TY - JOUR
T1 - Chunk incremental learning for cost-sensitive hinge loss support vector machine
AU - Gu, Bin
AU - Quan, Xin
AU - Gu, Yunhua
AU - Sheng, Victor S.
AU - Zheng, Guansheng
N1 - Funding Information:
This work was supported by the Project Funded by the Priority Academic Program Development (PAPD) of Jiangsu Higher Education Institutions, the Natural Science Foundation (No. BK20161534), Six talent peaks project (No. XYDXX-042) and the 333 Project (No. BRA2017455) in Jiangsu Province, the U.S. National Science Foundation (IIS-1115417), and the National Natural Science Foundation of China (No. 61573191).
Funding Information:
This work was supported by the Project Funded by the Priority Academic Program Development (PAPD) of Jiangsu Higher Education Institutions, the Natural Science Foundation (No. BK20161534 ), Six talent peaks project (No. XYDXX-042) and the 333 Project (No. BRA2017455) in Jiangsu Province, the U.S. National Science Foundation ( IIS-1115417 ), and the National Natural Science Foundation of China (No. 61573191 ).
Publisher Copyright:
© 2018
PY - 2018/11
Y1 - 2018/11
N2 - Cost-sensitive learning can be found in many real-world applications and represents an important learning paradigm in machine learning. The recently proposed cost-sensitive hinge loss support vector machine (CSHL-SVM) guarantees consistency with the cost-sensitive Bayes risk, and this technique provides better generalization accuracy compared to traditional cost-sensitive support vector machines. In practice, data typically appear in the form of sequential chunks, also called an on-line scenario. However, conventional batch learning algorithms waste a considerable amount of time under the on-line scenario due to re-training of a model from scratch. To make CSHL-SVM more practical for the on-line scenario, we propose a chunk incremental learning algorithm for CSHL-SVM, which can update a trained model without re-training from scratch when incorporating a chunk of new samples. Our method is efficient because it can update the trained model for not only one sample at a time but also multiple samples at a time. Our experimental results on a variety of datasets not only confirm the effectiveness of CSHL-SVM but also show that our method is more efficient than the batch algorithm of CSHL-SVM and the incremental learning method of CSHL-SVM only for a single sample.
AB - Cost-sensitive learning can be found in many real-world applications and represents an important learning paradigm in machine learning. The recently proposed cost-sensitive hinge loss support vector machine (CSHL-SVM) guarantees consistency with the cost-sensitive Bayes risk, and this technique provides better generalization accuracy compared to traditional cost-sensitive support vector machines. In practice, data typically appear in the form of sequential chunks, also called an on-line scenario. However, conventional batch learning algorithms waste a considerable amount of time under the on-line scenario due to re-training of a model from scratch. To make CSHL-SVM more practical for the on-line scenario, we propose a chunk incremental learning algorithm for CSHL-SVM, which can update a trained model without re-training from scratch when incorporating a chunk of new samples. Our method is efficient because it can update the trained model for not only one sample at a time but also multiple samples at a time. Our experimental results on a variety of datasets not only confirm the effectiveness of CSHL-SVM but also show that our method is more efficient than the batch algorithm of CSHL-SVM and the incremental learning method of CSHL-SVM only for a single sample.
KW - Chunk incremental learning
KW - Cost-sensitive learning
KW - Hinge loss
KW - Support vector machines
UR - http://www.scopus.com/inward/record.url?scp=85048503210&partnerID=8YFLogxK
U2 - 10.1016/j.patcog.2018.05.023
DO - 10.1016/j.patcog.2018.05.023
M3 - Article
AN - SCOPUS:85048503210
VL - 83
SP - 196
EP - 208
JO - Pattern Recognition
JF - Pattern Recognition
SN - 0031-3203
ER -