Definition
Reinforcement learning (RL) is a method for training systems using feedback about the quality of their actions or outputs.
A simplified example is a model generating two possible responses. If one response is preferred, the training process can reinforce behavior associated with that outcome and penalize less desirable behavior. This creates a learning signal that encourages the model toward better outcomes.
One important application is Reinforcement Learning from Human Feedback (RLHF), where human preferences provide the feedback used to improve model behavior.
The concept is related to behavioral reinforcement: actions associated with positive outcomes become more likely, while actions associated with negative outcomes become less desirable.