Reinforcement Learning and Agents: From Bandits to Policy Gradients

<p>Follow Priya, an ML engineer at a scrappy ad-tech startup, as she builds a reinforcement learning system piece by piece: starting with multi-armed bandits for ad creative selection, extending to Markov Decision Processes and value iteration for sequential ad bidding, moving to model-free Q-learning for live auctions, and finally implementing policy gradients (REINFORCE) from scratch in numpy to handle continuous bids. Along the way you will build gym-style environments, an evaluation harness, off-policy evaluation, and learn to spot reward hacking before it costs real money.</p>

Tools: Reinforcement Learning, Multi-Armed Bandits, Markov Decision Processes, Q-Learning, Policy Gradients, Python