This is the last stop of our DL architectures and new trends in DL journey. In this chapter, we learned that Bayesian deep learning combines the merits of both Bayesian learning and deep learning. It models uncertainty, which in a way tells us how much we trust the predictions. Capsule networks capture oriental and relative spatial relationships between objects. We believe they will become more mature and popular in the future.
Meta-learning, that is, learning to learn, is an exciting topic in the DL research community. We have implemented a meta-learning model, that is, Siamese Neural Networks with Keras, and applied it to a face recognition problem. In fact, there are many other interesting things going on in DL that are worth looking into, such as deep reinforcement learning, active learning, and automated machine learning. Are there any other new trends you noticed...
