Reward Design & Delayed Credit Assignment

Shape rewards without breaking optimality, and assign credit correctly across long, sparse horizons

Learn potential-based reward shaping and why it provably preserves the optimal policy, how reward hacking and Goodhart's Law wreck real products, and the mechanics of credit assignment over long horizons — n-step returns and eligibility traces (TD(λ), forward and backward views) with worked numeric examples.

Practice questions (5)

• Audit a Proposed Shaping Reward for a Warehouse Robot

  Advanced · Free
  View →

• Diagnose and Fix an Outrage-Promoting Recommender

  Advanced
  View →

• A Notification Agent That Learned to Spam

  Advanced
  View →

• Choose an n for a Delayed-Reward Trading Agent

  Advanced
  View →

• Compute Eligibility Traces and Backward-View Updates by Hand

  Advanced
  View →

Quizzes (1)