Data Condensation in Large Databases by Incremental Learning with Support Vector Machines

An algorithmfor data condensation using support vector machines (SVM's)is presented. The algorithm extracts datapoints lying close to the class boundaries,whichform a much reducedbut critical setfor classification. Theproblem of large memory requirementsfor training SVM's in batch mode is circumvented by adopting an active incremental learning algorithm. The learning strategy is motivatedfrom the condensed nearest neighbor classification technique. Experimental resultspresented show that such active incremental learning enjoy superiority in terms of computation time and condensation ratio, over related methods.

• Type: Conference
• Year: 2000
• Where: ICPR
• Authors: Pabitra Mitra, C. A. Murthy, Sankar K. Pal
• Keywords: Active Incremental Learning, Computer Vision, ICPR 2000, Incremental Learning Algorithm, Incremental Learning Enjoy