Here I will share my journey through Generative Adversarial Networks also known as GAN. I tried to implement couple of GAN from scratch which boosted my understanding of neural networks.
I started from very simple GAN known as Fully Connected GAN, then updated it with Deep Convolutional GAN for better results. It followed others and will continue to do so. As we all know GANs are bit unstable in during training time, therefore, there are some changes in loss functions. These transactions between different loss functions and their intutions really amazed me. And while implementating, it was really an adventurous journey.
Throught the journey I have learnt KL, JS divergence and how they impact on the outcome and then EMD which is short form of Earth Mover Distance.