Image Image Image

Post new topic Reply to topic  [ 1 post ] 
Author Message
 Post subject: Safe Learning in Zero-sum Games
PostPosted: Sun Aug 12, 2012 2:42 pm 
Senior member
User avatar

Posts: 124
Favourite Bot: coming
Safe Learning in Zero-sum Games

Authors : Kevin Waugh, Michael Bowling

Abstract :

A popular and effective approach for decision-making in zero-sum adversarial settings is to use a precomputed equilibrium strategy.
This approach is safe as it guarantees the highest expected reward against a worst-case adversary.
Unfortunately, this safety guarantee can come at a cost as it does not actively take advantage of an opponent’s weaknesses.
Despite this flaw, alternative approaches are far less appealing.
Even in small games, online learning and complex opponent modeling techniques fall short as they can require a substantial number of observations to achieve a reasonable level of play and far more for exploiting a weaker opponent.
In this paper we present two methods that marry a safety guarantee of an equilibrium strategy with the exploitive potential of online learning algorithms.
The first, ε-safe learning, learns to exploit an opponent using only ε-equilibrium strategies.
The second, equilibrium restricted learning, uses an online learner within the support of an equilibrium. We prove safety guarantees for both techniques, with the former’s guarantee stronger than the latter’s, and demonstrate effective learning, with the latter’s performance stronger than the former’s.

 Profile E-mail  
Display posts from previous:  Sort by  
Post new topic Reply to topic  [ 1 post ] 

Who is online

Users browsing this forum: No registered users and 3 guests

You cannot post new topics in this forum
You cannot reply to topics in this forum
You cannot edit your posts in this forum
You cannot delete your posts in this forum
You cannot post attachments in this forum

Search for:
Jump to: