Improvements to Platt's SMO algorithm for SVM classifier design

被引：1223

作者：

Keerthi, SS ^{[1
]}

Shevade, SK

Bhattacharyya, C

Murthy, KRK

机构：

[1] Natl Univ Singapore, Dept Mech & Prod Engn, Singapore 119260, Singapore

[2] Indian Inst Sci, Dept Comp Sci & Automat, Bangalore 560012, Karnataka, India

来源：

NEURAL COMPUTATION | 2001年 / 13卷 / 03期

关键词：

D O I：

10.1162/089976601300014493

中图分类号：

TP18 [人工智能理论];

学科分类号：

081104 ; 0812 ; 0835 ; 1405 ;

摘要：

This article points out an important source of inefficiency in Platt's sequential minimal optimization (SMO) algorithm that is caused by the use of a single threshold value. Using clues from the KKT conditions for the dual problem, two threshold parameters are employed to derive modifications of SMO. These modified algorithms perform significantly faster than the original SMO on all benchmark data sets tried.

引用

页码：637 / 649

页数：13

共 12 条

[1]

[Anonymous], 1990, SUPPORT VECTOR LEARN

[2]

BENNETT KP, 1992, OPTIMIZATION METHODS, V1, P23, DOI DOI 10.1080/10556789208805504

[3]

Burges C. J., 1998, DATA MINING KNOWLEDG, V3

[4]

FRIES TT, 1998, SUPPORT VECTOR NETWO

[5]

Joachims T., 1998, ADV KERNEL METHODS S

[6]

KEERTHI SS, 1999, CD9914 NAT U SING DE

[7]

KEERTHI SS, 1999, TRISL9903 IND I SCI

[8]

MANGASARIAN OL, 1998, SUCCESSIVE OVERRELAX

[9]

PLATT J, 1999, ADV NEURAL INFORMATI

[10]

SHEVADE SK, 1999, CD9916 NAT U SING

← 1 2 →