Beginning Deep Learning with TensorFlow: Work with Keras, MNIST Data Sets, and Advanced Neural Networks

© The Author(s), under exclusive license to APress Media, LLC, part of Springer Nature 2022

L. Long, X. ZengBeginning Deep Learning with TensorFlowhttps://doi.org/10.1007/978-1-4842-7915-1_13

13. Generative Adversarial Networks

Xiangming Zeng¹   and Liangqu Long²

(1)

State College, PA, USA

(2)

Shenzhen, Guangdong, China

 

What I cannot create, I have not yet fully understood.

—Richard Feynman

Before the invention of the generative adversarial network (GAN), the variational autoencoder was considered to be theoretically complete and simple to implement. It is very stable when trained using neural networks, and the resulting images are more approximate, but the human eyes can still easily distinguish real pictures and machine-generated pictures.

In 2014, ...