Reinforcement Learning
Reinforcement learning is a field of machine learning in which an agent learns by interacting with an environment and receiving rewards or penalties for its actions. Rather than learning from labeled examples, the agent experiments, sees the results, and gradually adopts the strategy that maximizes its total reward over time.
The core loop is simple: the agent observes the current state, chooses an action, and receives a reward and a new state. Over many trials it learns a policy, a mapping from situations to actions, that balances exploring new actions with exploiting ones that have worked before.
Classic examples include agents that learn to play video games to get the most points, or systems that choose which advertisement to show a given customer. In industrial settings, reinforcement learning is used to tune control settings or robot motions, usually trained first in simulation.
Reinforcement learning differs from supervised learning, which learns from examples labeled with the right answer, and from unsupervised learning, which finds structure in unlabeled data. Reinforcement learning from human feedback (RLHF) applies the idea to align language models with human preferences.