9381fc93ad66f9ec4b2eef71147a6665-AuthorFeedback.pdf

The goal of the Atari experiment was to demonstrate33 that the same setup could work at scale in combination with SOTARL agents. Wehaveadditionally updated our main Atari result, which aggregate results across games,toalsoquantify theuncertainty. Our approach attempts to leverage the same trajectory data in a more effective way to improve the46 learningprocess.