Smart and intelligent processes have become common in business and industry. Computer networks monitor and control a myriad of physical processes, and their protection against deliberate intrusions, false data injections, and other forms of cyber attacks has become of paramount importance. Economic and financial data are being affected by systemic and systematic risks. Random deviations of event-related potentials in neuroscience may not easily be recognized even by trained specialists. Wireless communications tend to be affected by a myriad of factors that diminish the quality of signals. Therefore, generally speaking, our ability to distinguish genuine signals from various forms of noise (anomalies, aberrations, etc.) has become more important than ever.  

Many sophisticated statistical and other methods have been suggested for tackling such problems, but they concentrate on detecting so-called overt anomalies, which are those that noticeably change the regime (e.g., means, variances, autocovariances, etc.) of genuine signals. Those methods are based on probabilistic arguments, stochastic processes techniques, deep learning, and artificial neural networks, to name a few.  

This paper advances anomaly detection in a distinctly different direction by developing an anomaly-detection method, which, unlike many existing methods, successfully tackles so-called covert anomalies.  These are anomalies that are usually small in size and thus do not immediately derail the system’s performance. The anomalies can therefore stay undetected for much longer periods of time and thus do considerable damage. The covert anomalies may even mimic genuine, time-series driven, random signals and are thus hard to notice even to the trained eyes. The proposed in this paper anomaly-detection method can successfully detect even such systematic anomalies (e.g., intrusions), as the provided extensive simulation studies demonstrate.   

 The paper gives a considerable impetus to the advancement of covert-signal detection methodologies and their practical implementation, thus facilitating timely risk recognition, assessment, and classification, all of which enable the development and implementation of appropriate system vulnerability and reliability assessment tools designed for a variety of specific risks, their mitigation strategies, and system resilience advancement.