Reinforcement Learning: Core Concepts and Algorithms

Reinforcement learning problem framing

Reinforcement learning (RL) problem framing models an agent interacting with a dynamic environment over discrete time steps. At each step, the agent observes a state, selects an action, and receives a reward based on the transition to a next state. The objective is to learn a policy (a rule for choosing actions given states) that maximizes expected cumulative reward, typically expressed as discounted return. A common formalization is the Markov decision process (MDP), which assumes the next state depends only on the current state and action (full observability). If the agent cannot directly observe the true state or observations are noisy, the problem is framed as a partially observable MDP (POMDP).

0/2 modes complete

Exploration–exploitation dilemma

0/2 modes complete