基于bregman距离和等式约束正则化AdaBoost算法

AdaBoost algorithm based on bregman distance and equality constraint regularization

基于J.Kivinen和M.K.Warmuth提出的一种基于正则化的在线学习模式,提出基于bregman距离和等式约束正则化弱分类器权值更新模式,实现了AdaBoostS,AdaBoostIE,AdaBoostRE,AdaBoostDE和AdaBoostE五种弱分类器权更新算法。在实验部分,利用实际数据对提出的五种算法与Real AdaBoost、Gentle AdaBoost和Modest AdaBoost算法作了比较。

Based on regularizing online learning pattern proposed by J.Kivinen and M.K.Warmuth, update model of weight of weak classifier via bregman distance and equality constraint is devised. Five update algorithms of weight of weak classifier, Ada-BoostS, AdaBoostIE, AdaBoostRE, AdaBoostDE and AdaBoostE are achieved. In the experiments on real datasets, the algorithms performance of five update algorithms with state of art algorithms in assembly classifier research is compared.