Transcript Michael Jordan - University of California, Berkeley

Decision-Making under Uncertainty
http://www.cs.berkeley.edu/projects/muri
University of California, Berkeley
Venkat Anantharam
Laurent El Ghaoui
Michael I. Jordan
Stuart Russell
S. Shankar Sastry
University of California, Davis
Roger Wets
Stanford University
Daphne Koller
Ben van Roy
Claire Tomlin
June 20, 2001
MURI Review Meeting
Faculty Investigators









Venkat Anantharam – information theory, stochastic control
Laurent El Ghaoui – convex optimization, control theory
Michael Jordan – graphical models, statistics
Daphne Koller – decision theory, game theory, AI
Stuart Russell –reinforcement learning, decision theory, AI
Shankar Sastry – hybrid control, systems theory
Claire Tomlin – hybrid control, systems theory
Ben van Roy – reinforcement learning, dynamic programming
Roger Wets – variational analysis, stochastic optimization
June 20, 2001
MURI Review Meeting
Postdoctoral Researchers
 Sekhar Tatikonda – information theory, distributed control
 Yair Weiss – probabilistic inference
June 20, 2001
MURI Review Meeting
Research Concentration Areas
 (RCA 1) Exploration of basic methods, such as probability and
decision theory, Bayesian and related methods, and utility theory
 (RCA 2) Bound the effects of missing or incorrect information
 (RCA 3) Incorporate Quality of Service issues, such as trading off
time for certainty
 (RCA 4) Dealing with varying bandwidth and connectivity in the
networked battlefield
 (RCA 5) Fusing uncertain information of different kinds
 (RCA 6) Ways of structuring problems for optimal understanding
 (RCA 7) Present possible and alternative courses of action to the
user, quantitatively ranked according to the system’s beliefs
June 20, 2001
MURI Review Meeting
Investigator to RCA Map









Venkat Anantharam – RCA1, RCA4
Laurent El Ghaoui – RCA1, RCA2, RCA3
Michael Jordan – RCA1, RCA2, RCA5, RCA7
Daphne Koller – RCA1, RCA2, RCA3, RCA5, RCA6, RCA7
Stuart Russell – RCA1, RCA3, RCA5, RCA6, RCA7
Shankar Sastry – RCA1, RCA2, RCA6
Claire Tomlin – RCA1, RCA2, RCA6
Ben van Roy – RCA1, RCA2, RCA3
Roger Wets – RCA1, RCA2
 Sekhar Tatikonda – RCA1, RCA4
 Yair Weiss – RCA1, RCA2, RCA3
June 20, 2001
MURI Review Meeting
Research Spectrum
Physical
environment
Information
environment
June 20, 2001
Algorithms
and
Interfaces
MURI Review Meeting
Human
decision
maker
Industrial Contacts








Intel (Gary Bradski)
Google (Peter Norvig)
Microsoft Research (David Heckerman, Eric Horvitz, Jack Breese)
AT&T Labs Research (Michael Kearns, Yann LeCun, Satinder Singh)
Affymetrix (David Kulp, Cyrus Harmon)
Alphatech (John Fox)
IBM Watson Research Center (Shivakumar Vaithyanathan)
Honeywell Labs (Datta Godbole, Mary Jo Hoffman, Jorge Tierno)
June 20, 2001
MURI Review Meeting
DoD/Government Contacts











SPAWAR, San Diego, CA
Third Fleet Command Ship, USS Coronado
Army Research Laboratories
SPAWAR, Norfolk, VA
Fort Leavenworth (TRADOC)
AFRL (Siva Banda)
AFOSR (Marc Jacobs)
NASA (George Meyer)
NASA (Asaf Degani)
Darpa ITO (John Bay)
Darpa DSO (Eric Eisenstadt)
June 20, 2001
MURI Review Meeting
Educational Activities
Courses specifically designed for relevance to MURI








Advanced Topics in Learning and Decision-Making--Berkeley
Statistical Learning Theory---Berkeley
Introduction to Convex Optimization---Berkeley
Optimization Seminar---Berkeley
Neuro-Dynamic Programming and Reinforcement Learning--Stanford
Stochastic Decision Models---Stanford
Analysis and Control of Nonlinear Systems---Stanford
Stochastic Programming---Davis
June 20, 2001
MURI Review Meeting
Meetings



Biweekly research meetings at Berkeley
Full-day workshop---September 20, 2000
(Berkeley)
Full-day workshop---March 28, 2001 (Stanford)
June 20, 2001
MURI Review Meeting
Agenda












9:00 – 9:15 Michael Jordan (overview)
9:15 – 9:55 Shankar Sastry – Pursuit-evasion games, map building,
embedded software
9:55 – 10:35 Daphne Koller – Markov decision problems with factored value
functions, multi-agent Markov decision problems
10:35 – 11:15 Stuart Russell – Programmable reinforcement learning systems
11:15 – 1:00 Lunch
1:00 – 1:40 Claire Tomlin – Distributed control of multiple vehicle systems
1:40 – 2:20 Michael Jordan – Feature selection, edge selection and variable
selection
2:20 – 3:10 Yair Weiss – Approximate inference via approximations of free
energy
3:10 – 3:30 Break
3:30 – 4:10 Laurent El Ghaoui – Estimation and optimization: Gaps and
bridges
4:10 – 4:30 Sekhar Tatikonda – Network information theory
4:30 – 5:00 Open discussion
June 20, 2001
MURI Review Meeting