Reinforcement Learning (RL)
Working knowledgeTechniques and Architectures
Also called: RL
Reinforcement learning is a way of training a system by trial and error, rewarding the actions that work and penalising the ones that do not, until it learns a sequence of decisions that scores well. It suits problems where the right answer is a strategy rather than a label, such as routing a complex supply chain, navigating a robot, or running an automated trading system.
In practice
Reinforcement learning needs a reward you can define and measure, and that constraint decides whether it is usable on your problem at all. Where the objective is fuzzy, the system optimises the proxy you wrote down rather than the outcome you wanted, so the design of the reward is the risk, not the algorithm.