ConceptAI & ML
Discount factor in policy gradients
Option 1 is used in practice over Option 2 because omitting the gamma-to-the-t-minus-1 multiplier ensures the policy gradient is weighted correctly at every time step rather than favoring early steps.
Know first
Named as a prerequisite by the lectures that define this. The number is how many say so. A grey one has no page yet.
Discounted return1Policy gradient theorem1
Lecturers
1Connections
2The lectures that define it or assume it, and the concepts next to it.