Learning and Classifying under Hard Budgets
- Aloak Kapoor, Dept of Computing Science, University of Alberta
- Russ Greiner, Dept of Computing Science; PI of AICML
Since resources for data acquisition are seldom infinite, bothlearners and classifiers must act intelligently under hard budgets. In this paper, we consider problems in which feature values are unknown toboth the learner and classifier, but can be acquired at a cost. Our goalis a learner that spends its fixed learning budget bL acquiring training data, to produce the most accurate 'active classifier' that spends atmost bC per instance. To produce this fixed-budget classifier, the fixed budget learner must sequentially decide which feature values to collect tolearn the relevant information about the distribution. We explore severalapproaches the learner can take, including the standard 'round robin' policy (purchasing every feature of every instance until the bL budget is exhausted). We demonstrate empirically that round robin is problematic (especially for small bL), and provide alternate learning strategies that achieve superior performance on a variety of datasets.
Citation
A. Kapoor, R. Greiner. "Learning and Classifying under Hard Budgets". European Conference on Machine Learning (ECML), Porto, Portugal, pp 166-173, October 2005.| Keywords: | Classification, cost-sensitive, budgeted learning, machine learning |
| Category: | In Conference |
BibTeX
@incollection{Kapoor+Greiner:ECML05,
author = {Aloak Kapoor and Russ Greiner},
title = {Learning and Classifying under Hard Budgets},
Pages = {166-173},
booktitle = {European Conference on Machine Learning (ECML)},
year = 2005,
}Last Updated: October 04, 2007Submitted by Russ Greiner
