Generative Adversarial-Synergetic Networks for Anomaly Detection

Hongjun Li, Chaobo Li, Ze Zhou

Abstract: Anomaly detection is an important and demanding problem in social harmony. However, due to the uncertainty, irregularity, diversity and scarcity of abnormal samples, the performance is often poor. This paper presents Generative Adversarial-Synergetic Networks (GA-SN) to improve the discriminative ability. It is built on the Adversarial Discriminant Network (ADN) for detecting anomaly and the Synergetic Generative Network (SGN) for extracting pivotal and discriminant feature. The ADN takes advantage of adversary structure between discriminator and the asymmetric generator. As an important hub of ADN and SGN, the asymmetric generator is combined with promotor by synergy in SGN. Different from other methods which based on generative adversarial network, our method pays more attention to the discriminant performance rather than the reconstruction performance. Our method was verified on MNIST and UCSD Ped2 dataset, where MNIST obtained higher values on both area under curve and F1-score, and 12.56% equal error rate is obtained on UCSD Ped2. The experimental results show the superiority of proposed method compared to existing methods. So the GA-SN combines synergy and adversarial relationship, which helps to improve discriminative performance in detection.