We present an efficient algorithm to actively select samples which accurately model a target function below a specified threshold. We describe several heuristics for choosing these samples, and show how these heuristics perform on synthetic and real target functions. We then show how these algorithms can be used to make a sophisticated statistical procedure for finding confidence regions two orders of magnitude more data efficient.

17 pages

*Robotics Institute, Carnegie Mellon University
**Department of Statistics, Carnegie Mellon University

This page maintained by reports@cs.cmu.edu