CRITERIA AND TECHNIQUES FOR PROCESSING NOISY DATA WITH ANOMALOUS VALUES

Abstract

An approach to solve the problem of processing noisy data with anomalous values when the statistics of noise and anomalies are unknown is proposed. This approach is based on a method of generalized maximum likelihood and a superset of cost functions. The superset allows to tune the data processing onto the current noise environment. This tuning is performed by setting the values of three free parameters related to the scale, the heaviness of tails, and the form of random values distribution, as well as to the presence of anomalous values. In the general case, the proposed approach requires to solve a multimodal optimization problem.