|Table of Contents|

An Implementation Method for Minimal VC Dimensional Classifier(PDF)

南京师范大学学报(工程技术版)[ISSN:1006-6977/CN:61-1281/TN]

Issue:
2008年01期
Page:
75-79
Research Field:
Publishing date:

Info

Title:
An Implementation Method for Minimal VC Dimensional Classifier
Author(s):
Wei HuirongXu Jianhua
School of Mathematics and Computer Science,Nanjing Normal University,Nanjing 210097,China
Keywords:
support vector m ach ines kerne l functiona l param eter decompositional m ethod com plex optim ization m ethod
PACS:
O234
DOI:
-
Abstract:
In support o f vectorm ach ine the param eters o f kerne l function have to be adjusted in advance. H owever, the constra ined non linear prog ramm ing of m in im al VC d im ensiona l c lassifie r invo lves the param eter o f RBF kerne,l w hich could be determ ined adaptive ly. In this paper, a fast im plem entation m ethod based on the com plex optim iza tion m ethod, pena lty func tion me thod and grad ient descentm e thod is designed to so lve such a nonlinea r problem. Them ethod has no t on ly good perform ance o f c lassification, but a high speed to dea l w ith la rge data sets. The expe rim en tal resu lts on four benchm ark data se ts demonstra te that our a lgor ithm runs faster and ob tains h ighe r prec is ion than the fam ous SVM l ight algorithm does.

References:

[ 1] Vapn ik V N. 统计学习理论的本质[M ]. 张学工, 译. 北京: 清华大学出版社, 1999.
Vapn ik V N. The Nature o f Statistica lLearn ing Theo ry[M ]. Zhang Xuegong, Translated. Be ijing: Ts inghuaUn iversity Press, 1999. ( in Chinese)
[ 2] V apn ik V N. An ove rv iew of statistical learning theo ry[ J]. IEEE Transactions on Neura lNe tw orks, 1999, 10( 5): 988 999.
[ 3] 张学工. 关于统计学习理论与支持向量机[ J]. 自动化学报, 2001, 26 ( 1): 32 -42.
Zhang Xuegong. About sta tistica l learn ing theory and support vectorm ach ines[ J] . ActaAutom atica Sin ica, 2001, 26 ( 1) : 32- 42. ( in Chinese)
[ 4] W ahba G, Lin Y, Zhang H. M a rg in- like quantities and genera lized approx ima te cross va lida tion for support vector m ach ines [ C] / / Neural Netwo rks fo r Signa l Process ing IX ( Proceed ings o f the 1999 IEEE S igna l Pro cessing Soc ietyW orkshop) . M ad-i son, USA: IEEE Com puter So ciety Press, 1999: 12- 20.
[ 5] W ahbaG, L in X, Gao F, et a.l The b ias-variance tradeo ff and the random ized GACV[ C] / / Advances in In fo rm ation Processing System s 11. Cam bridgeMA: M IT Press, 1999: 620- 626.
[ 6] Chapelle O, Vapnik V, Bousquet O. Choosing mu ltip le param e ters for support vectorm ach ines[ J] . M ach ineLearn ing, 2002, 46( 3): 131- 159.
[ 7] Chung K M, KaoW C, Sun C L, et a.l Radius m arg in bounds for suppo rt vectorm achines w ith the RBF ke rnel[ J]. Neural Com puta tion, 2003, 15( 11): 2 643 -2 681.
[ 8] Xu J H. Des igning non linear c lassifiers through m in im izing VC dim ension bound[ J]. Lecture Notes in Computer Sc iences, 2005, 3496: 900- 905.
[ 9] 徐士良. 常用算法程序集( C语言描述) [M ]. 北京: 清华大学出版社, 2004.
Xu Sh iliang. Set Procedures Comm only U sed A lgor ithm ( C Description Language) [M ]. Be ijing: TsinghuaUn iv ers ity Press, 2004. ( in Chinese)
[ 10] 袁亚湘, 孙文瑜. 最优化理论与方法[M ]. 北京: 科学出版社, 2003.
Yuan Yax iang, SunW enyu. Optim ization Theory andM ethods[M ] . Beijing: Sc ience Press, 2003. ( in Ch inese)
[ 11] H imm e lb lau D M. 实用非线性规划[M ]. 北京: 科学出版社, 1981.
H imm e lb lau D M. Practical Nonlinea r Programm ing[M ] . Beijing: Sc ience Press, 1981. ( in Ch inese)
[ 12] 刘宝光编著. 非线性规划[M ]. 北京: 北京理工大学出版社, 1988. L iu Baoguang. Non linear Prog ramm ing[M ]. Be ijing: Be ijing Institute o f Techno logy Press, 1988. ( in Chinese)
[ 13] Osuna E, Freund R, G iro si F. An improved tra in ing algorithm fo r support vectorm achines[ C] / / IEEEW orkshop on Neural Netwo rks fo r S ignal Processing. Am elea Island, FL: IEEE, 1997: 276- 285.
[ 14] Joachins T. M aking large- scale SVM learn ing practical[ C ] / / Scho lkopf B, Burges C J C, Smo la A J. Advances in Ke rnel M ethods-Support Vector Learn ing. Cam bridgeMA: M IT Press, 1998: 41- 56.
[ 15] P latt J C. Sequential m in im al optim iza tion: a fast algor ithm fo r tra ining suppo rt vector m ach ines[ R]. M icroso ft Research Technical Report: MSR- TR- 98- 14, 1998

Memo

Memo:
-
Last Update: 2013-04-24