Reinforcement Learning (RL) is a learning paradigm where an agent interacts with an environment over time, taking actions and receiving feedback in the form of rewards. The goal of the agent is to learn a policy — a strategy for choosing actions — that maximizes the total cumulative reward (also called the return).
Unlike supervised learning, RL doesn’t rely on labeled input-output pairs. Instead, the agent learns from experience, exploring the environment and using trial and error to improve its behavior. The feedback is often delayed, meaning the consequences of actions may not be known immediately, which makes learning more complex.
Reinforcement learning problems are often modeled as Markov Decision Processes (MDPs), where decisions depend on both the current state and action, and outcomes have probabilistic transitions. Key components in RL include states, actions, rewards, value functions, and policies.
RL has been successfully applied in areas like robotics, game playing (e.g. AlphaGo, OpenAI Five), recommendation systems, finance, and autonomous vehicles. It also serves as the basis for many cutting-edge approaches in artificial general intelligence (AGI). However, RL often requires large amounts of data and careful tuning, and remains an active area of research.