Lets have some knowledge refresh through the core ideas that from the foundation in which an agent operates, to how it evaluates actions, learns from rewards and gradually discover the optimal strategy. Will explore advanced topics such Deep Q-Networks, Policy Gradient methods, or Agentic AI, … these concepts provide the essential building blocks for understanding how intelligent systems learn to make decisions.
The concept
| Environment | Defines the world in which the agent operates, including states, transitions, and rewards. |
| Agent | The decision-maker that learns the best policy through experience. |
| State | Represents the current situation. The Bellman Equation estimates the long-term value of being in a state. |
| Action | Represents the possible decisions. Q-Learning estimates the long-term value of taking each action in every state. |
| Reward | mmediate feedback used to evaluate decisions. Living penalties shape rewards to encourage efficiency, while TD Learning uses rewards to update value estimates. |