Reinforcement Learning in Python — Temporal Difference Learning

Led by Geoffrey Hinton Simulacrum

About this course

Learning from incomplete episodes — TD(0) prediction, SARSA (on-policy control) and Q-Learning (off-policy control), the algorithms behind modern RL.

Drawing on the foundational research of Geoffrey Hinton in neural networks and learning algorithms.

1. Module 1

   TD(0) Prediction and SARSA

   Led by Geoffrey Hinton Simulacrum

   The question

   Temporal difference introduction · TD(0) prediction (one-step bootstrapping) · the TD update rule and comparison with MC · TD(0) prediction in code · bias-variance trade-off (TD vs MC) · SARSA (State-Action-Reward-State-Action) · on-policy TD control...

   Outcome

   Demonstrates understanding and implementation of td(0) prediction and sarsa.

   Sub-units

   1. ○ 1.1 TD(0) Prediction and SARSA Open this unit → ✓ ↻
   Begin the course → Print certificate of completion
2. Module 2

   Q-Learning

   Led by Geoffrey Hinton Simulacrum

   The question

   Q-Learning (off-policy TD control) · the Q-Learning update rule: Q(s,a) ← Q(s,a) + α[r + γ max Q(s',a') - Q(s,a)] · why Q-Learning is off-policy (learns about the greedy policy while following an exploratory policy) · Q-Learning in code · comparison ...

   Outcome

   Demonstrates understanding and implementation of q-learning.

   Sub-units

   1. ○ 2.2 Q-Learning Open this unit → ✓ ↻
   Open this module → Print certificate of completion