Introduction to Reinforcement Learning

These flashcards cover key terms and concepts from the lecture on Reinforcement Learning, focusing on definitions, challenges, and components of RL.

Reinforcement Learning (RL)

The science of decision making, merging statistical learning with optimal control.

Agent

An entity that receives observations, executes actions, and collects rewards in a reinforcement learning framework.

Reward signal

A scalar signal quantifying the quality of an action, crucial in guiding the agent's decisions.

Policy

A map from state to action, which can be deterministic or stochastic, guiding the agent's behavior.

Value function

A prediction of future cumulative rewards, evaluating the quality of a policy for each state and action.

Markov state

A state is considered Markov if the future state depends solely on the current state and action, not on past states.

Exploration-exploitation trade-off

The dilemma of choosing between exploring new actions to gain more information and exploiting known actions that yield high rewards.

Full observability

A condition where the complete state of the environment can be reconstructed from a history of observations.

Partial observability

A condition where the environment's state cannot be fully reconstructed; only a proxy can be inferred from the history.

Data acquisition

The process of gathering information from the environment, which is vital for improving the agent's policy.