Generative Adversarial Networks(GAN) in industry & academia

The industry and academia already have given a positive nod to the GAN architecture. The rise of GAN is a sure thing and in the coming time, the growth of GAN architecture is unavoidable. The outcome of computer-vision has completely changed with the inclusion of GAN; image classification, image inpainting, image to image translation, facial recognition have become much easier now, thanks to the GAN. The major challenge however with the GAN is that it needs huge image data sets and to process huge data with limited GPU/CPU is indeed a real challenge today. GAN can be simplified into two parts: the generator network and the discriminator network. Most of the time, convolutional neural networks(CNN) are chosen to accommodate generator and discriminator. Therefore, the limitations of conventional convolution layers, pooling layers, flattening, and dense layers are also equally applied to GAN. One can visualize the basic architecture of GAN in the below figure.

Recently, generative adversarial networks have adopted by almost all potential AI companies. The following are the major applications of GAN(Ref Wiki) 

1. Fashion, art, and advertising 
2. Science 
3. Video games 
4. Concerns about malicious applications 
5. Miscellaneous applications 

~AISavvy,2020.