It's really cool work. Congrats! @RobertDadashi Regarding the part 1 paper https://arxiv.org/abs/1901.11524 in figure 11 b) perhaps the similar behavior of PI and natural PG in the polytope stems from connection between PI and Newton's method showed by @martypute and Brumelle (1979)
Our two-parter on viewing RL and representation learning from a geometric angle is out https://arxiv.org/abs/1901.11524 https://arxiv.org/abs/1901.11530 -- can't wait to see what new algorithms will come out of this! w/ @RobertDadashi @wwdabney & other fantastic collaborators
-
-
-
Thanks, & definitely! We know about the connection between the simplex method and PI (from
@martypute's book), but hadn't thought about the second order methods angle.
End of conversation
New conversation -
Loading seems to be taking a while.
Twitter may be over capacity or experiencing a momentary hiccup. Try again or visit Twitter Status for more information.