Machine Learning Department
School of Computer Science, Carnegie Mellon University


Actively Learning Level-Sets of
Composite Functions

Brent Bryan, Jeff Schneider*

December 2007


Keywords: Active learning, level-sets

Scientists frequently have multiple types of experiments and data sets on which they can test the validity of their models and the plausible or optimal regions for the model parameters. Identifying these parameter regions reduces to finding a level set on a function defined as a composite of the evaluations of each experiment or data set for a parameter setting. An active learning algorithm for this problem must at each iteration select a parameter setting to be tested and decide which experiment type to use for the test. We propose an active learning algorithm for identifying level sets of composite functions. Empirical tests on synthetic functions and on real data for a 7D cosmological model show it significantly reduces the number of samples required to identify desired regions.

16 pages

*Robotics Institute, Carnegie Mellon University

SCS Technical Report Collection
School of Computer Science homepage

This page maintained by