Online unsupervised kernel learning algorithms
Anthony Kuh, Muhammad Sharif Uddin, Phyllis Ng · 2017
In recent years more attention has been focused on unsupervised learning algorithms for feature selection, clustering, and even training of deep neural networks. In this overview paper we discuss work we have conducted on online kernel unsupervised learning algorithms. The algorithms are based on using a least squares cost function, a support vector machine framework, choosing a sparsity criterion, and principles from adaptive filter theory. The algorithms have been applied to anomaly detection problems and particularly detecting bad data on the power grid. We conclude the paper by discussing a broader suite of online learning algorithms including a subspace algorithm and lower cost stochastic gradient algorithms such as the kernel LMS algorithm.