Maximum-Entropy Sampling

Abstract

The goal of maximum entropy sampling is to choose a most informative subset of s random variables from a set of n random variables, subject to side constraints. A typical side constraint might be a budget restriction, where one has a cost for observing each random variable. Other possibilities include logical constraints (e.g. multiple choice or precedence constraints). In many situations, one can assume that the random variables are Gaussian, or that they can be suitably transformed.