Timezone: »

 
Poster
On GANs and GMMs
Eitan Richardson · Yair Weiss

Wed Dec 05 02:00 PM -- 04:00 PM (PST) @ Room 210 #59
in Wed Poster Session B »

A longstanding problem in machine learning is to find unsupervised methods that can learn the statistical structure of high dimensional signals. In recent years, GANs have gained much attention as a possible solution to the problem, and in particular have shown the ability to generate remarkably realistic high resolution sampled images. At the same time, many authors have pointed out that GANs may fail to model the full distribution ("mode collapse") and that using the learned models for anything other than generating samples may be very difficult.

In this paper, we examine the utility of GANs in learning statistical models of images by comparing them to perhaps the simplest statistical model, the Gaussian Mixture Model. First, we present a simple method to evaluate generative models based on relative proportions of samples that fall into predetermined bins. Unlike previous automatic methods for evaluating models, our method does not rely on an additional neural network nor does it require approximating intractable computations. Second, we compare the performance of GANs to GMMs trained on the same datasets. While GMMs have previously been shown to be successful in modeling small patches of images, we show how to train them on full sized images despite the high dimensionality. Our results show that GMMs can generate realistic samples (although less sharp than those of GANs) but also capture the full distribution, which GANs fail to do. Furthermore, GMMs allow efficient inference and explicit representation of the underlying statistical structure. Finally, we discuss how GMMs can be used to generate sharp images.

Keywords: Adversarial Networks · Generative Models · Latent Variable Models
[ Paper]  [ Poster] 

Author Information

Eitan Richardson (The Hebrew University of Jerusalem)
Yair Weiss (Hebrew University)

Yair Weiss is an Associate Professor at the Hebrew University School of Computer Science and Engineering. He received his Ph.D. from MIT working with Ted Adelson on motion analysis and did postdoctoral work at UC Berkeley. Since 2005 he has been a fellow of the Canadian Institute for Advanced Research. With his students and colleagues he has co-authored award winning papers in NIPS (2002),ECCV (2006), UAI (2008) and CVPR (2009).

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors