Cloud-GAN: Cloud Generation Adversarial Networks for anomaly detection

Xianhua Zeng, Yang Zhuo, Tianxing Liao, Jueqiu Guo

Abstract: Highlights•For the first time, we introduce the cloud model into GAN, and construct a novel GAN called Cloud-GAN for anomaly detection.•We adopt an adaptive weighting technique to optimize the training process of traditional variational autoencoders based on adaptive weights, so as to enhance the full use of different layer features of the model, thereby avoiding indiscriminate and biased learning of the model.•We combine Cloud-GAN based on adaptive weight with Gaussian kernel density function, and send the low-dimensional representation and the relative Euclidean distance between input and output from the generating network of Cloud-GAN into the kernel function for modeling, so as to carry out effective anomaly detection.