Adaptive classification by maximizing class separability with respect to the unlabeled data

Robert S. Lynch Jr.; Peter K. Willett

doi:10.1117/12.487565

8 August 2003 Adaptive classification by maximizing class separability with respect to the unlabeled data

Proceedings Volume 5107, System Diagnosis and Prognosis: Security and Condition Monitoring Issues III; (2003) https://doi.org/10.1117/12.487565
Event: AeroSense 2003, 2003, Orlando, Florida, United States

Abstract

In this paper, the Adaptive Mean-Field Bayesian Data Reduction Algorithm is discussed, which utilizes a method that maximizes the class separability of unlabeled training data. The algorithm is based on a Dirichlet distribution model for each class. In this new method, the Dirichlet model is extended such that dissimilar distributions are encouraged amongst the classes with respect to unlabeled data, and with respect to data containing missing values. It has previously been shown for two class cases that the theoretical probability of error is lower bounded by 0.25 under the original Dirichlet model. Thus, the new model has been developed with the idea of encouraging error probabilities below this lower bound given the data contains missing information, such as the class labels. Results are illustrated with simulated data as applied to sequential classification using Page's test. In general it is shown that the new method's performance is superior to that of the original Dirichlet model, where it is apparent that any previously acquired unlabeled data are being utilized in the training set to improve the correct classification of future test data samples.

Citation Download Citation

Robert S. Lynch Jr. and Peter K. Willett "Adaptive classification by maximizing class separability with respect to the unlabeled data", Proc. SPIE 5107, System Diagnosis and Prognosis: Security and Condition Monitoring Issues III, (8 August 2003); https://doi.org/10.1117/12.487565

ACCESS THE FULL ARTICLE

INSTITUTIONAL
Select your institution to access the SPIE Digital Library.

SELECT YOUR INSTITUTION

PERSONAL
Sign in with your SPIE account to access your personal subscriptions or to use specific features such as save to my library, sign up for alerts, save searches, etc.

PERSONAL SIGN IN

No SPIE Account? Create one

PURCHASE THIS CONTENT

SUBSCRIBE TO DIGITAL LIBRARY

50 downloads per 1-year subscription

Members: $195

Non-members: $335 ADD TO CART

25 downloads per 1 - year subscription

Members: $145

Non-members: $250 ADD TO CART

PURCHASE SINGLE ARTICLE

Includes PDF, HTML & Video, when available