DIMACS - Graduate Student Combinatorics Seminar


Title: Multi-armed Bandit Problems

Speaker: Brian Garnett, Rutgers University

Date: Wednesday, April 15, 2015 12:10pm

Location: Graduate Student Lounge, 7th Floor, Hill Center, Rutgers University, Busch Campus, Piscataway, NJ


Abstract:

The basic setup of the multi-armed bandit problem is that there are several slot machines with unknown reward distributions, and you have time/resources for some (maybe unknown) number of plays. If your goal is to maximize your winnings, you'll have to balance the desire to stick with a seemingly good machine with the competing desire to acquire more information about other machines. I'll discuss some strategies depending on the goal and variation of the problem.

Further information can be found at http://math.rutgers.edu/~nhf12/GCS/