Summary

This was an interesting chapter, and I hope you enjoyed reading it as much as I enjoyed writing it. It's at present the hot topic of research. This chapter introduced generative models and their classification, namely implicit generative models and explicit generative models. The first generative model that was covered is VAEs; they're an explicit generative model and try to estimate the lower bound on the density function. The VAEs were implemented in TensorFlow and were used to generate handwritten digits.

This chapter then moved on to a more popular explicit generative model: GANs. The GAN architecture, especially how the discriminator network and generative network compete with each other, was explained. We implemented a GAN using TensorFlow for generating handwritten digits. This chapter then moved on to the more successful variation of GAN: the DCGAN. We implemented a DCGAN to generate celebrity images. This chapter also covered the architecture details of CycleGAN, a recently...