Value-Based Methods: Q-Learning to DQN

From tabular TD learning to Deep Q-Networks — and why the combination almost doesn't work

Learn temporal-difference learning, tabular Q-learning and its off-policy Bellman update, why tables collapse at scale, function approximation with neural networks, and the experience replay, target network and double-Q fixes that make Deep Q-Networks trainable — framed around the deadly triad of function approximation, bootstrapping and off-policy learning.

Practice questions (6)

• Trace a Manual Q-Learning Update for a Cloud Autoscaler

  Advanced · Free
  View →

• Migrate a Tabular Agent to Function Approximation

  Advanced
  View →

• Diagnose a Diverging Training Run

  Advanced
  View →

• Quantify and Fix Overestimation Bias in a Trading Agent

  Advanced
  View →

• Design and Debug a Replay Buffer for a Recommendation Agent

  Advanced
  View →

Quizzes (1)