Embedded Files
(1) Bandit 1 training visualization
Blue: Reinforce, Orange: Adam, Green: Our method.
The action distribution is plotted and normalized for visualization.
(2) Double pendulum swing-up: visualize trajectory distribution
Reinforce
Adam
Our method
(3) Landscape of expected reward after gaussian filter
The global optimum is marked with the red dotted line.
Page updated
Google Sites
Report abuse