UTCS Faculty Candidate - J. Zico Kolter/CS and AILaboratory at MIT, "Learning, Inference, and Control for Sustainable Energy", ACES 2.302

Contact Name: 
Jenna Whitney
Date: 
Apr 12, 2012 11:00am - 12:00pm

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

http://apps.cs.utexas.edu/talkschedules/cgi/list_events.cgi

Type o

f Talk: Faculty Recruitment

Speaker/Affiliation: J. Zico Kolter/CS and
AI Laboratory at MIT

Talk Audience: UTCS Faculty, Graduate Students

, Undergraduate Students, and Outside Interested Parties

Date/Time: T

hursday, April 12, 2012, 11:00 am

Location: ACES 2.302

Host: J

. Zico Kolter

Talk Title: Learning, Inference, and Control for Susta

inable Energy

Abstract:
Sustainable energy issues pose one of the la

rgest challenges facing society: 84% of the world''s energy currently comes
from fossil fuels, raising major issues with climate change, energy secu

rity, and the long-term availability of these sources. Although energy do

mains span a huge range of different areas, a common theme in many modern

energy tasks is the availability of large amounts of data, and the need to
learn models, make inferences, and control the system based upon this da

ta. These are problems that require new methods in machine learning, prob

abilistic inference, and control, and where such algorithms can have a pr

ofound impact on the energy space. In this talk I will look at two particu

lar tasks spanning different extremes of energy consumption and generation

and show how new algorithmic methods can play a pivotal role in each.

First, on the energy consumption side, I will present new techniques for

energy disaggregation, the task of taking an aggregate power signal and de

composing it into separate devices. This ability helps us understand how en

ergy is consumed in a building, and studies have shown that just presentin

g this information to users can directly lead to large energy savings. Unl

ike previous approaches to this problem, my work considers models that loo

k jointly at the entire signal and exploit the rich temporal structure of t

he data. The key technical challenge here is the task of making inferences
in these high-dimensional, factorized, temporal models, and I will pres

ent new algorithms I have developed, based upon convex relaxations of infe

rence, that greatly outperform existing approaches on this task. Second,
on the energy generation side, I will present work on maximizing power ou

tput for wind turbines in low-wind conditions. In particular, I will pres

ent a novel policy learning approach, based upon trust-region optimization

, which is able to maximize power using much less data than existing learn

ing techniques. We demonstrate that the method produces 30% more power tha

n a purely model-based approach on an experimental wind turbine.

Bi

o:
J. Zico Kolter is a postdoctoral fellow in the Computer Science and Ar

tificial Intelligence Laboratory at MIT. He received his his Ph.D. in Comp

uter Science from Stanford University in 2010 and his B.S. from Georgetown

University in 2005. His research revolves around sustainable energy domain

s, with a focus on core learning, inference, and control tasks within th

is space. His work in this area include projects in energy disaggregation

, wind turbine control, and modeling building energy consumption. His pa

st work also looked at learning and control methods in other domains, incl

uding autonomous cars in extreme maneuvers, quadruped locomotion, and fea

ture selection in reinforcement learning. He is the recipient of an NSF Co

mputing Innovation Postdoctoral Fellowship, a former recipient of an NSF G

raduate Research Fellowship, and has received best paper awards at the SIG

KDD and AIAA Infotech@Aerospace conferences.