[1]严超,王元庆,李久雪,等.AdaBoost分类问题的理论推导[J].东南大学学报(自然科学版),2011,41(4):700-705.[doi:10.3969/j.issn.1001-0505.2011.04.009]
 Yan Chao,Wang Yuanqing,Li Jiuxue,et al.Theory deduction of AdaBoost classification[J].Journal of Southeast University (Natural Science Edition),2011,41(4):700-705.[doi:10.3969/j.issn.1001-0505.2011.04.009]
点击复制

AdaBoost分类问题的理论推导()
分享到:

《东南大学学报(自然科学版)》[ISSN:1001-0505/CN:32-1178/N]

卷:
41
期数:
2011年第4期
页码:
700-705
栏目:
计算机科学与工程
出版日期:
2011-07-20

文章信息/Info

Title:
Theory deduction of AdaBoost classification
作者:
严超1王元庆1李久雪2张兆扬3
(1南京大学电子科学与工程学院,南京210093)
(2东南大学信息科学与工程学院,南京 210096)
(3上海大学新型显示技术及应用集成教育部重点实验室,上海 200444)
Author(s):
Yan Chao1Wang Yuanqing1Li Jiuxue2Zhang Zhaoyang3
(1School of Electric Science and Engineering, Nanjing University, Nanjing 210093, China)
(2School of Information Science and Engineering, Southeast University, Nanjing 210096,China)
(3Key Laboratory of Advanced Display and System Application of Ministry of Education, Shanghai University, Shanghai 200444,China)
关键词:
多分类AdaBoost算法归一化因子贝叶斯推理
Keywords:
multi-classification AdaBoost algorithm normalization factor Bayes inference
分类号:
TP301.6
DOI:
10.3969/j.issn.1001-0505.2011.04.009
摘要:
为解决AdaBoost算法在二分类问题及多分类问题上缺乏共同的理论基础,算法系列的系统性难以得到理论诠释这一问题,首先,从算法思想的层面对AdaBoost算法与最优贝叶斯推理的关系进行了探讨; 然后对AdaBoost算法的训练流程及相关参量进行了定量分析; 最后从基本不等式定理入手,重点推导了AdaBoost算法由二分类问题向多分类问题延展的理论依据,探讨了AdaBoost算法的本质.总结并证明了AdaBoost算法的2条理论基础:当非负数之和是一个定值时,其差值越大则其乘积越小; 非负数的算术平均数大于等于它们的几何平均数.并且分别就二分类问题和多分类问题对AdaBoost算法的应用提出了优化策略.
Abstract:
AdaBoost two-classification and AdaBoost multi-classification lack mutual theory principals, so the unity of AdaBoost algorithm could not be represented theorically. To solve this problem, firstly, the connection of AdaBoost algorithm and Bayes Inference is probed; secondly, the training process and relative parameters of AdaBoost algorithm are analyzed quantitatively; thirdly, with fundamental inequality principals, the extension process of AdaBoost algorithm from two-classification application to multi-classification application is reasoned. Two intrinsic theories are summarized and proved: if the sum of some non-negative numbers is fixed, their product will become smaller when their values difference become greater; arithmetic average of some non-negative numbers is greater than their geometric average. In addition, some improvements to two-classification and multi-classification applications are suggested.

参考文献/References:

[1] Freund Yoav,Schapire Robert E.A decision-theoretic generalization of on-line learning and an application to boosting[J].Journal of Computer and System Sciences 1997,13(55):119-139.
[2] Schapire Robert E,Singer Yoram.Improved boosting algorithms using confidence-rated predictions[J].Machine Learning,1999,37(3):297-336.
[3] Zhao Xiaohui,Liang Fuzhong,Yu Yao,et al.The AdaBoost algorithm with prior probabilities and the visualization demonstrated in GIS for geo-hazard forecasting[C]//Ninth International Conference on Hybrid Intelligent Systems.Chengdu,China,2009:436-441.
[4] Qahwaji R,Omari M,Colak T,et al.Using the real,gentle and modest AdaBoost learning algorithms to investigate the computerised associations between coronal mass ejections and filaments[C]//International Conference on Communications,Computers and Applications.Bradford,United Kingdom,2008:37-42.
[5] Windeatt T,Ardeshir G.Decision tree simplification for classifier ensembles[J].International Journal of Pattern Recognition and Artificial Intelligence,2004,18(5):749-776.
[6] Hao Wei,Luo Jiebo.Generalized multiclass AdaBoost and its applications to multimedia classification[C]//Computer Vision and Pattern Recognition Workshop.New York,USA,2006:112-116.
[7] Zhu Ji,Zou Hui,Saharon Rosset,et al.Multi-class AdaBoost[J].Statistics And Its Interface,2009,3(2):349-360.
[8] Shi Hongbo,Lü Xiaoyong.The Nave Bayesian classifier learning algorithm based on Adaboost and parameter expectations[C]//Third International Joint Conference on Computational Science and Optimization.Wuhan,China,2010:665-670.
[9] Li Weihua,Liu Weiyi,Yue Kun.Recovering the global structure from multiple local bayesian networks[J].International Journal on Artificial Intelligence Tools,2008,17(6):1067-1088.
[10] Liu Weipeng.Bayesian method[EB/OL].(2008-09-21) [2010-12-10].http://blog.csdn.net/pongba/archive/2008/09/21/2958094.aspx.
[11] Wang Zhanjun,Fang Chi,Ding Xiaoqing.Asymmetric Real Adaboost[C]//The 19th International Conference on Pattern Recognition.Tampa,USA,2008:234-239.
[12] Yan Chao,Wang Yuanqing.Research of the real Adaboost algorithm[J].Computer Science,2010,9(37):209-212.

相似文献/References:

[1]顾明亮,夏玉果,张长水,等.基于AdaBoost的汉语方言辨识[J].东南大学学报(自然科学版),2008,38(4):585.[doi:10.3969/j.issn.1001-0505.2008.04.008]
 Gu Mingliang,Xia Yuguo,Zhang Changshui,et al.AdaBoost based Chinese dialect identification[J].Journal of Southeast University (Natural Science Edition),2008,38(4):585.[doi:10.3969/j.issn.1001-0505.2008.04.008]

备注/Memo

备注/Memo:
作者简介:严超(1986—),男,博士生;王元庆(联系人),男,博士,教授,博士生导师,yqwang@nju.edu.cn.
基金项目:国家自然科学基金重点资助项目(608320036)、新型显示技术及应用集成教育部重点实验室资助项目(P200902)、南京大学研究生创新基金资助项目(2011CL03)、江苏省研究生培养创新工程资助项目.
引文格式: 严超,王元庆,李久雪,等.AdaBoost分类问题的理论推导[J].东南大学学报:自然科学版,2011,41(4):700-705.[doi:10.3969/j.issn.1001-0505.2011.04.009]
更新日期/Last Update: 2011-07-20