In the first seven chapters of this book, deep neural networks with varied architectures have demonstrated their ability to learn from image, text, and transactional data. Even though deep learning has been developing rapidly over recent years, its evolution doesn't seem to be decelerating anytime soon. We are seeing new deep learning architectures being proposed almost every month, and new solutions becoming state-of-the-art every now and then. Hence, in this last chapter, we would like to talk about a few ideas in deep learning that we found to be impactful this year and that should be more prominent in the future.
In this chapter, we will look at the following topics:
- Bayesian neural networks
- Limitation of deep learning models
- Implementation of Bayesian neural networks
- Capsule networks
- Limitation of convolutional neural network (CNNs) ...
