Authors:
Christoph Augenstein
1
;
Norman Spangenberg
1
and
Bogdan Franczyk
2
Affiliations:
1
University of Leipzig, Information Systems Institute, Grimmaische Straße 12, Leipzig and Germany
;
2
University of Leipzig, Information Systems Institute, Grimmaische Straße 12, Leipzig, Germany, Wroclaw University of Economics, ul. Komandorska 118/120, Wroclaw and Poland
Keyword(s):
Neuronal Nets, Deep Learning, Anomaly Detection, Architecture, Data Processing.
Related
Ontology
Subjects/Areas/Topics:
Applications of Expert Systems
;
Artificial Intelligence and Decision Support Systems
;
Enterprise Information Systems
;
Industrial Applications of Artificial Intelligence
;
Information Systems Analysis and Specification
;
Tools, Techniques and Methodologies for System Development
Abstract:
Anomaly detection means a hypernym for all kinds of applications finding unusual patterns or not expected behaviour like identifying process patterns, network intrusions or identifying utterances with different meanings in texts. Out of different algorithms artificial neuronal nets, and deep learning approaches in particular, tend to perform best in detecting such anomalies. A current drawback is the amount of data needed to train such net-based models. Moreover, data streams make situation even more complex, as streams cannot be directly fed into a neuronal net and the challenge to produce stable model quality remains due to the nature of data streams to be potentially infinite. In this setting of data streams and deep learning-based anomaly detection we propose an architecture and present how to implement essential components in order to process raw input data into high quality information in a constant manner.