Study of SVM-based Incremental Learning for User Adaptation in Multi-class Classification Environment

彭彬彬,孙正兴,金翔宇

DOI: https://doi.org/10.3321/j.issn:0469-5097.2004.02.017

2004-01-01

Abstract:Because of the difference in users' handwritings, drawing styles, and accents, it is necessary to build user adaptation systems in order to identify the users' intentions in the online graphics recognition system. Support Vector Machines (SVM) are learning systems that use a hypothesis space of linear functions in a high dimensional feature space, trained with learning algorithm from optimization theory that implements a learning bias derived from statistical learning theory. SVM-based incremental learning, which can make a user-relevant recognition system quickly adapt to specific users' preference without losing its general performance, is an elegant solution for user adaptation problems in on-line graphics recognition system.Two learning strategies (repetitive learning and incremental learning), two incremental learning algorithms, and two classifier structures (one-against-one and one-against-all) are compared under the multi-class classification environment in our online graphics recognition system. Theoretical analysis and experimental results both show: (1) incremental learning can adapt the classifier to new obtained samples much faster than repetitive learning without losing any precision; (2) the SVM-based incremental learning algorithm of Syed et al. 's is superior to that of Xiao et al's; (3) one-against-one structure is superior to one-against-all structure for a multi-class incremental learning environment.