CS 264 : Beyond Worst - Case Analysis Lecture # 20 : Algorithm - Specific Algorithm Selection ∗

A major theme of CS264 is to use theory to derive good guidance about which algorithm to use to solve a given problem in a given domain. For most problems, there is no “one size fits all” algorithm, and the right algorithm to use depends on the set of inputs relevant for the application. In today’s lecture, we’ll turn this theme into a well-defined mathematical problem, formalized via statistical learning theory. Alternatively, we can think of today’s lecture as a more general approach to the self-improving algorithms discussed in the last lecture. Before presenting the formalism, we consider three motivating examples.