Is supervised learning a subset of reinforcement learning? It seems like the definition of supervised learning is a subset of reinforcement learning, with a particular type of reward function that is based on labelled data (as opposed to other information in the environment). Is this an accurate portrayal?
 A: It's true that any supervised learning problem can be cast as an equivalent reinforcement learning problem: Let states correspond to the input data. Let actions correspond to predictions of the output. Define reward as the negative of the loss function used for supervised learning. Maximize expected reward. In contrast, reinforcement learning problems can't generally be cast as supervised learning problems. So, from this perspective, supervised learning problems are a subset of reinforcement learning problems.
But, trying to solve a supervised learning problem using a general reinforcement learning algorithm would be rather pointless; all this does is throw away structure that would have made the problem easier to solve. Various issues arise in reinforcement learning that aren't relevant to supervised learning. And, supervised learning can benefit from approaches that don't apply in the general reinforcement learning setting. So, although there are some common underlying principles and shared techniques between the fields, one doesn't typically see supervised learning discussed as a type of reinforcement learning.
References
Barto and Dietterich (2004). Reinforcement learning and its relationship to supervised learning.
