well as the new software framework DeepMind has published to facilitate experimentation in this space. See you there!

https://arxiv.org/abs/1902.00506 - the paper
https://github.com/deepmind/hanabi-learning-environment - the research platform
https://boardgamegeek.com/boardgame/98778/hanabi - the card game

Where & When:
CS 3310
*4:30pm*, 2/13

Finn Kuusisto, PhD
Postdoctoral Fellow
Regenerative Biology Laboratory
Morgridge Institute for Research
www.morgridge.org