Online Unsupervised Kernel Affine Projection Algorithms

In recent years, the application of unsupervised learning techniques has become of growing importance in a number of fields, e.g., feature selection, clustering etc. While an array of fast supervised learning algorithms have been developed over the years, the number of fast unsupervised learning algorithms is lacking. In this paper, we present a fast unsupervised kernel affine projection algorithm using a least-squares one-class support vector machine framework and coherence sparsification criterion. A kernel NLMS type algorithm is then developed as a special case. To validate the efficacy of the proposed algorithms, we then perform simulations to detect outliers in datasets.