Lesson 305
Reinforcement Learning
Agents · Rewards · Q-Learning
1:00How an agent learns optimal behaviour from scratch by acting in an environment, collecting reward signals, and backing credit through the Bellman equation.
By the end, you can
- Describe the agent–environment loop and name what the environment returns after each action.
- Contrast reinforcement learning with supervised learning and explain the credit-assignment problem.
- Define state, action, reward, discount factor, policy, value function, and Q function in an MDP.
- Explain what the Markov property implies about the agent's memory requirements.
- Compute a discounted return given a sequence of rewards and a gamma value.
- Apply one step of the Q-learning update formula given Q, alpha, r, gamma, and max Q(s').
- Explain why Q-learning is model-free and off-policy.
- Describe how the Bellman equation propagates value backward from a goal.
- Explain what Deep RL adds and name at least two landmark applications.
Up next in AI, Machine Learning & Course Review
