Active Learning in Discrete Input Spaces (2002)
Tags
Active Learning, AD-trees, Association Rules, Cached Sufficient Statistics, Efficient Statistical Algorithms, Optimization
Abstract
Traditional design of experiments (DOE) from the statistics literature focuses on optimizing an output parameter over a space of continuous input parameters. Here we consider DOE, or active learning, for descrete input spaces. A trivial example of this is the k-armed bandit problem, which is the case of having a single input attribute of arity k. We address the full problem of many attributes where it is impossible to test every combination of attribute-value pairs even once within the given number of experiments, but we expect to be able to generalize on the results of experiments. We further pose the problem of active learning on fixed experiment sets where we can not choose any possible setting of the input variables, but instead must choose from a fixed set of available experiments. We discuss discrete DOE and fixed experiment sets in marketing and pharmaceutical domains. We propose several active learning algorithms based on the idea of building a function approximator for the experiments taken so far and using its predictions and confidence intervals to select future experiments. The algorithms are tested using commonly available data sets. We conclude with our ideas for extending these algorithms.
Full text
Download (application/pdf, 216.6 kB)
Approximate BibTeX Entry
@inproceedings{schneider-active,
Year = {2002},
Booktitle = {Proceedings of the 34th Interface Symposium},
Author = {
Jeff Schneider, Andrew
Moore
},
Title = {Active Learning in Discrete Input Spaces}
}