Reinforcement Learning
Learning by doing — agents, rewards, and the long game
How machines learn from trial and error — from Q-learning to AlphaGo to the RL that trained ChatGPT.
Advanced3 modules · 10 lessons~87 min900 XPAligned with Hugging Face Deep RL CourseStart with: ML FoundationsStart with: Probability & Statistics
Progress0/10
Module
The RL Loop
Agents, rewards, and the problem of acting
Agents, Environments & Rewards8 min · 80 XPAgents, Environments & Rewards
Markov Decision Processes9 min · 85 XPMarkov Decision Processes
Exploration vs. Exploitation8 min · 85 XPExploration vs. Exploitation
Module
Value-Based Methods
Learn what actions are worth — the policy comes free
Q-Learning: What Is an Action Worth?9 min · 90 XPQ-Learning: What Is an Action Worth?
Deep Q-Networks9 min · 95 XPDeep Q-Networks
The Tricks That Made It Work8 min · 90 XPThe Tricks That Made It Work
Module
Policy & Modern RL
Learn the policy directly — and teach language models to behave
Policy Gradients9 min · 90 XPPolicy Gradients
PPO: The Workhorse8 min · 90 XPPPO: The Workhorse
RLHF: RL Meets Language Models10 min · 100 XPRLHF: RL Meets Language Models
Where RL Actually Works9 min · 95 XPWhere RL Actually Works
Capstone project
Train an agent
Train an agent on a classic environment (CartPole, FrozenLake, or a simple grid world). Tune the reward or hyperparameters once and document what changed.
Deliverable: Training code, a learning-curve plot, and a paragraph on what the reward taught — and didn't.
Milestones
Finish every lesson to earn the Reinforcement Learning Certificate.