Efficient Multiclass Boosting Classification with Active Learning Export

@inproceedings{citeulike:1650290, abstract = {We propose a novel multiclass classi¯cation algorithm Gen- tle Adaptive Multiclass Boosting Learning (GAMBLE). The algorithm naturally extends the two class Gentle AdaBoost algorithm to multiclass classi¯cation by using the multi- class exponential loss and the multiclass response encoding scheme. Unlike other multiclass algorithms which reduce the K-class classi¯cation task to K binary classi¯cations, GAMBLE handles the task directly and symmetrically, with only one committee classi¯er. We formally derive the GAM- BLE algorithm with the quasi-Newton method, and prove the structural equivalence of the two regression trees in each boosting step. To scale up to large datasets, we utilize the generalized Query By Committee (QBC) active learning framework to focus learning on the most informative samples. Our em- pirical results show that with QBC-style active sample se- lection, we can achieve faster training time and potentially higher classi¯cation accuracy. GAMBLE's numerical supe- riority, structural elegance and low computation complexity make it highly competitive with state-of-the-art multiclass classi¯cation algorithms.}, author = {Huang, Jian and Ertekin, Seyda and Song, Yang and Zha, Hongyuan and Giles, Lee C.}, citeulike-article-id = {1650290}, citeulike-linkout-0 = {http://clgiles.ist.psu.edu/papers/SDM2007-boosting-active-learning.pdf}, journal = {Seventh SIAM International Conference}, keywords = {active-learning, boosting, classification, multi-class}, posted-at = {2007-09-13 04:40:22}, priority = {3}, title = {Efficient Multiclass Boosting Classification with Active Learning}, url = {http://clgiles.ist.psu.edu/papers/SDM2007-boosting-active-learning.pdf}, year = {2007} }

TY - CONF ID - citeulike:1650290 L3 - citeulike-article-id:1650290 N2 - We propose a novel multiclass classi¯cation algorithm Gen- tle Adaptive Multiclass Boosting Learning (GAMBLE). The algorithm naturally extends the two class Gentle AdaBoost algorithm to multiclass classi¯cation by using the multi- class exponential loss and the multiclass response encoding scheme. Unlike other multiclass algorithms which reduce the K-class classi¯cation task to K binary classi¯cations, GAMBLE handles the task directly and symmetrically, with only one committee classi¯er. We formally derive the GAM- BLE algorithm with the quasi-Newton method, and prove the structural equivalence of the two regression trees in each boosting step. To scale up to large datasets, we utilize the generalized Query By Committee (QBC) active learning framework to focus learning on the most informative samples. Our em- pirical results show that with QBC-style active sample se- lection, we can achieve faster training time and potentially higher classi¯cation accuracy. GAMBLE's numerical supe- riority, structural elegance and low computation complexity make it highly competitive with state-of-the-art multiclass classi¯cation algorithms. JF - Seventh SIAM International Conference TI - Efficient Multiclass Boosting Classification with Active Learning KW - active-learning KW - boosting KW - classification KW - multi-class AU - Huang, Jian AU - Ertekin, Seyda AU - Song, Yang AU - Zha, Hongyuan AU - Giles, Lee PY - 2007/// UR - http://clgiles.ist.psu.edu/papers/SDM2007-boosting-active-learning.pdf ER -