A new online learning with Kernels method in novelty detection
Guoqi Li, Changyun Wen, Zhengguo G. Li · 2011
A new optimization problem together with its model for online learning with kernels in novelty detection is formulated in a Reproducing Kernel Hilbert Space (RKHS). By exploiting the techniques of Lagrange dual problem in a similar way to Vapnik's support vector machine (SVM), the optimization problem is solved iteratively and this gives an algorithm named online learning with kernels denoted as (OLKN). The algorithm is applied to novelty detection including real time background substraction. Such successful applications illustrate the effectiveness of the OLKNin novelty detection.