Seminar on Computational Learning and Adaptation


  Achieving Efficient and Cognitively Plausible Learning in Backgammon

Scott Sanner
CMU, Stanford
ACT-R Research Group
Pittsburgh, PA
ssanner@cs.stanford.edu

Traditionally, computer applications to game domains have taken a bruteforce approach, relying on sheer computational power to overcome the complexity of the domain. Although many of these programs have been quite successful, it is interesting to note that humans can still perform extremely well against computer opponents in many game domains. Thus we are compelled to ask, if no human could match the computational power of most of these programs, are there methods for learning and performance in game domains that more closely reflect human cognition? In response to this question, this talk will cover an attempt to model how humans learn and play games by developing a Backgammon-playing algorithm based on the ACT-R cognitive architecture. Analysis of this algorithm shows that it is efficient and commensurate with human abilities suggesting that it provides a cognitively plausible theory of learning in Backgammon.


Date: Thurs., Jan 18

Time: 4:15-5:30PM

Place: Cordura 100


Return to the seminar schedule