70 likes | 278 Views
CS570 Artificial Intelligence Machine Learning. Team 10 Kwak Nohyun Lee youn-ki Hwang young-bae. SVM(Support Vector Machine). We used SVM light software, which is a implementation of Support Vector Machines (SVMs) in C, from WWW.
E N D
CS570 Artificial IntelligenceMachine Learning Team 10 Kwak Nohyun Lee youn-ki Hwang young-bae
SVM(Support Vector Machine) • We used SVM light software, which is a implementation of Support Vector Machines (SVMs) in C, from WWW. • Complex concept, but using SVM S/W was very ease. • In textbook, it was described that SVM with 25,000 support vectors achieved an error rate of 1.1%.
SVM Input Format • Each line in input file means each example. • <line> .=. <target> <feature>:<value> … <feature>:<value> • <target> .=. +1 | -1 | 0 | <float> • <feature> .=. <integer> | "qid" • <value> .=. <float> • Ex)+1 1:0.2 2:0.44 3:0.5 -1 1:0.43 3:0.12 9284:0.2
Image Normalization • Bilinear Interpolation origin image resized image • Estimate from 4 pixel around the matching position • Resize to 30X30 by the experimental experience • Big size -> slow • Small size -> data loss • It mean 30X30 dimension for SVM format
Learning Process • In 0, 1, 2, … 9 data • used 70% for training set • used 30% for test • For positive examples, 7000 ( 70%) data • For negative examples, 9000 data, • 1000 per other digits by the experimental experience
Reference • http://www.cs.cornell.edu/People/tj/svm_light/ • T. Joachims, Making large-Scale SVM Learning Practical. Advances in Kernel Methods - Support Vector Learning, B. Schölkopf and C. Burges and A. Smola (ed.), MIT-Press, 1999.