Reinforcement Learning
Agents, rewards, and policies — from Markov Decision Processes to Deep Q-Networks and Policy Gradient.
7 lessons
01Markov Decision Processes
The formal language of RL — states, actions, rewards, and the Bellman equation that ties them together.
Q-Learning
Learning optimal action-values from experience alone — temporal difference, off-policy updates, and exploration.
Deep Q-Networks
Scaling Q-learning with neural networks — and the two tricks (replay, target net) that made it stable.
Policy Gradient & Actor-Critic
Optimizing the policy directly — REINFORCE, the variance problem, and the actor-critic fix.
Exploration Strategies & Model-Based RL
How agents balance trying new actions vs. exploiting what they know — and how world models enable sample-efficient learning.
From Policy Gradient to RLHF (PPO)
How vanilla policy gradient becomes PPO — and how PPO becomes RLHF the moment the policy is a language model.
Reinforcement Learning Quiz
Test your understanding of MDPs, Q-learning, DQN, policy gradients, exploration, model-based RL, and PPO/RLHF.