Online Matching with Stochastic Rewards

Graph Theory Seminar
Tuesday, February 19, 2013 - 12:05
1 hour (actually 50 minutes)
Skiles 005
Duke University
   The online matching problem has received significant attention in recent years because of its connections to allocation problems in internet advertising, crowd sourcing, etc. In these real-world applications, the typical goal is not to maximize the number of allocations; rather it is to maximize the number of “successful” allocations, where success of an allocation is governed by a stochastic event that comes after the allocation. These applications motivate us to introduce stochastic rewards in the online matching problem. In this talk, I will formally define this problem, point out its connections to previously studied allocation problems, give a deterministic algorithm that is close to optimal in its competitive ratio,  and describe some directions of future research in this line of work. (Based on joint work with Aranyak Mehta.)