So far in our journey, we have been dealing with simple regression and classification tasks. We regressed observations against continuous values (that is, when predicting the stock market) and classified features into categorical labels (while conducting sentiment analysis). These are two cornerstone activities pertaining to supervised ML. We showed a specific target label for each observation our network comes across while training. Later on in this book, we will cover some unsupervised learning techniques with neural networks by using Generative Adversarial Networks (GANs) and autoencoders. Today, however, we employ neural networks to something quite different from these two caveats of learning. This caveat of learning can be named reinforcement learning.

Reinforcement learning is noticeably distinct from the aforementioned variations...