An On-Line Learning Approach with Support Vector Dormain Classifier

As one kind of one-class classifier, support vector domain classifier (SVDC) has worked well for the batch model learning problems. But with real-world database increase in size, there is a need to scale up inductive learning algorithm to handle more training data. On-line learning technique is one possible solution to the scalability problem, where data is processed in parts, and the result combined so as to use less memory. This paper presented an on-line learning algorithm based on SVDC, and the basic idea of the proposed algorithm is to obtain the initial target concepts using SVDC during the training phase and then update these target concepts by an updating model. Different from the existed on-line learning approaches, in our algorithm, the model updating procedure equals to solve a quadratic programming (QP) problem, and the updated model still owns the property of spars solution. Compared with other existed on-line learning algorithms, the inverse procedure of our algorithm (i.e. decreasing learning) is easy to conduct without extra computation