M. Afzal Upal
Ottawa Hills, OH
Phone: 416-***-****
Email: abg2n3@r.postjobfree.com
Education
Ph.D. (Computer Science)
MSc. (Computer Science)
B.Sc. (Computer Science)
University of Alberta, Edmonton
Thesis: Learning to improve plan quality.
Supervisor: Dr. Renee' Elio
University of Saskatchewan, Saskatoon
Thesis: Monte Carlo Comparison of Non-hierarchical unsupervised classifiers
Supervisor: Dr. Eric Neufeld
University of Saskatchewan, Saskatoon
2000
1996
1994
Professional Appointments
Computer Scientist
Assistant Professor
Assistant Professor
Computer Scientist
Assistant Professor
Graduate Teaching Assistant
Software Engineer
Defence R&D Canada, Toronto
Occidental College (oxy.edu), Los Angeles, CA
Electrical Engineering and Computer Science Department, University of
Toledo (utoledo.edu)
Information Extraction and Transport Inc. (iet.com) Arlington, VA
Faculty of Computer Science
Dalhousie University (dal.ca), Halifax, Canada
Department of Computing Science
University of Alberta, Edmonton, Canada
Product Direct Network, Saskatoon, Canada
8/2008 to present
6/2007 to 4/2009
7/2003 to 5/2007
3/2001 to 12/2002
7/1999 to 3/2001
9/1995 to 6/1994
1/1995 to 8/1995
6. Courses Taught
Occidental College, Cognitive Science Program
COGS 295: Topics in Cognitive Science (Spring 2008)
COGS 110: Models of Cognition (Spring 2008)
Cultural Studies Program Section 28: Computation, Cognition, & Culture
(Fall 2007)
COGS 101: Introduction to Cognitive Science (Fall 2007, Spring 2008)
University of Toledo Electrical Engineering & Computer Science (EECS)
Department
EECS 1570: Linear Data Structures (Spring 2006)
EECS 1580: Non-linear Data Structures (Spring 2007)
EECS 2100: Computer Organization and Assembly Language Programming (Fall
2003, Fall 2004, Fall 2005, Fall 2006)
EECS 4000: Senior Design (Spring 2005)
EECS 4580/5580/7580: Survey of Artificial Intelligence (Spring 2004)
EECS 4740/5740: Advanced Artificial Intelligence (Spring 2006, Spring 2007)
EECS 5930: Electrical Engineering & Computer Science Graduate Seminar (Fall
2006)
Dalhousie University Faculty of Computer Science (CSCI)
CSCI 3134: Functional & Logic Programming (Spring 2000, Summer 2000)
CSCI 3136: Principles of Programming Languages (Spring 2000, Summer 2000,
Fall 2000)
CSCI 4163: Human Computer Interaction (Fall 1999)
CSCI 6504: Advanced Topics in Software Agents (Fall 2000)
University of Saskatchewan Department of Computer Science
CMPT 100: Introduction to Computer Science for non-majors (Winter 1995)
7. Programming Experience
7.1 Information Extraction & Transport (IET) Inc. Projects
Ultra*Log: As a part of this DARPA sponsored project, the IET team
working under my leadership
successfully extended the world's largest agent-based planning, execution,
and replanning system.
Automating military logistics is a challenging planning problem because of
(a) the millions of different
object types and thousands of heterogeneous interacting organizational
units involved, (b) the complex
continual interplay between planning, execution, and replanning, and (c)
stringent performance
requirements. Instead of focusing on traditional domain independent AI
planning techniques, DARPA's
Advanced Logistics Planning Project (ALP) modeled the military units
involved in military logistics
planning using the agent-based computing approach known as Cogntive Agent
Architecture or Cougaar1
for short. Cougaar is able to compute a detailed Level 5 military logistics
plan given an Oplan within an
hour. One of the major limitations of earlier versions of Cougaar was its
inability to allow military
commanders to pose what-if queries (such as, "what would happen if these
vehicles were flown and not
driven") and compute alternative course(s) of action for the hypothetical
scenarios while retaining the
original course of action for comparison. It also did not dynamically
evaluate the effect of a change in
the state of the world on the quality of a plan. As the leader of IET's
Ultra*Log project team, I extended
the basic Cougaar architecture to (a) allow Cougaar users to pose what-if
queries (Upal 2003), (b) be able
to see the resulting alternative plans, (c) be able to determine the
probability of success and failure of
each plan, and (d) tradeoff the alternatives in a decision theoretic
framework (Upal & Fung 2003). Time
efficiency was our primary design concern followed by space efficiency.
Because of time efficiency,
designing a probabilistic reasoning solution that would allow probabilistic
information about chances of
success/failure of each plan to be propagated from the lowest level agents
to higher level agents was a
key challenge. I designed a message passing version of Judea Pearl's
polytree algorithm that conformed
to Ultra*Log architectural principles of asynchrony and distributed
processing. Space efficiency
concerns demanded a conservative approach towards copying Cougaar objects
for creating alternative
versions of "reality" for processing what-if queries. I designed an
approach called "internal-delta" which
only copied an object to a modified world when that object had to be
modified in the new world (Upal
2003). Otherwise, the objects from the unchanged world were allowed to be
inherited by the changed
what-if worlds. Our evaluation indicated that this approach was successful
in saving space in most
Ultra*Log planning scenarios (Upal & Fung 2003).
EELD: As part of DARPA's Evidence Extraction and Link Discovery project,
I led the design and
development of automated software for measuring the performance of link
discovery systems. The link
discovery task is different from traditional data-mining and classification
tasks which involve analyzing
data to discover and represent associations based on the aggregate
statistical characteristics of a sample
of instances drawn from some population. The link discovery task begins
with such association patterns
and sifts through various databases to find instances of a given pattern. I
analyzed similarities between
classification and the link discovery to design metrics for measuring the
performance of link discovery
systems. The link discovery system evaluation software that I implemented
was used to evaluate the
performance of link discovery systems being developed under the EELD
program.
RKF: As a member of DARPA's Rapid Knowledge Formation project I designed
challenge problems to
challenge the state of the art in knowledge acquisition.
7.2 Product Direct Network Projects
OPAL: OPAL was one of web's first electronic auction environments that
allowed companies in
Canada's oil patch to trade surplus equipment. An Informix SQL database
formed the backend of the
system. My job involved maintaining the Apache web-server and design and
develop software in
programming in Informix 4GL and Perl to build common gateway interfaces
(CGI).
3