We just accomplished an important learning journey of DL architectures with restricted Boltzmann machines and autoencoders! Throughout this chapter, we got more familiar with RBMs and their variants. We started with what RBMs are, the evolution paths of RBMs, and how they become the state-of-the-art solutions to recommendation systems. We implemented RBMs in TensorFlow from scratch and built an RBM-based movie recommender. Beyond a shallow architecture, we explored a stacked version of RBMs called deep belief networks and employed it in image classification, which was implemented in TensorFlow from scratch.
Learning autoencoders is the second half of the journey, as they share similar ideas of finding latent representation of the input by input data reconstruction. After discussing what autoencoders are and talking about their evolution path, we illustrated a variety of...
