CS885 Lecture 7a: Policy Gradient

[@SphereofTime]

26:15 how alphago uses montecarlotreesearch

[@user-or7ji5hv8y]

wow, thanks for sharing the lecture

[@Shikhar_]

The AlphaGo example was so neat!

[@MrNgovanmao]

minute 21:13, the line before stochastic gradient should be E[\gamma^n G_n \frac{
abla pi_\theta(A_n|S_n)}{ pi_\theta(A_n|S_n)}]

[@hemanthsharma5630]

Thanks for the great lecture prof. :)

[@shanggao781]

Thanks Prof Poupart.

[@tuliomoreira7494]

Amazing lecture. Does anyone know why at 12:14 we are discounting twice?

[@yuxiang3147]

It's not discounting twice, the first gamma^n is used to represent the state/reward distribution after executing the policy for n steps, the second gamma^t is used to represent the future expected reward.

[@НиколайНовичков-е1э]

Thank you!

[@lordjared2572]

35:34 Resign is just the same as calling gg. :)))