Share Email Print

Proceedings Paper

Optimal policy for labeling training samples
Author(s): Lester Lipsky; Daniel Lopresti; George Nagy
Format Member Price Non-Member Price
PDF $14.40 $18.00
cover GOOD NEWS! Your organization subscribes to the SPIE Digital Library. You may be able to download this paper for free. Check Access

Paper Abstract

Confirming the labels of automatically classified patterns is generally faster than entering new labels or correcting incorrect labels. Most labels assigned by a classifier, even if trained only on relatively few pre-labeled patterns, are correct. Therefore the overall cost of human labeling can be decreased by interspersing labeling and classification. Given a parameterized model of the error rate as an inverse power law function of the size of the training set, the optimal splits can be computed rapidly. Projected savings in operator time are over 60% for a range of empirical error functions for hand-printed digit classification with ten different classifiers.

Paper Details

Date Published: 4 February 2013
PDF: 9 pages
Proc. SPIE 8658, Document Recognition and Retrieval XX, 865809 (4 February 2013); doi: 10.1117/12.2005942
Show Author Affiliations
Lester Lipsky, Univ. of Connecticut (United States)
Daniel Lopresti, Lehigh Univ. (United States)
George Nagy, Rensselaer Polytechnic Institute (United States)

Published in SPIE Proceedings Vol. 8658:
Document Recognition and Retrieval XX
Richard Zanibbi; Bertrand Coüasnon, Editor(s)

© SPIE. Terms of Use
Back to Top