Graph generative adversarial network

In this report, we briefly explain the building blocks of Generative Adversarial Network (GAN), recent research on generalization of Convolution Neural Network (CNN) to graphs, and experimented on further usage of graph convolution on other types of model. We also proposed a simple method to upsampl...

全面介紹

Saved in:
書目詳細資料
主要作者: Tjeng, Stefan Setyadi
其他作者: Xavier Bresson
格式: Final Year Project
語言:English
出版: 2018
主題:
在線閱讀:http://hdl.handle.net/10356/73993
標簽: 添加標簽
沒有標簽, 成為第一個標記此記錄!
實物特徵
總結:In this report, we briefly explain the building blocks of Generative Adversarial Network (GAN), recent research on generalization of Convolution Neural Network (CNN) to graphs, and experimented on further usage of graph convolution on other types of model. We also proposed a simple method to upsample graphs. Experiments include usage of graph convolution on Variational Autoencoder (VAE) and GAN. Comparison of result between traditional and proposed graph VAE are observed showing graph VAE achieving better performance. An experimentation on Graph GAN shows the model is unable to converge. Problem analysis and idea for improvement on graph convolution and upsampling are given.