BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Talks.cam//talks.cam.ac.uk//
X-WR-CALNAME:Talks.cam
BEGIN:VEVENT
SUMMARY:Wasserstein Natural Gradients for Reinforcement Learning - Ferenc 
 Huszár
DTSTART:20201201T131500Z
DTEND:20201201T141500Z
UID:TALK152023@talks.cam.ac.uk
CONTACT:Mateja Jamnik
DESCRIPTION:"Join us on Zoom":https://zoom.us/j/99166955895?pwd=SzI0M3pMVE
 kvNmw3Q0dqNDVRalZvdz09\n\nPolicy Gradient methods can learn complex behavi
 ours in difficult reinforcement learning tasks but often struggle with dat
 a-inefficiency: they make slow progress requiring frequent rollouts or sim
 ulations of the environment. A key to speeding these methods up is to inco
 rporate the information geometry of policies into the optimisation. This c
 an be done via trust regions (TRPO)\, additive penalties (PPO)\, or via na
 tural gradients.\n\nIn this talk I present new optimization approach which
  can be applied to policy optimisation as well as evolution strategies for
  reinforcement learning. The procedure uses a computationally efficient Wa
 sserstein natural gradient (WNG) descent that takes advantage of the geome
 try induced by a Wasserstein penalty to speed optimization. I will illustr
 ate the differences between different natual gradient descent schemes and 
 discuss experiments on challenging tasks which demonstrate improvements in
  both computational cost and performance over advanced baselines.\n\nThis 
 talk is largely based on "https://arxiv.org/abs/2010.05380":https://arxiv.
 org/abs/2010.05380\n
LOCATION:Zoom
END:VEVENT
END:VCALENDAR