A new method for solving overfitting problem of gentle AdaBoost

Shuqiong Wu; Hiroshi Nagahashi

doi:10.1117/12.2050093

10 January 2014 A new method for solving overfitting problem of gentle AdaBoost

Shuqiong Wu, Hiroshi Nagahashi

Proceedings Volume 9069, Fifth International Conference on Graphic and Image Processing (ICGIP 2013); 90691P (2014) https://doi.org/10.1117/12.2050093
Event: Fifth International Conference on Graphic and Image Processing, 2013, Hong Kong, China

Abstract

AdaBoost is a machine learning technique which integrates many weak classifiers into one strong classifier to enhance its classification performance. Gentle AdaBoost is a variant of AdaBoost which introduces Newton steps to the boosting process. It is proved that the overall performance considering both the training error and generalization error of Gentle AdaBoost is better than other AdaBoost variants on low-noise data. However, it suffers from overfitting problem when the training data include high noise. To solve this problem, we propose a new approach to limit the weight distortion according to a stretched distribution of the whole sample weights. Experimental results have shown that our algorithm obtains a better generalization error on both standard and noise-input datasets. Moreover, our method does not increase the calculation time compared with Gentle AdaBoost.

Citation Download Citation

Shuqiong Wu and Hiroshi Nagahashi "A new method for solving overfitting problem of gentle AdaBoost", Proc. SPIE 9069, Fifth International Conference on Graphic and Image Processing (ICGIP 2013), 90691P (10 January 2014); https://doi.org/10.1117/12.2050093

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