Reinforcement Learning

← Back to Learning Types

An agent learns to make decisions by interacting with an environment and receiving reward signals. Used in robotics, game playing, and LLM alignment (RLHF).

Key Properties

• States Actions and Rewards
• RLHF

Core Components

• Agent — the learner/decision maker
• Environment — what the agent interacts with
• State — current situation
• Action — what the agent can do
• Reward — feedback signal
• Policy — strategy mapping states to actions

Key Algorithms

• Q-Learning, Deep Q-Networks (DQN)
• Policy Gradient methods (REINFORCE, PPO, A3C)

Related

• Supervised Learning (contrast: no explicit reward signal)

---

ml reinforcement-learning rlhf