Recent Advancements in Multi-armed Bandit: from Clinical Trials to Sponsored Search Auctions
DEI - Seminar Room
April 13th, 2012
2.30 - 3.30 p.m.
The multi-armed bandit model effectively describes a wide range of online decision-making problems such as identifying the most effective treatment in clinical trial, finding the advertisement which is more likely to be clicked in a web advertising platform, discovering the shortest path in a graph, and so on. In this talk, we will cover some recent advancements in the multi-armed bandit extending the standard setting to novel problems such as the best-arm identification (eg, returning the best treatment after a finite number of patients), active bandit setting (eg, accurate estimation of the reliability of different production lines), and the application of bandits to strategic settings (eg, web advertising).
Artificial intelligence, robotics and computer vision