UTCS Colloquium/AI - Marc Deisenroth/University of Washington: "Probabilistic Inference and Learning for Control", ACES 2.402

Contact Name: 
Jenna Whitney
Jan 28, 2011 2:00pm - 3:00pm

There is a sign-up schedule for this event that can be found at



Type of Talk: UTCS Colloquium/AI

Speaker/Affiliation: Marc

Deisenroth/University of Washington

Date/Time: Friday, January 28, 2

011 2:00 p.m.

Location: ACES 2.402

Host: Peter Stone

Talk Ti

tle: "Probabilistic Inference and Learning for Control"

Talk Abstract:

We propose PILCO, a data-efficient and fully probabilistic model-based

framework for autonomously learning transition dynamics and controllers

nin the absence of expert knowledge. In most autonomous learning

s either task-specific domain knowledge and/or many trials are
required t

o learn a task. In practical applications, however, full
knowledge abou

t the underlying dynamics or thousands of trials might be

ractical. PILCO learns a probabilistic dynamics model
from data only. By

representing and incorporating model uncertainty into
the decision making
process, PILCO reduces model bias and fully
automatically learns to sol

ve fairly complicated control problems in
only a few trials. Across multi

ple complicated control tasks, PILCO
achieves an unprecedented degree of
automation and an unprecedented
speed of learning.

Speaker Bio:

arc Peter Deisenroth received a Dr.-Ing. degree from the Karlsruhe

ute of Technology in 2009.
Since 2010 he has been a postdoctoral research

er in the Robotics and
State Estimation Lab, University of Washington,

Seattle. He is an
adjunct researcher at Intel Labs Seattle and the CBL La

b, University of
Cambridge (UK). From 2006 to 2009, he was a researcher
at the Max Planck
Institute for Biological Cybernetics in Tuebingen (Ger

many) and at the
University of Cambridge (UK). His research focuses on B

inference, machine learning, robotics, and control.