Figure 2 - uploaded by Lu Mi
Content may be subject to copyright.
Generated images by GAN and WGAN models trained on MNIST after 1,100k,500k,1000k iterations.

Generated images by GAN and WGAN models trained on MNIST after 1,100k,500k,1000k iterations.

Source publication
Preprint
Full-text available
Both generative adversarial network models and variational autoencoders have been widely used to approximate probability distributions of datasets. Although they both use parametrized distributions to approximate the underlying data distribution, whose exact inference is intractable, their behaviors are very different. In this report, we summarize...

Contexts in source publication

Context 1
... order to provide a quantitative analysis, we used entropy of the synthetic data distribution to measure When mode collapse happens, the entropy will keep decreasing. For example, we show the training process of GAN and WGAN, shown in Fig. 2. After 1000k iterations, the final images sampled from GAN only contain digit 1, which indicates that it is prone to mode collapse. In contrast, for WGAN, the final result is composed of various different digits, the mode collapse issue is mitigated significantly. The entropy shown in Fig. 3 of each iteration decreases rapidly, and ...
Context 2
... accuracy on the test dataset. Then we used the clas- sifier to recognize the handwritten digits generated from GAN and WGAN and calculated the entropy of the generative distribution for each training iter- ation. Let p i represent the probability of each digit i sampled from the generative network at each iter- process of GAN and WGAN, shown in Fig. 2. Af- ter 1000k iterations, the final images sampled from GAN only contain digit 1, which indicates that it is prone to mode collapse. In contrast, for WGAN, the final result is composed of various different digits, the mode collapse issue is mitigated significantly. The entropy shown in Fig. 3 of each iteration de- creases rapidly, and ...

Similar publications

Preprint
Full-text available
Dynamic heterogeneous networks describe the temporal evolution of interactions among nodes and edges of different types. While there is a rich literature on finding communities in dynamic networks, the application of these methods to dynamic heterogeneous networks can be inappropriate, due to the involvement of different types of nodes and edges an...
Chapter
Full-text available
Information spread on networks can be efficiently modeled by considering three features: documents’ content, time of publication relative to other publications, and position of the spreader in the network. Most previous works model up to two of those jointly, or rely on heavily parametric approaches. Building on recent Dirichlet-Point processes lit...
Preprint
Full-text available
This paper proposes a random network model for blockchains, a distributed hierarchical data structure of blocks that has found several applications in various industries. The model is parametric on two probability distribution functions governing block production and communication delay, which are key to capture the complexity of the mechanism used...
Preprint
Full-text available
The speed of urban construction is accelerating, and the number of motor vehicles in the city is also growing rapidly, which has brought many serious traffic problems. Using modern technology to identify the routes of vehicles can effectively relieve the pressure of traffic travel. However, due to the huge number of urban population, it is difficul...
Article
Full-text available
Mediation analysis is increasingly used in the social sciences. Extension to social network data, however, has proved difficult because statistical network models are formulated at a lower level of analysis (the dyad) than many outcomes of interest. This study introduces a general approach for micro-macro mediation analysis in social networks. The...