An improved algorithm of OMKC based on the optimized perceptron with the best kernel
Bingjie Cheng, Shangping Zhong · 2014
Online Multiple Kernel Classification (OMKC) algorithm has been a popular method for exploring effective online combination of multiple kernel classifiers. The framework for OMKC is commonly obtained by learning multiple kernel classifiers and simultaneously their linear combination. However, the traditional perceptron algorithm which OMKC algorithm bases on does not achieve a much smaller mistake rate. In this paper, we put forward a novel algorithm based on OMKC using an improved perceptron algorithm. Our perceptron algorithm is applied with the best kernel. The algorithm produces an online validation procedure to search for the best kernel among the pool of kernels using the first 10% training examples. By using histograms analysis, our proposed algorithm can achieve smaller mistake rate and less time consuming. Extensive experimental results on twelve data sets demonstrate the effectiveness and efficiency of our algorithm.