TadGAN is an unsupervised time series anomaly detection method that uses generative adversarial networks. It introduces a cycle-consistent GAN architecture to map time series data to reconstructed time series. TadGAN trains the GAN with cycle consistency loss and Wasserstein loss. It evaluates anomalies based on reconstruction error scores calculated using simple and dynamic time warping methods, as well as normalized critic scores. The paper aims to improve upon other time series GAN methods like WGAN, BiGAN, and CycleGAN through this approach.