Computational Performance of the Parameters Estimation in Extreme Seeking Entropy Algorithm

Abstract

This paper is dedicated to the evaluation of the computational time performance of the algorithms that estimate the parameters of the generalized Pareto distribution, namely Method of Moments, Maximum likelihood estimator and Quasi-maximum likelihood algorithms. The generalized Pareto distribution is utilized by the Extreme Seeking Entropy algorithm to detect novelty in data. The algorithm is evaluating the weight increments of the simple adaptive filter that are obtained via incrementally learning algorithm. The computational time performance is examined in the experiment with the detection of step-change parameters of the signal generator. Its output contains also additive Gaussian noise.