Amy Greenwald is a computer scientist whose work focuses on
artificial intelligence, specifically multi-agent interactions and game theory.
She did her undergraduate work in a dual degree program at the University of
Pennsylvania, where she studied both computer science and economics, which, in
essence, remain her focus. But, says Dr. Greenwald, "I can't say that early on
I knew that was exactly what I wanted to do. Computers and economics made for a
great program and a great opportunity, so I took advantage of it, but my career
zigzagged around a little bit before I came back to it."
That zigzagging took her
first to Oxford, where she had won a scholarship. With the focus on computer
theory at Oxford, Dr. Greenwald says that she "veered off from economics and
computer science into logic and computer science." She got a master's degree at
Oxford and returned to the United States to start her Ph.D. at Cornell. Because
Cornell has a very strong group in logic and computer science, she stayed with
that focus. After a few years, she realized that combination wasn't what she
wanted, and in 1995 she left Cornell and went to New York City.
While interviewing for jobs and considering how best to
continue her doctoral studies, Dr. Greenwald sat in on a class at City
University of New York because one of her advisors at Cornell had recommended
the professor, Dr. Rohit Parikh. "Of all the things I did in that period," says
Dr. Greenwald, "I liked the courses at City University the best." Although she
decided to go to New York University (NYU) for her Ph.D., Dr. Parikh served on
her thesis committee, and the collaboration between the two prospered.
The NYU computer science building happened to be right
across the street from the business school. "I was just starting out at NYU,"
says Dr. Greenwald, "and I was looking for a new thesis topic. I knew I was
going to do something with economics and computer science. One day I happened
to sit in on a game theory class at the business school, and that completely
changed my career interest."
Since then, Dr. Greenwald has been doing research on
computer science and game theory. She worked as a postdoc on the KDI project
Automated Learning in Network Traffic Control, along with her advisor
from NYU, Dr. Bhubaneswar Mishra (the project's principal investigator) and Dr.
Parikh (co-principal investigator), among others.
For the project, Dr. Greenwald did work on resource
allocation. The team started with a problem called the "Santa Fe bar problem,"
which assumes that there is a bar in Santa Fe that has live music on Thursday
nights. The bar seats 60 people, but every Thursday night 100 people want to
go. The problem is to figure out, on any given Thursday night, whether to
goand risk finding out there's not enough roomor stay home, only to learn
that there were plenty of seats and then wish you had gone. The team modeled
the program game theoretically, and eventually, using low-rationality
algorithms, they were able find a way for a different set of 60 people to go to
the bar each time. Dr. Greenwald says, "We were viewing this just like sending
packets along a network link. It's a similar problem. It's as if you wanted,
for example, to send 100 packets and only had capacity for 60."
Today an assistant professor in computer science at Brown
University, Dr. Greenwald continues to focus on game theory. She is actively
involved in an international forum called Trading Agent Competition (TAC),
which promotes research into the trading agent problem. In TAC Classic, a
travel agent must put together a travel package for clients that includes
everything the clients want (flights, hotels, etc.), but each component is sold
separately in simultaneous auctions. "There's a lot of machine learning in this
game," says Dr. Greenwald, "because we're trying to make predictions about what
prices will be, and in particular we're trying to predict the behavior of the
other agents in the game." In the newest version of the game, called TAC SCM
(Supply Chain Management), agents must bid to sell their products, while at the
same time getting all the components they need and predicting prices. This set
of steps duplicates many of the challenges inherent in supporting effective
supply chain practices. For more information on this work, visit the TAC Web
site at http://www.sics.se/tac
"This is a very practical and very relevant problem, and a
very, very hard one," says Dr. Greenwald. This research can be used by any
company that needs to figure out its procurement schedule, as well as how it's
going to put together their components, when to sell them, and what the price
To learn more about Dr. Greenwald's work, visit her Web site
Back to Top of