Mixture Density Conditional Generative Adversarial Network Models\n (MD-CGAN)

2020/04/07 by Jaleh Zand, Stephen Roberts, Zand, Jaleh +1
Computer Science · #Generative Adversarial Networks and Image Synthesis #Gaussian Processes and Bayesian Inference #Anomaly Detection Techniques and Applications

paper · pdf · doi:10.48550/arxiv.2004.03797

Abstract

Generative Adversarial Networks (GANs) have gained significant attention in\nrecent years, with impressive applications highlighted in computer vision in\nparticular. Compared to such examples, however, there have been more limited\napplications of GANs to time series modelling, including forecasting. In this\nwork, we present the Mixture Density Conditional Generative Adversarial Model\n(MD-CGAN), with a focus on time series forecasting. We show that our model is\ncapable of estimating a probabilistic posterior distribution over forecasts and\nthat, in comparison to a set of benchmark methods, the MD-CGAN model performs\nwell, particularly in situations where noise is a significant component of the\nobserved time series. Further, by using a Gaussian mixture model as the output\ndistribution, MD-CGAN offers posterior predictions that are non-Gaussian.\n

Related

• Probabilistic Generative Adversarial Networks
• Generative Adversarial Network for Probabilistic Forecast of Random Dynamical System
• Flow-GAN: Combining Maximum Likelihood and Adversarial Learning in\n Generative Models
• Benchmarking Regression Methods: A comparison with CGAN
• Time Series Simulation by Conditional Generative Adversarial Net