An Efficient Exploration Basis for Learning - Elad Hazan - Technion lecture

An Efficient Exploration Basis for Learning Lecture by Elad Hazan associate professor of operations research at the Technion - Faculty of Industrial Engineering and Management - March 27, 2014 Numerous machine learning problems require an exploration basis -- a mechanism to explore the action space. We'll describe a novel geometric notion of an exploration basis and give efficient algorithms to construct such a basis. We'll describe applications to the multi-armed bandit model for complex decision problems such as ad placement and online routing as well as to optimal experiment design.