Reinforcement Learning
Reinforcement Learning (RL) is a machine learning paradigm where an agent learns to make decisions in an environment to maximize a reward signal.
Reinforcement Learning (RL) is the computational study of learning from interaction. Unlike supervised learning, where a model is given the 'correct' answer, an RL agent must discover which actions yield the highest reward by trying them. This involves a fundamental tradeoff between Exploration (trying new actions to find better ones) and Exploitation (using known actions that yield high rewards). RL is the primary technology behind autonomous robotics, game-winning AI (like AlphaGo), and the alignment of Large Language Models (RLHF).
graph LR
Center["Reinforcement Learning"]:::main
Pre_machine_learning["machine-learning"]:::pre --> Center
click Pre_machine_learning "/terms/machine-learning"
Rel_deep_learning["deep-learning"]:::related -.-> Center
click Rel_deep_learning "/terms/deep-learning"
Rel_game_theory["game-theory"]:::related -.-> Center
click Rel_game_theory "/terms/game-theory"
Rel_logistic_regression["logistic-regression"]:::related -.-> Center
click Rel_logistic_regression "/terms/logistic-regression"
classDef main fill:#7c3aed,stroke:#8b5cf6,stroke-width:2px,color:white,font-weight:bold,rx:5,ry:5;
classDef pre fill:#0f172a,stroke:#3b82f6,color:#94a3b8,rx:5,ry:5;
classDef child fill:#0f172a,stroke:#10b981,color:#94a3b8,rx:5,ry:5;
classDef related fill:#0f172a,stroke:#8b5cf6,stroke-dasharray: 5 5,color:#94a3b8,rx:5,ry:5;
linkStyle default stroke:#4b5563,stroke-width:2px;
🧒 Explain Like I'm 5
🎮 Training a computer program like a puppy: rewarding good behavior and ignoring bad behavior until it learns to be helpful.
🤓 Expert Deep Dive
## RLHF: Aligning Human and Machine
Reinforcement Learning from Human Feedback (RLHF) is the secret sauce behind modern chatbots like ChatGPT. Since it's impossible to write a mathematical formula for 'a good, helpful answer,' we show the model pairs of answers and let humans rank them. An RL agent is then trained to predict these rankings, creating a 'Reward Model' that guides the LLM toward safe and helpful output.