Distributed Generative Adversarial Networks for Anomaly Detection

摘要

Cognitive radio networks can be used to detect anomalous and adversarial communications to achieve situational awareness on the radio frequency spectrum. This paper proposes a distributed anomaly detection scheme based on adversarially-trained data models. While many anomaly detection methods typically depend on a central decision-making server, our distributed approach makes better use of decentralized resources, and decreases reliance on a single point of failure. Using a novel combination of generative adversarial network (GAN) elements, participating cognitive radio devices learn a representation of local network activity data through a non-cooperative (strategic) game. Deviations from this expected network activity are flagged as anomalies and treated as possible network security threats, improving situational awareness. Tested on a range of time series datasets, the performance of the proposed distributed scheme matches that of state-of-the-art, centralized anomaly detection methods.