[Lisa_seminaires] [Tea Talk] Glen Berseth (UBC) July 20 2018 10:30 AA3195

Michael Noukhovitch mnoukhov at gmail.com
Mar 17 Juil 06:00:15 EDT 2018


This week we have *Glen Berseth * from * UBC*, who will be interning with
Chris Pal, giving a talk on *July 20 2018* at *10:30* in room *AA3195*

Will this talk be streamed <https://mila.bluejeans.com/1365483656/webrtc>?
Yes!
Want to meet with Glen? Glen is only available until 2pm on Friday, so send
him an email directly (gberseth at cs.ubc.ca) and perhaps invite him to lunch
:)

Given how many tea talks we've had, it should be a breeze to find your way
by moving your legs and walking to come to this tea talk!
Michael

*P.S.* We still don't have a speaker for next week's talk, so if you're
interested in giving a talk please slack or email me!

*TITLE* Scalable Deep Reinforcement Learning for Physics-Based Motion
Control

*KEYWORDS *hierarchical RL,  meta-learning, multi-task learning

*ABSTRACT*
Motion control in physics-based animation is challenging due to complex
dynamics and discontinuous contacts. Many previous control methods that
produce walking motions are very stiff, only work in particular
environments and require significant manual tuning to get functioning. In
this work, we progress the state-of-the-art in physics-based character
animation in a number of directions using machine learning methods. We
present three contributions that build upon the current research on motion
control using deepRL. First, we show that decomposing tasks into a
hierarchy increase learning efficiency by operating across multiple time
scales on a complex locomotion and navigation task. Second, we investigate
improved action exploration methods to sample more promising actions on
robots and in simulation using forward dynamics distributions. This
sampling strategy has been shown to improve sample efficiency for a number
of problems, including many from the OpenAI Gym. Last, we consider a new
algorithm to progressively learn and integrate new skills producing a
robust and multi-skilled physics-based controller. This algorithm combines
the skills of experts together and then applies transfer learning methods
to initialize and accelerate the learning of new skills.

*BIO* Glen Berseth is a PhD candidate at the Department of Computer Science
at the University of British Columbia where he works on reinforcement
learning, machine learning and motion planning. He received the BSc degree
in Computer Science from York University in 2012 and his MSc from York
University under the supervision of Petros Faloutsos in 2014 for
optimization and authoring crowd simulations. He has published in a wide
range of areas including computer animation, machine learning and robotics
and is an NSERC scholarship award holder.
-------------- section suivante --------------
Une pièce jointe HTML a été nettoyée...
URL: <http://mailman.iro.umontreal.ca/pipermail/lisa_seminaires/attachments/20180717/5ebde20b/attachment.html>


Plus d'informations sur la liste de diffusion Lisa_seminaires