GANs are probably one of the most interesting types of deep neural networks. Ever since GANs were first introduced by Goodfellow et al. in 2014, there have been more and more research projects and applications built around them, some of which are really intriguing. The following are the interesting ones we picked up on:
- Image generation, such as cat pictures, fake celebrity faces, and even modern artwork
- Audio or video synthesis, for instance, DeepMind's WaveNets, which are able to generate human speech and 3D reconstruction for astronomical images
- Time-series generation, such as medical data and high-frequency data for stock-market prediction
- Statistical inference, such as Amazon's algorithm that designs clothing by analyzing a bunch of pictures
So, why are GANs so powerful? When it comes to GANs, it is better to first talk about the...
