Machine Learning in Elixir

Learning from Experience

One of the largest subfields in machine learning this book didn’t cover is reinforcement learning. Reinforcement learning is concerned with creating agents capable of making intelligent actions and learning from reward signals in an environment. Reinforcement learning is fundamentally different from both supervised and unsupervised learning. In a reinforcement learning problem, there’s no concept of labeled or unlabeled data. Instead, the learning process is entirely online as an agent interacts with its environment.

The most common training ground for reinforcement learning agents is video games. For example, the Arcade Learning Environment^[48] is an environment designed to allow researchers and hobbyists to create ...