Novelty detection via linear adaptive filters
Matouš Cejnek · Cvut DSpace (Czech Technical University) · 2020
Novelty detection is an important signal processing task. This task is essential for many industry, and biomedical applications. This thesis is presenting research on the topic of novelty detection utilizing parameters of linear adaptive filters. A new method of adaptive novelty detection is presented in this thesis - Error and Learning Based Novelty Detection. The goal of this thesis is to present the new method as a viable tool for online unsupervised novelty detection in non-stationary and drifted data. The method is supported with various experimental evidence collected from multiple studies. These studies cover multiple traditional applications like system change point detection and outlier detection. The results are obtained from experiments with real and synthetic data.