Budgeted Learning of Naive Bayes Classifiers - MSc
- Dan Lizotte, University of Michigan
Sometimes, data is not free. We consider the situation where the learner must `purchase' its training data, subject to a fi xed budget. In particular, we examine classifi er learning where observing the value of a feature of a training example has an associated cost, and the total cost of all feature values acquired during training must remain less than the fixed budget. This thesis compares methods for sequentially choosing which feature value to purchase next, given the remaining budget and user's current knowledge of Naive Bayes model parameters. This problem is similar to "active learning", but active learning scenarios assume the ability to purchase class labels, whereas we are interested in purchasing feature values. Also, work on active learning has traditionally focused on myopic (greedy) approaches and uniform/round-robin policies for query selection, but we show that such methods are often suboptimal and present a tractable method for incorporating knowledge of a xed budget in the information acquisition process. This thesis extends ideas from [MLG03], and provides a more complete treatment of the topics covered in [LMG03].
Citation
D. Lizotte. "Budgeted Learning of Naive Bayes Classifiers - MSc". MSc Thesis, University of Alberta, September 2003.| Keywords: | budgeted learning, active learning, machine learning |
| Category: | MSc Thesis |
BibTeX
@mastersthesis{Lizotte:03,
author = {Dan Lizotte},
title = {Budgeted Learning of Naive Bayes Classifiers - MSc},
School = {University of Alberta},
year = 2003,
}Last Updated: May 08, 2009Submitted by Russ Greiner
