AdaBoost - 山本 研究室.PDF

AdaBoost The Top Ten Algorithms in Data Mining Chapter 7 2011/07/22 ◆ ()Algorithms 1.1960 -K-nn(1951?), K-means(1967) 2.197080 -EM-Algorithm(1977), NB(1981?), CART(1984) 3.1990 -Apriori(1994), SVM(1995?1992?), -AdaBoost(1995), PageRank(1998) SVM AdaBoost ◆ - - - (Overfitting) 1 M = 3 1 M = 9 t t 0 0 −1 −1 0 1 0 1 x x PRML1 ◆ - -SVMBoosting y = 1 y = 0 y = −1 PRML7 SVM: margin -AdaBoost Boosting() (Adaptive Boosting) ◆ -Boosting(Ensemble Learning) 1Bugging, Boosting, Random Forest - 1.Base(Weak) Learner() 2. - /(Boost) ◆ Notation - - -2 - - - - - - -0-1 ◆ Boosting(1) -: 1.Base(Weak) Learner() 2. -2: -50% (50%) - (51% ) - ◆ Boosting(2) - - - → ◆ Boosting(2) - - - → A General Boosting Procedure 1. Initialize 2. From t = 1 to T. (T: )