Summary

In this chapter, we were introduced to RNNs and covered the major differences between the architectures of RNNs and FFNNs. We looked at BPTT and how weight matrices are updated. We learned how to use RNNs using Keras and solved a problem of author attribution using RNNs in Keras. We looked at the shortcomings of RNNs by looking at vanishing gradients and exploding gradients. In the next chapters, we will look into architectures that will address these issues.