Carnegie Mellon University
Department of Computer Science
AI Seminar 00/01 Schedule
Time: Tuesday 3:30-4:30pm
Place: Wean Hall 5409
The AI seminars are open to the public and will
be held on most Tuesdays at 3:30pm in Wean Hall, Carnegie Mellon University.
Special AI seminars can be arranged for visitors on dates other than Tuesdays.
The schedule will be updated daily. To volunteer to give an AI seminar
or to nominate an outside speaker, contact Dr.
Tai Sing Lee at tai@cs.cmu.edu.
All AI seminars will be posted to both cboards
and bboards (cs & robotics). Meanwhile, if you would like to/not
to be reminded by email, please subscribe/unsubscribe ai-seminar-announcements
mailing list by sending email to Michael.Lewicki@cs.cmu.edu.
Simply put "Subscribe/Unsubsribe ai-seminar-announcements" in
the subject of your message.
09/19/00 - Daphne Koller Stanford
University
Probabilistic Relational Models: Representation,
Inference, & Learning
Bayesian networks are a compact and natural representation for
complex
probabilistic models. They use graphical notation to encode domain
structure: the direct probabilistic dependencies between variables
in the
domain. Bayesian networks have been applied successfully in a
variety of
applications; they can also be learned directly from data, allowing
us to
automatically extract the most significant correlations in the
domain.
Bayesian networks, however, are an unsuitable representation for
complex
domains involving many entities that interact with each other. In
the first
part of the the talk, I will describe probabilistic relational
models
(PRMs), which extend the language of Bayesian networks with the
expressive
power of object-relational languages. A PRM models the uncertainty
over the
attributes of objects in the domain and uncertainty over the
relations
between the objects. This language is capable of compactly
modeling domains
that are substantially more complex than ones for which standard
Bayesian
networks are appropriate. The same structure that allows compact
representation can also support substantially faster inference than
in
equivalent Bayesian network models. In the second part of the
talk, I
will show how we can use probabilistic relational models to learn
the
probabilistic dependency structure in a relational domain, using a
relational database as our starting point. I will discuss
applications of
this new learning technology to various domains, including complex
biological data sets.
Joint work with: Nir Friedman, Lise Getoor, Avi Pfeffer, and Ben
Taskar.
Daphne
Koller received her PhD from Stanford University in 1994. After a
two-year postdoc at Berkeley, she returned to Stanford, where she is now
an Assistant Professor in the Computer Science Department. She has a broad
range of interests spanning artificial intelligence, economics, and theoretical
computer science. Her main research interest is in creating large-scale
systems that reason and act under uncertainty. The theme underlying her
work is the integration of ideas from decision theory and economics into
these systems. This task raises the need for compact and natural knowledge
representation schemes and for efficient inference and learning algorithms
that utilize these schemes. Daphne Koller is the author of over 45 refereed
publications, which have appeared in AI, theoretical computer science,
and economics venues. She is on the editorial board of the Journal of Artificial
Intelligence Research. She was awarded the Arthur L. Samuel Award for her
PhD thesis work and the Sloan Foundation Faculty Fellowship in 1996.
02/27/01 - Jeffrey Mark Siskind
NEC Research Institute, Inc.
Grounding the Lexical Semantics of Verbs in Visual Perception
using Force Dynamics and Event Logic
In this talk, I will present an implemented system, called Leonard, that
classifies simple spatial motion events, such as `pick up' and `put down',
from video input. Unlike previous systems that classify events based on their
motion profile, Leonard uses changes in the state of force-dynamic relations,
such as support, contact, and attachment, to distinguish between event types.
Since force-dynamic relations are not visible, Leonard must construct
interpretations of its visual input that are consistent with a physical theory
of the world. Leonard models the physics of the world via kinematic stability
analysis and performs model reconstruction via prioritized circumscription
over this analysis. In this talk, I will present an overview of the entire
system, along with the details of both the model reconstruction process and the
subsequent event-logic inference algorithm that can infer occurrences of
compound events from occurrences of primitive events. This inference
algorithm uses a novel representation, called spanning intervals, to give a
concise representation of the large interval sets that occur when representing
liquid and semi-liquid events. I will illustrate how Leonard handles a
variety of complex visual-input scenarios that cannot be handled by approaches
that are based on motion profile, including extraneous object in the field of
view, sequential and simultaneous event occurrences, and non-occurrence of
events. I will also present a live example illustrating the end-to-end
performance of Leonard classifying an event from video input.
Jeffrey Mark Siskind
Ph.D., MIT, Electrical Engineering and Computer Science, 1992.
Jeff joined NECI in 1997 after a postdoctoral position at the University of Pennsylvania, and faculty positions at the University of Toronto, the Technion, and the University of Vermont. His research straddles cognitive science and computer systems. He currently focuses on three areas: visual event perception, computational modeling of child language acquisition, and optimizing compilers for mostly-functional programming languages.
12/05/00 - Michael Collins
AT&T Labs-Research
Statistical Models for Natural Language Parsing
This talk will discuss the problem of machine learning applied to
natural language parsing: given a set of example sentence/tree pairs,
the task is to learn a function from sentences to trees which
generalizes well to new sentences.
In the first part of the talk I will review recent work on
probabilistic, history-based approaches. Much of the recent success of
these methods has been due to the incorporation of lexically
conditioned parameters. I will discuss the importance of head words
and dependency parameters, and also the use of estimation methods such
as decision trees or maximum entropy methods.
While history-based models have several advantages, it can be awkward
to encode some constraints within this framework. It is often easy to
think of features which might be useful in discriminating between
candidate trees for a sentence, but much more difficult to alter the
model to take these features into account. In the second part of the
talk I will review more recent work on learning methods which promise
to be considerably more flexible in incorporating features. I will
discuss how three such approaches -- boosting, support vector machines
and markov random fields -- can be applied to parsing, and the
similarities and relative advantages of the three approaches.
Michael Collins
did his undergraduate studies in Electrical
Engineering at Cambridge University, and went on to do a Masters in
Speech and Language Processing, also at Cambridge. He received his PhD
from University of Pennsylvania in 1998, and has been at AT&T
labs-research since January 1999, most recently in the AI
department. His research interests are in machine-learning approaches
to natural language processing.
01/30/01 - Christopher Raphael Dept of Math and Statistics, U Mass, Amherst.
Can the Computer Learn to Play Music Expressively?
I will discuss my recent work in creating a computer program
that plays the role of a sensitive musical accompanist in a
non-improvisatory piece for soloist and accompaniment.
An accompanist must synthesize a number of different sources of
information: The accompanist must follow the presciption of the
musical score, must "listen to" the soloist, and must learn from
rehearsals, all while obeying an internal sense of musicality.
During live performance, my accompaniment system combines these
sources of information into to a coherent probabilistic model, a
Bayesian belief network, from which it can deduce the optimal course
of action in real time. I will provide a demonstration.
Christopher Raphael
is currently an Assistant Professor in the Department
of Mathematics and Statistics at the University of Massachusetts, Amherst
where his interests include Bayesian belief networks and hidden Markov models
and their applications to various recognition problems.
Prior to his current appointment he was a Research Scientist in
the Speech and Natural Language Processing group at BBN working
in optical character recognition.
As a winner of the San Francisco Symphony Young Artist Competition,
he soloed with the San Francisco Symphony in 1978 and has held a fellowship
at Tanglewood. He played principal oboe in the Santa Cruz Symphony
from 1982 to 1985 and soloed with that orchestra on several occasions.
04/04/01 - Makoto Yokoo NTT Communication Science Labs, JAPAN
Theory of Internet Auctions: Effect of False-name Bids
on Auction Protocols
In this talk, we examine the effect of a new type of frauds
called false-name bids, which can be a serious problem in Internet
auctions. False-name bids are bids submitted by a single agent under
multiple fictitious names (e.g., multiple e-mail addresses). If there
exists no false-name bid, the Generalized Vickrey Auction protocol
(GVA) satisfies three desirable properties, i.e., individual
rationality, Pareto efficiency, and incentive compatibility. On the
other hand, when false-name bids are possible, we prove that it is
theoretically impossible for an auction protocol to simultaneously
satisfy these three properties. We develop a new auction protocol
called Leveled Division Set (LDS) protocol, which satisfies individual
rationality and incentive compatibility.
Makoto Yokoo is currently a distinguished technical member in NTT
Communication Science Laboratories, Kyoto, Japan. His research
interests include multi-agent systems, search/constraint satisfaction,
and mechanism design among agents. He recently published a book
"Distributed Constraint Satisfaction: Foundation of Cooperation in
Multi-Agent Systems" from Springer. He is a board member of
International Foundation of Multi-agent Systems (IFMAS), which is the
host organization of International Conference on Multi-agent systems
(ICMAS). He is also editorial board members and program committee
members of various international journals and conferences, including
Journal on Artificial intelligence Research, IJCAI.
Webpage maintained by Stella
X. Yu