[Date Prev][Date Next][Date Index]
CSLI Calendar, Wednesday, 21 March 2007, vol. 22:27
CSLI CALENDAR OF PUBLIC EVENTS
______________________________________________________________________
21 March 2007 Stanford Vol. 22, No. 27
______________________________________________________________________
A weekly publication of the
Center for the Study of Language and Information (CSLI)
Stanford University, Cordura Hall, Stanford, CA 94305-4101
http://www-csli.stanford.edu/
____________
ACTIVITIES FROM 21 MARCH 2007 TO 30 MARCH 2007
WEDNESDAY, 21 MARCH 2007
3:30pm SRI CCB Seminar Series [21-Mar-07]
EK255, SRI International
"Combining mathematical modeling and text-mining to aid
real-life biological research"
Andrey Rzhetsky
Biomedical Informatics, Columbia University
http://genome6.cpmc.columbia.edu/andrey/andrey.html
(I have no url for the CCB seminar series, if anyone knows it
that would be great)
Abstract below
6:30pm SF Bay ACM Talk [21-Mar-07]
Hewlett Packard, Pruneridge and Wolfe, Cupertino, Bldg. 48, Oak Room
"The Scribblenet: APIs, Machines and Magic Words -
Building the Do What I Mean Engine"
Aaron Straup Cope
Yahoo!
http://sfbayacm.org/
Abstract below
THURSDAY, 22 MARCH 2007
11:00am Music 319: CCRMA Hearing Seminar [22-Mar-07]
CCRMA Seminar Room, The Knoll
"On the sensation of time"
Hynek Hermansky
Martigny Switzerland
http://ccrma-mail.stanford.edu/mailman/listinfo/hearing-seminar
Abstract below
2:00pm Berkeley Intel Research Open House [22-Mar-07]
Intel Research Berkeley, 2150 Shattuck, Ste. 1300
Open House from 2 to 4, come see what they do
http://www.intel-research.net/berkeley/Seminars.asp
1:30pm Kailath Colloqium [22-Mar-07]
Clark Center Auditorium
http://isl.stanford.edu/kailathlecture/
(free registration required)
3:00pm Kailath Lecture [22-Mar-07]
Clark Center Auditorium
"Codes and Graphical Models: A Personal Survey"
David Forney
MIT
http://isl.stanford.edu/kailathlecture/
(this is part of the colloquium above)
Abstract below
4:00pm SRI AI Seminar Series [22-Mar-07]
EJ228, SRI International
"The Smart Personal Assistant: An Overview"
Wayne Wobcke
Computer Science and Engineering, University of New South Wales
http://www.ai.sri.com/seminars/
Abstract below
4:00pm PARC Forum [22-Mar-07]
George Pake Auditorium at PARC
"Beer: The Best Beverage in the World"
Charlie Bamforth
UC Davis
http://www.parc.com/forum/
4:00pm UC Berkeley CIS Seminar [22-Mar-07]
Soda Hall 310 (UC Berkeley)
"Market mechanisms for agent coordination"
Sven Koenig
U. of Southern California
http://www.cs.berkeley.edu/~feisha/cis-seminar
Abstract below
FRIDAY, 23 MARCH 2007
all day Kailath Colloqium [23-Mar-07]
Packard 101
http://isl.stanford.edu/kailathlecture/
(free registration required)
12 noon Stanford Talk [23-Mar-07]
Wallenberg Hall Learning Theater (Bldg. 160)
"Microformats"
Tantek Celik
Technorati
http://events.stanford.edu/
Abstract below
3:00pm SRI STAR Lab Seminars [23-Mar-07]
EJ 124 (SRI International)
"Recent Advances in Automatic Speech Summarization"
Sadaoki Furui
Computer Science, Tokyo Institute of Technology
http://www-speech.sri.com/seminars/
Abstract below
3:00pm Berkeley Information Access Seminar [23-Mar-07]
107 South Hall (Berkeley)
"The Survivability of Texts"
Roger C. Schonfeld
Manager of Research, Ithaka
http://courses.ischool.berkeley.edu/i296a-1/s07/schedule.html
Abstract below
4:00pm UC Berkeley Oxyopia Lecture [23-Mar-07]
489 Minor Hall (UC Berkeley)
"Compensatory processes in color vision"
Michael Webster
Psychology, University of Nevada, Reno
http://optometry.berkeley.edu/opt_txtpp/ce/oxyopias.html
Abstract below
4:15pm CSLI Seminar on Computational Learning and Adaptation [23-Mar-07]
Cordura Hall 100
"iLink and Web 2.0: Machine Learning on Dynamic Content-Driven
Social Networks"
Sugato Basu
SRI AI Center
http://cll.stanford.edu/scla/schedule.shtml
Abstract below
MONDAY, 26 MARCH 2007
TUESDAY, 27 MARCH 2007
2:00pm Berkeley International Computer Science Institute [27-Mar-07]
ICSI, Suite 600 (UC Berkeley)
"Ensembles of classifiers"
Alberto Suerez
Universidad Autonoma de Madrid and visiting researcher at ICSI
http://www.icsi.berkeley.edu/
Abstract below
WEDNESDAY, 28 MARCH 2007
4:00pm SRI AI Seminar Series [28-Mar-07]
EJ228, SRI International
"Bridging the Gap Between Formal and Computational Semantics"
Stephen Pulman
Oxford University
http://www.ai.sri.com/seminars/
Abstract below
THURSDAY, 29 MARCH 2007
all day Stanford 50 Conference [29-Mar-07]
Stanford
"State of the Art and Future Directions of Computational
Mathematics and Numerical Computing"
A conference celebrating the 50th anniversary of George
Forsythe's arrival at Stanford and the 75th birthday of
Professor Gene Golub
http://compmath50.stanford.edu/
(registration fee of $125, students free [except for banquet])
12 noon Stanford Semantics and Pragmatics Workshop [29-Mar-07]
Margaret Jacks Hall 460:126
"Bridging the Gap Between Formal and Computational Semantics"
Stephen Pulman
Oxford University
http://www-linguistics.stanford.edu/semgroup/
(see abstract for the 28 March SRI AI seminar below)
4:00pm SRI AI Seminar Series [29-Mar-07]
EJ228, SRI International
"From an Adaptive Reminding System to a Virtual Coach"
Nadine Richard
Japanese National Institute of Informatics
http://www.ai.sri.com/seminars/
Abstract below
4:00pm PARC Forum [29-Mar-07]
George Pake Auditorium at PARC
"Optical Interference Special Effect Pigments:
From Banknote Security to Nail Polish"
Alberto Argoitia
Flex Products, Division of JDSU
http://www.parc.com/forum/
FRIDAY, 30 MARCH 2007
all day Stanford 50 Conference [30-Mar-07]
Stanford
"State of the Art and Future Directions of Computational
Mathematics and Numerical Computing"
A conference celebrating the 50th anniversary of George
Forsythe's arrival at Stanford and the 75th birthday of
Professor Gene Golub
http://compmath50.stanford.edu/
(registration fee of $125, students free [except for banquet])
SATURDAY, 31 MARCH 2007
all day Stanford 50 Conference [31-Mar-07]
Stanford
"State of the Art and Future Directions of Computational
Mathematics and Numerical Computing"
A conference celebrating the 50th anniversary of George
Forsythe's arrival at Stanford and the 75th birthday of
Professor Gene Golub
http://compmath50.stanford.edu/
(registration fee of $125, students free [except for banquet])
____________
Stanford Blood Center: Shortage of all types except AB+. For an
appointment: http://bloodcenter.stanford.edu/ or call 650-723-7831.
It only takes an hour of your time and you get free cookies.
____________
NOTE
I note that cross-bay the Intel Research Lab at Berkeley has an open
house tomorrow (22 March) from 2pm to 4pm. They promise an
opportunity to meet University faculty, Intel researchers, students
and see demos and posters in areas such as systems, networking,
programming languages, statistical databases, technology for emerging
regions, urban computing, and mobility. See
http://www.intel-research.net/ for more information.
And best wishes to the students taking finals this week and have a
good spring break.
____________
SRI CCB SEMINAR SERIES
on Wednesday, 21 March 2007, 4:00pm - 5:30pm
EK255, SRI International
(I have no url for the CCB seminar series, if anyone knows it
and can tell me that would be great)
"Combining mathematical modeling and text-mining to aid
real-life biological research"
Andrey Rzhetsky
Biomedical Informatics, Columbia University
http://genome6.cpmc.columbia.edu/andrey/andrey.html
The information overload in molecular biology is a mere example of the
status common to all fields of the current science and culture: An
ever-strengthening avalanche of novel data and ideas overwhelms
specialists and non-specialists alike. The help of relieving the
information overload may come from the text-miners who can
automatically extract and catalogue facts described in books and
journals.
My talk will touch the following questions: What can large-scale
analyses of scientific literature tell us about both active and
forgotten knowledge? What can such analyses tells us about the
scientific community itself? How do mathematical models help us to
differentiate true and false statements in literature? How will text-
mining help us to find cures for human and non-human maladies?
____________
SF BAY ACM TALK
on Wednesday, 21 March 2007, 6:30pm - 9:00pm
Hewlett Packard, Pruneridge and Wolfe, Cupertino, Bldg. 48, Oak Room
http://sfbayacm.org/
"The Scribblenet: APIs, Machines and Magic Words -
Building the Do What I Mean Engine"
Aaron Straup Cope
Yahoo!
Flickr is one of the most successful photo hosting sites. It has
brought a lot of innovation into the field through it's use of tagging
and geocoding. Now machine tags and the Flickr API are launching the
next wave of rich internet applications using Ajax and powerful
backend services.
The "Scribblenet" is a gentle romp through the design and philosophy
of walking the line between making it easy enough for people to bother
putting data in to a system and still useful enough to make it worth
the trouble of getting it out again.
Using the Flick API (commonly known as "Application Programming
Interfaces" but perhaps better understood, today, as "Anti Platform
Initiatives" or "'Architecture of Participation' Interfaces") and
machine tags as examples, the talk will discuss why applications, and
developers, should open up and let go as a first step in building
computing's elusive "Do What I Mean" engine.
About the Speaker: Aaron is Canadian by birth, American by descent,
North American by experience et Montrealais au fond. Aaron works at
Flickr doing mobile and geo related hackin...I mean,
engineering. Aaron does not normally speak in the third person and by
all accounts "there's flesh under all that RDF-talk."
____________
MUSIC 319: CCRMA HEARING SEMINAR
on Thursday, 22 March 2007, 11:00am
CCRMA Seminar Room, The Knoll
http://ccrma-mail.stanford.edu/mailman/listinfo/hearing-seminar
I've very happy to announce the Hynek Hermansky will be presenting
his work at next week's CCRMA Hearing Seminar.
Hynek has done some of the most important work connecting the
engineering of speech recognition systems to the science of auditory
perception. His work on perceptual linear prediction (PLP) turned out
to look a lot like the MFCC solution so common in today's speech
recognition systems. Lately he's been thinking about the temporal
course of speech and how it's processed by our brain. --Malcolm Slaney
"On the sensation of time"
Hynek Hermansky
Martigny Switzerland
Spectral content of sounds has been studied for millennia as an
important information carrier. However, in most situations, it is
changes in the spectral content, rather than the spectral content
itself, that carry the targeted information. We describe emerging
speech processing techniques that that use medium-spans (several
hundreds of ms) of the speech signal to evaluate local dynamics of
spectral components and discuss their possible advantages in
extraction of linguistic information from speech. Some properties of
these techniques, partially derived directly from large amounts of
labelled speech data, appear to be consistent with spectro-temporal
properties of mammalian auditory cortex.
Hynek Hermansky is employed at the IDIAP Martigny, Switzerland, and is
an Adjunct Professor at the Swiss Federal Institute of Technology at
Lausanne, Switzerland (among a number of other, unfortunately also
mostly unpaid positions). He has been working in speech processing for
over 30 years. He is a Fellow of IEEE for "Invention and development
of perceptually-based speech processing methods". He holds
Dr.Eng. Degree from the University of Tokyo, and Dipl. Ing. Degree
from Brno University of Technology, Czech Republic. His main research
interests are in acoustic processing for speech recognition.
____________
KAILATH LECTURE
on Thursday, 22 March 2007, 3:00pm
Clark Center Auditorium
http://isl.stanford.edu/kailathlecture/
"Codes and Graphical Models: A Personal Survey"
David Forney
MIT
Codes and Graphical Models: A Personal Survey
David Forney
Adjunct Professor, MIT
In the past decade, the promise held forth by Shannon in 1948 has at
last been practically realized. Turbo codes and low-density
parity-check (LDPC) codes now approach the Shannon limit within tenths
of a dB, and are being implemented in almost all new digital
communications systems.
A unified conceptual foundation for these capacity-approaching codes
and their decoding algorithms has been developed in a new field called
"codes (or systems) on graphs." This field may be viewed as a
generalization of classical discrete-time linear system theory, in
which the conventional sequential discrete time axis is replaced by a
general graph.
This field evolved out of various antecedents, including:
- the study of convolutional codes as linear finite-state systems over
finite fields;
- trellis and tail-biting trellis realizations of block codes;
- Tanner graph representations of LDPC codes;
- Wiberg's unification of all these topics, and his cut-set bound.
We will offer a guided high-level tour through these developments,
from a personal perspective.
We will also mention some currently open problems. For example, the
classical theory of minimal realizations of linear codes and systems
extends rather completely to linear (or group) codes/systems defined
on cycle-free graphs. However, certain fundamental obstacles stand in
the way of constructing minimal realizations of linear codes/systems
defined on graphs with cycles, such as tail-biting realizations.
Nevertheless, some progress has been made.
About the Speaker: G. David Forney, Jr. is one of the foremost
contributors to the theory and the practice of digital
communications. In his 1965 doctoral thesis, he proposed the use of
concatenated codes, which later became the standard for space
communications. Subsequently, at Codex Corporation, a start-up company
in Cambridge, MA, he designed and implemented the first commercially
successful family of high-speed telephone-line modems, which later
became an international standard. In the 1970s, Dr. Forney made
fundamental contributions to the theory of convolutional codes, the
workhorse of most modern communication systems. In exploring the
theory of such codes, he made significant connections to linear system
theory, and also discovered the optimal demodulation technique for
channels with intersymbol interference. In the 1980s, he made key
contributions to the theory and practice of trellis-coded modulation,
and of combined coding and equalization, which were used in later
generations of high-speed modem standards. Most recently, he has been
working on the theory of codes and systems on graphs, which provides a
unified conceptual foundation for the various capacity-approaching
codes that have been discovered in the last 15 years.
____________
SRI AI SEMINAR SERIES
on Thursday, 22 March 2007, 4:00pm - 5:30pm
EJ228, SRI International
http://www.ai.sri.com/seminars/
"The Smart Personal Assistant: An Overview"
Wayne Wobcke
Computer Science and Engineering, University of New South Wales
The Smart Personal Assistant (SPA) enables users to access e-mail and
calendar information using natural language dialogue through a PDA
device. The user interface to the SPA must present the system as a
single unified set of back-end task assistants, enabling the user to
conduct a dialogue in which it is easy to switch between these
domains. The SPA is implemented using JACK Intelligent Agents, and
includes a special BDI Coordinator agent with plans both for
coordinating the actions of the task assistants and for encoding the
systems dialogue model.
This talk will provide an overview of the Smart Personal Assistant
project, including the rationale for the system design, the research
and engineering issues addressed in building the SPA, and the results
of a recent user study to test both the effectiveness of dialogue
management and the usability of the system. The research contributions
of the project include an agent-based approach to dialogue management
and a method for the use of Ripple Down Rules in e-mail and calendar
management a rule-based technique that guarantees a high degree of
accuracy whilst not being burdensome for the user to create and
maintain rule sets. More general issues discussed include strategies
used in dialogue processing to compensate for speech recognition
errors, and the design of usability studies for personal assistant
systems. The emphasis given to these topics can be varied according to
the interests of the audience. Joint work with Anh Nguyen, Van Ho and
Alfred Krzywicki.
About the Speaker: Wayne Wobcke is an Associate Professor in the
School of Computer Science and Engineering at the University of New
South Wales. He is also Program Manager for the Natural Adaptive User
Interfaces/Smart Personal Assistants Program of the Cooperative
Research Centre for Smart Internet Technology (a government funded
university-industry collaborative research centre).
His current research interests centre on the development of logics of
rational agency that are computationally grounded in PRS-type agent
architectures, and on personal assistant applications, focusing on
dialogue management and computational models of teamwork. Between 1998
and 2000, he worked at British Telecom Laboratories in the U.K. on the
Intelligent Assistant project, which was recognized in the award of
the British Computer Society Medal for Innovation in Information
Technology.
____________
UC BERKELEY CIS SEMINAR
on Thursday, 22 March 2007, 4:00pm-5:00pm
Soda Hall 310 (UC Berkeley)
http://www.cs.berkeley.edu/~feisha/cis-seminar
"Market mechanisms for agent coordination"
Sven Koenig
U. of Southern California
In this talk, I will give an overview of our research on market
mechanisms for the allocation of resources in cooperative domains. As
example, I will use exploration tasks where a team of mobile robots
needs to visit a number of given targets in known or partially unknown
terrain. An important characteristic of these multi-robot routing
tasks is that the assignment of targets to robots can turn out to be
suboptimal as the robots gain more information about the
terrain. Auctions promise to assign and re-assign targets to robots
efficiently in terms of both the required amount of computation and
communication since information is compressed into numeric bids that
the robots can compute in parallel. I will discuss the advantages and
disadvantages of different auction mechanisms, including recent
theoretical results that show that sequential single-item auctions can
provide constant factor performance guarantees in known terrain even
though they run in polynomial time. Time permitting, I will also
discuss generalizations of sequential single-item auctions, such as
sequential single-item auctions with bundles.
This is joint work with D. Kempe, P. Keskinocak, M. Lagoudakis, V.
Markakis, A. Meyerson, C. Tovey and our students.
____________
STANFORD TALK
on Friday, 23 March 2007, 12 noon
Wallenberg Hall Learning Theater (Bldg. 160)
http://events.stanford.edu/
"Microformats"
Tantek Celik
Technorati
"Designed for humans first and machines second, microformats are a
set of simple, open data formats built upon existing and widely
adopted standards."
-- http://microformats.org
"Microformats are about using the standards we all know and love to
convey as much semantic meaning as possible. Think of them as semantic
best practices. They use current XHTML tags such as address, cite, and
blockquote and attributes such as rel, rev, and title to create
semantically appropriate blocks of code. Microformats are great
because they are both usable and elegant---and all you need to do to
get started with them is familiarize yourself with the best ways to
apply the tags and attributes you already use."
-- http://www.digital-web.com
A Q&A session will immediately follow Tantek's presentation.
Related Links:
http://www.microformats.org/
http://en.wikipedia.org/wiki/Microformats
About the Speaker: Tantek is the Chief Technologist for
Technorati.com. Prior to Technorati he worked at Microsoft, Oracle,
and Apple. He graduated from the Computer Science department here in
1991.
____________
SRI STAR LAB SEMINAR
on Friday, 23 March 2007, 3:00pm
EJ 124 (SRI International)
http://www-speech.sri.com/seminars/
"Recent Advances in Automatic Speech Summarization"
Sadaoki Furui
Computer Science, Tokyo Institute of Technology
Speech summarization technology, which extracts important information
and removes irrelevant information from speech, is expected to play an
important role in building speech archives and improving the
efficiency of spoken document retrieval. However, speech summarization
has a number of significant challenges that distinguish it from
general text summarization. Fundamental problems with speech
summarization include speech recognition errors, disfluencies, and
difficulties of sentence segmentation. Typical speech summarization
systems consist of speech recognition, sentence segmentation, sentence
extraction, and sentence compaction components. Most research up to
now has focused on sentence extraction, using LSA (Latent Semantic
Analysis), MMR (Maximal Marginal Relevance), or feature-based
approaches, among which no decisive method has yet been found. Proper
sentence segmentation is also essential to achieve good summarization
performance. How to objectively evaluate speech summarization results
is also an important issue. Several measures, including families of
SumACCY and ROUGE measures, have been proposed, and correlation
analyses between subjective and objective evaluation scores have been
performed. Although these measures are useful for ranking various
summarization methods, they do not correlate well with human
evaluations, especially when spontaneous speech is targeted.
____________
BERKELEY INFORMATION ACCESS SEMINAR
on Friday, 23 March 2007, 3:00pm - 5:00pm
107 South Hall (Berkeley)
http://courses.ischool.berkeley.edu/i296a-1/s07/schedule.html
"The Survivability of Texts"
Roger C. Schonfeld
Manager of Research, Ithaka
I've been working on a history of the survivability of texts in the
United States over the course of the past 150 years. This history has
major implications for policymaking about how to organize future
preservation services for both print and electronic materials. From a
systemwide organizational perspective, there are significant
implications for the network accessible environment. I will give a
brief overview of my work and invite discussion.
About the Speaker: Roger Schonfeld is Manager of Research for Ithaka,
a not-for-profit organization working to help higher education take
advantage of advances in information technologies. Roger leads a
research program presently focused on the transition away from print
and related preservation issues; on understanding the community's
information-services needs; and on improving our understanding of new
resource models for teaching and learning. Roger is the author of
JSTOR: A History (Princeton University Press, 2003), which examines
business models for the shift to an online environment for scholarly
texts by focusing on how JSTOR developed into a self-sustaining
not-for-profit organization. He has also published The Nonsubscription
Side of Periodicals (Council on Library and Information Resources,
2004) and, with Brian Lavoie, the most comprehensive examination of
the systemwide print book collection, "Books without Boundaries: A
Brief Tour of the System-wide Print Book Collection," Journal of
Electronic Publishing, 2006. Previously, Roger was a research
associate at The Andrew W. Mellon Foundation. Mar 30: Spring break: No
seminar meeting.
____________
UC BERKELEY OXYOPIA LECTURE
on Friday, 23 March 2007, 4:00pm
489 Minor Hall (UC Berkeley)
http://optometry.berkeley.edu/opt_txtpp/ce/oxyopias.html
"Compensatory processes in color vision"
Michael Webster
Psychology, University of Nevada, Reno
Color appearance is strongly shaped by adaptation to the observer's
color environment. These adjustments may play an important role in
maintaining color constancy despite changes in the spectral
sensitivity of the observer, and are illustrated both for changes over
time (e.g., after cataract surgery) and over space (e.g., at different
retinal locations). Conversely, the same processes should lead to
differences in color perception when the environment changes, and
these are illustrated through measurements of color appearance for
individuals living in different environments. Color percepts may be
adjusted not only for the average spectrum but also for higher-order
properties of the stimulus, for instance calibrating how hue changes
with saturation. For example, broadening the bandwidth of a spectrum
alters the pattern of cone excitations, yet for lights with Gaussian
spectra the perceived hue is independent of bandwidth, suggesting that
constant hues are tied to a fixed property of the stimulus (e.g., the
spectral peak) rather than a fixed physiological response (e.g., the
relative cone responses). This suggests that hue judgments may be
largely compensated for the filtering effects of the eye's spectral
sensitivity, and that this compensation embodies specific inferences
about natural color signals.
____________
CSLI SEMINAR ON COMPUTATIONAL LEARNING AND ADAPTATION
on Friday, 23 March 2007, 4:15pm to 5:30pm
Cordura Hall 100
http://cll.stanford.edu/scla/schedule.shtml
"iLink and Web 2.0:
Machine Learning on Dynamic Content-Driven Social Networks"
Sugato Basu
SRI AI Center
In the last decade, machine learning and data mining techniques have
seen widespread successful application to different Internet
technologies, including web search, product recommendation, spam
detection, spelling correction, and news clustering.
However, the web is fast undergoing a paradigm shift, moving from
being a mechanism for delivering static web-content in the existing
Web 1.0 model to a platform facilitating dynamic collaborative content
creation in the emerging Web 2.0 paradigm. This trend is reflected in
the growing popularity of new social web-services, for example,
tagging (Flickr) compared to photo editing (Ofoto), and blogging
(Blogger) compared to homepage hosting (Geocities).
This talk will outline how this new emphasis on rapid creation and
sharing of user-generated content (UGC) over large social networks has
given rise to dynamic content-driven social networks, and a new set of
challenging machine learning problems in this context. Focusing on a
project (iLink) that the speaker is currently working on, the talk
will discuss research problems like online learning of topic models
over streaming text data, large-scale topic analysis over UGC in
social networks, and learning to route messages in a social query
model.
____________
BERKELEY INTERNATIONAL COMPUTER SCIENCE INSTITUTE
on Tuesday, 27 March 2007, 2:00 - 3:00 p.m.
Main Lecture Hall, ICSI, 1947 Center Street, Sixth Floor, Berkeley
http://www.icsi.berkeley.edu/
"Ensembles of classifiers"
Alberto Suerez
Universidad Autonoma de Madrid and visiting researcher at ICSI
Numerous experimental studies show that pooling the decisions of
classifiers in an ensemble can lead to improvements in the
generalization performance of weak learners. In the first part of this
talk we present two novel algorithms to generate ensembles of
classifiers. The first method takes advantage of the intrinsic
variability of the iterative growing and pruning tree construction
procedure devised by Gelfand et al. [Gelfand, S.B., Ravishankar, C.S.
and Delp, E.J. (1991), "An Iterative Growing and Pruning Algorithm for
Classification Tree Design", IEEE Transactions On Pattern Analysis and
Machine Intelligence Vol. 13, no. 2 138-150].
The second procedure consists in generating classifiers of the same
type built using perturbed versions of the training data. These
perturbed data are obtained by injecting random uncorrelated noise in
the class labels of the original training examples. The final
classification is obtained by averaging over classifiers obtained with
independent realizations of the noise.
The second part of the talk is devoted to exploring the importance of
the order of aggregation in bagging ensembles. We present several
heuristic rules to order the classifiers in bagging ensembles prior to
aggregation.
Early stopping in the aggregation of the classifiers ordered with
these rules allow the identification of subensembles that require less
memory for storage, have a faster classification speed and can perform
better than the original bagging ensemble. Furthermore, this ensemble
pruning procedure does not seem to deteriorate the robust performance
of bagging in noisy classification tasks.
About the Speaker: Alberto Suerez received the degree of Licenciado in
Chemistry from the Universidad Autonoma de Madrid, Madrid, Spain, in
1988, and the Ph.D. degree in Physical Chemistry from the
Massachusetts Institute of Technology (MIT), Cambridge, MA, in
1993. After holding postdoctoral positions at Stanford University
(USA), at the Universite Libre de Bruxelles (Belgium), as a research
fellow financed by the European Commission within the program
"Training and Mobility of Researchers", and at the Katholieke
Universiteit Leuven (Belgium), he is currently a professor in the
Computer Science Department of the Universidad Autonoma de Madrid
(Spain). He has worked on relaxation theory in condensed media,
stochastic and thermodynamic theories of nonequilibrium systems,
lattice-gas automata, and decision tree induction. His current
research interests include machine learning, quantitative and
computational finance, time series analysis and information processing
in the presence of noise.
____________
SRI AI SEMINAR SERIES
on Wednesday, 28 March 2007, 4:00pm - 5:30pm
EJ228, SRI International
http://www.ai.sri.com/seminars/
This talk will also be given in the Stanford Semantics and Pragmatics
Workshop, Margaret Jacks Hall 460:126 on Thursday, 29 March 2007 at 12
noon.
"Bridging the Gap Between Formal and Computational Semantics"
Stephen Pulman
Oxford University
The literature in formal linguistic semantics contains a wealth of
fine grained and detailed analyses of many linguistic phenomena. But
very little of this work has found its way into implementations,
despite a widespread feeling (among linguists at least) that this cant
be very difficult: just fix a grammar to produce the right logical
forms and hook them up to a theorem prover. In this talk I take a
representative analysis of adjectival comparatives and ask what steps
one would have to go through so as to use this analysis in a
computational setting like open domain question-answering. I then try
to identify some general conclusions that can be drawn from this
exercise.
____________
SRI AI SEMINAR SERIES
on Thursday, 29 March 2007, 4:00pm - 5:30pm
EJ228, SRI International
http://www.ai.sri.com/seminars/
"From an Adaptive Reminding System to a Virtual Coach"
Nadine Richard
Japanese National Institute of Informatics
With the ever-growing load of information, events and various
commitments we need to handle, the management of personal calendars
and todo lists becomes more and more difficult and time-consuming. We
are currently developing an adaptive, emotional, and expressive
interface agent, which learns when and how to notify users about tasks
and events. As suggested by its name, the ultimate purpose of our
TamaCoach project is to investigate virtual coaching, in the context
of personal time management. The next step is thus to define the basic
requirements for building a virtual coach, especially in terms of
interruption strategies and expressiveness, in order to provide an
adaptive system that actually motivates the user in achieving
self-assigned tasks.
About the Speaker: Nadine Richard achieved her Ph.D. in 2001 at ENST
(Ecole Nationale Superieure des Telecommunications, Paris). Initially
working on the description of behaviours for virtual agents, she
applied her model to robotic agents and Ambient Intelligence during a
first post-doc at ENST. Since September 2005, she is a post-doc
researcher at NII (National Institute of Informatics, Tokyo), in the
human-agent interaction group led by Prof. Seiji Yamada.
____________
END MATERIAL
The CSLI Calendar appears weekly on most Wednesdays throughout the
year. Announcements, abstracts, and other information to appear in
the Calendar should be submitted to the editor, who reserves the right
to decide what does or does not go in the calendar
mailto:incalendar@csli.stanford.edu
Requests to be added to the mailing list should be sent to
majordomo@csli.stanford.edu. With the lines in the body of the text
of either
subscribe csli-calendar
for the long form or
subscribe csli-short-calendar
for the short form (i.e., no abstracts). Problems with subscribing or
unsubscribing should be sent to
owner-csli-calendar@csli.stanford.edu.
The full current issue is at
http://www-csli.stanford.edu/Archive/calendar/current.shtml
and the archives at
http://www-csli.stanford.edu/Archive/calendar/
People on many of the CSLI computers can type 'help csli-calendar' to
see the current issue.
The CSLI Calendar is also posted each week to the su.events usenet
newsgroup (only available from computers on the Stanford network)
Information about CSLI's research program is available at
http://www-csli.stanford.edu/
For maps to the Stanford University rooms see
http://www-csli.stanford.edu/events/locations.shtml
____________