Summary

In this chapter, we learned about a whole new array of neural networks that have turned the artificial intelligence world upside down. Generative networks were always important to us, but we could not reach human-comparable accuracy with them until recently. Although there are a few successful generative network architectures, we have discussed only the two most popular networks in this chapter.

Generative networks use basic architectures like CNNs or RNNs as the building blocks of the overall network, but use some nice techniques to make sure the network is learning to generate some output. So far, generative networks have been widely used in art, and we could easily predict that generative networks will become the foundation of many sophisticated networks, since the model has to learn data distribution to generate output. Perhaps the most promising use of generative networks won't be generation but learning data distribution through generation and using that information...