[Date Prev][Date Next][Date Index]
CSLI Calendar, Wednesday, 22 October 2008, vol. 24:8
CSLI CALENDAR OF PUBLIC EVENTS
______________________________________________________________________
22 October 2008 Stanford Vol. 24, No. 8
______________________________________________________________________
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 22 OCTOBER 2008 TO 31 OCTOBER 2008
WEDNESDAY, 22 OCTOBER 2008
12 noon Stanford Online Accessibility Program [22-Oct-08]
McCullough 115
"Universal by Design"
James Craig
Apple
Abstract below
3:45pm Psychology Department Colloquium [22-Oct-08]
Jordan Hall 420:041
"Using illusions to probe the functional architecture of human
vision"
Bart Anderson
University of Sydney
https://www.stanford.edu/dept/psychology/webd/node/906
Abstract below
4:15pm EE380: Computer Systems Colloquium [22-Oct-08]
Gates B01
"Scalable Privacy-Friendly Client Cloud Computing:
a gathering Perfect Disruption"
Carl Hewitt
MIT (emeritus)
http://carlhewitt.info
http://ee380.stanford.edu/contents.html
Abstract below
THURSDAY, 23 OCTOBER 2008
10:00am Stanford IT Open House [23-Oct-08]
Arrillaga Alumni Center
open to the Stanford community
http://itopenshouse.stanford.edu/
11:30am CCRMA Hearing Seminar [23-Oct-08]
CCRMA Seminar Room, The Knoll
"Temporal Features of Speech in the Auditory System: Normal
and Dyslexic Children"
Daniel Abrams
Stanford
http://ccrma-mail.stanford.edu/mailman/listinfo/hearing-seminar
Abstract below
4:00pm SRI AI Seminar Series [23-Oct-08]
EJ228, SRI International
"True Knowledge: Open-Domain Question Answering Using
Structured Knowledge and Inference"
William Tunstall-Pedoe
True Knowledge
http://www.ai.sri.com/seminars/
Abstract below
4:00pm PARC Forum [23-Oct-08]
George Pake Auditorium at PARC
"The Crime of Reason"
Robert B. Laughlin
Stanford University
http://www.parc.com/forum/
4:00pm UC Berkeley CIS Seminar [23-Oct-08]
Soda Hall 310 (UC Berkeley)
"On the Complexity of Linear Prediction:
Risk Bounds, Margin Bounds, and Regularization"
Ambuj Tewari
Toyota Technological Institute
http://www.eecs.berkeley.edu/CIS/seminars/seminars.html
Abstract below
4:15pm SSP10: Symbolic Systems Forum [23-Oct-08]
Bldg. 380:380C (Math Corner)
"How Democracy Resolves Conflict in Difficult Games"
Steven J. Brams
New York University
http://symsys.stanford.edu/ssp_events
Abstract below
6:00pm Media X Lecture Series [23-Oct-08]
Wallenberg Hall Learning theater (Bldg. 160:124)
Title to be announced
Joshua Rosen
iLeonardo
Co-sponsored with Stanford Humanities Lab
http://mediax.stanford.edu/
FRIDAY, 24 OCTOBER 2008
11:00am Berkeley Institute of Cognitive and Brain Seminar [24-Oct-08]
Tolman 5101 (Berkeley)
"The geometry of neural similarity spaces studied with fMRI"
Geoffrey Aguirre
University of Pennsylvania
http://icbs.berkeley.edu/
12:30pm CS547: Human-Computer Interaction Seminar [24-Oct-08]
Gates B01
"Information foraging theory"
Peter Pirolli
PARC
http://hci.stanford.edu/cs547/
Abstract below
3:00pm Berkeley Information Access Seminar [24-Oct-08]
107 South Hall (Berkeley)
"Representing Indigenous Knowledge"
Geoffrey Bowker
University of Santa Clara
http://courses.ischool.berkeley.edu/i296a-1/f08/schedule.html
Abstract below
3:15pm Friday Cognitive Seminar [24-Oct-08]
Jordan Hall 420:050
"Perceptual reorganization and the dual representation
hypothesis: How kids can see things they couldn't see before"
Davie Yoon,
Stanford Psychology
http://www-psych.stanford.edu/events_cognitive.html
(Stanford people only)
4:00pm Berkeley Institute of Cognitive and Brain Seminar [24-Oct-08]
Minor 489 (Berkeley)
"Geometric and photometric constraints in the perception
of contours, surfaces and materials."
Bart Anderson
University of Sydney
http://icbs.berkeley.edu/
4:15pm Information Systems Seminar [24-Oct-08]
Packard 101
Title to be announced
Helmut Boelcskei
ETH
http://isl.stanford.edu/colloquium.html
MONDAY, 27 OCTOBER 2008
TUESDAY, 28 OCTOBER 2008
2:00pm Berkeley International Computer Science Institute [28-Oct-08]
ICSI, Rm 607 (UC Berkeley)
"Semantic Role Labeling and Kernel Methods for Question
Answering and Spoken Language Understanding"
Alessandro Moschitti
University of Trento
http://www.icsi.berkeley.edu/
Abstract below
3:30pm Townsend Working Group in Neuroscience and Philosophy [28-Oct-08]
Dennes Room, 234 Moses Hall (UC Berkeley)
"An Overview of Issues in perceptual learning: The Whorfian
Hypothesis and Non-Hebbian learning"
Stan Klein
Optometry and Helen Wills Neuroscience Institute, UC Berkeley
http://cornea.berkeley.edu/people/stan.html
http://neurophilosophy.berkeley.edu/meetings.html
4:15pm Cognitive Systems Seminar [28-Oct-08]
Nora Suppes Hall
"Adaptive Control of Autonomous Underwater Vehicles"
Kanna Rajan
Monterey Bay Aquarium Research Institute (MBARI)
http://cll.stanford.edu/css
Abstract below
4:15pm Mathematical Logic Seminar [28-Oct-08]
Bldg. 160:319
"The Consistency Strength of NFU*"
Robert Solovay
Berkeley
http://www-logic.stanford.edu/logic-seminar.html
Abstract below
WEDNESDAY, 29 OCTOBER 2008
3:45pm Psychology Department Colloquium [29-Oct-08]
Jordan Hall 420:041
"Using Psychology and the Internet to Reduce Health
Disparities Worldwide"
Ricardo Muñoz
UC San Francisco
http://www-psych.stanford.edu/
4:15pm EE380: Computer Systems Lab Colloquium [29-Oct-08]
Gates B03
To be announced
http://ee380.stanford.edu/contents.html
5:30pm Tanner Lecture I [29-Oct-08]
Levinthal Hall, Humanities Center
"Ontogenetic Origins of Human Altruism"
Michael Tomasello
Max Planck Institute for Evolutionary Anthropology
http://ethicsinsociety.stanford.edu/ethics-events/tanner-lectures/
THURSDAY, 30 OCTOBER 2008
10:00am Tanner Lecture Discussion I [30-Oct-08]
Landau Economics Bldg, SIEPR A
Michael Tomasello
Max Planck Institute for Evolutionary Anthropology
Carol Dweck, Stanford Psychology
Elizabeth Spelke, Harvard Psychology
http://ethicsinsociety.stanford.edu/ethics-events/tanner-lectures/
4:00pm SRI AI Seminar Series [30-Oct-08]
EJ228, SRI International
"Managing Personal Tasks with Time Constraints and Preferences"
Ioannis Refanidis
University of Macedonia
http://www.ai.sri.com/seminars/
Abstract below
4:00pm UC Berkeley CIS Seminar [30-Oct-08]
Soda Hall 310 (UC Berkeley)
"Discriminative Methods with Structure"
Simon Lacoste-Julien
UC Berkeley
http://www.eecs.berkeley.edu/CIS/seminars/seminars.html
Abstract below
4:15pm SSP10: Symbolic Systems Forum [30-Oct-08]
Bldg. 380:380C (Math Corner)
"Web Access to Voting Records: Motivations and Issues"
Todd Davies
Symbolic Systems Program
http://symsys.stanford.edu/ssp_events
Abstract below
5:30pm Tanner Lecture II [30-Oct-08]
Levinthal Hall, Humanities Center
"Phylogenetic Origins of Human Collaboration"
Michael Tomasello
Max Planck Institute for Evolutionary Anthropology
http://ethicsinsociety.stanford.edu/ethics-events/tanner-lectures/
FRIDAY, 31 OCTOBER 2008 - Halloween
10:00am Tanner Lecture Discussion II [31-Oct-08]
Landau Economics Bldg, SIEPR A
Michael Tomasello
Max Planck Institute for Evolutionary Anthropology
Joan Silk, UCLA Anthropology
Brian Skyrms, Stanford Philosophy
http://ethicsinsociety.stanford.edu/ethics-events/tanner-lectures/
12:30pm CS547: Human-Computer Interaction Seminar [31-Oct-08]
Gates B01
"Power of Peers"
Justine Cassell
Northwestern University
http://hci.stanford.edu/cs547/
3:00pm Berkeley Information Access Seminar [31-Oct-08]
107 South Hall (Berkeley)
"The Extent of Geographic Resources Available on the Web"
Robert Pasley
Sheffield Univ., U.K.
http://courses.ischool.berkeley.edu/i296a-1/f08/schedule.html
Abstract below
3:15pm Philosophy Department Colloquium [31-Oct-08]
Bldg. 90:92Q
"Emotional Choice & Rational Choice"
Jon Elster
Columbia University
http://www-philosophy.stanford.edu/ce.html
____________
Stanford Blood Center: Shortage of O, A, B-, and 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.
____________
ANNOUNCEMENT
A new series entitled "Cognitive Systems Seminar" will have its
first talk on Tuesday, October 28 at 4:15pm in Nora Suppes Hall (right
next to Cordura Hall). The web site is
<http://cll.stanford.edu/css/>.
Goals of the Seminar
Visions of cognitive systems that sense, think, act, and interact at
human levels have driven the field of Artificial Intelligence since
its inception. Although progress toward this overarching goal has
been difficult to achieve (in part, because AI has emphasized
component technologies as the field has matured) researchers have made
great strides in understanding the elements of intelligent
behavior. There is now a resurgence of interest in integrated
cognitive systems.
The CSLI Seminar on Cognitive Systems aims to encourage and expand
this return to the field's foundational goal. It provides a forum for
researchers to exchange ideas about all aspects of cognitive systems,
including their purpose, design, application, and evaluation. The
Seminar takes an interdisciplinary perspective by examining such
systems from the viewpoints of AI, psychology, linguistics, software
engineering, and other fields. It addresses the importance of
component technologies for perception, inference, planning, execution,
language, and learning, by considering them in the context of
intelligent systems.
The Seminar takes place at Stanford University's Center for the Study
of Language and Information. Most talks will be on Tuesdays at 4:15
PM, when parking is free. As with previous seminar series, we
anticipate that most speakers and attendees will come not just from
Stanford but also from nearby corporate and government research
centers. We encourage open discussion to establish a common language
and lay the groundwork for future collaborations. If you would like to
be added to the seminar mailing list, or if you are interested in
giving a talk, send email to Dan Shapiro, dgs at csli.stanford.edu.
____________
STANFORD ONLINE ACCESSIBILITY PROGRAM
on Wednesday, 22 October 2008, 12 noon
McCullough 115
http://soap.stanford.edu/
"Universal by Design"
James Craig
Apple
Good designers and developers strive to create intuitive, supportive
interfaces that allow people to discover information independently and
use products efficiently. Accessibility for the disabled, if
considered, is most often an afterthought and sometimes dismissed as
"too hard" or "too boring."
In this session, James Craig will demonstrate through numerous
examples, from behavioral psychology to cyber-kinesiology, how
accessible, universal design benefits not only those with
disabilities, but society at large, enabled or disabled. everyone.
About the Speaker: James Craig has more than a decade of experience
designing and developing accessible web sites and software. He works
for Apple in Cupertino, lives in San Francisco, and volunteers for the
Web Standards Project (WaSP) and the World Wide Web Consortium (W3C).
____________
PSYCHOLOGY DEPARTMENT COLLOQUIUM
on Wednesday, 22 October 2008, 3:45pm
Jordan Hall 420:041
http://www-psych.stanford.edu/
"Using illusions to probe the functional architecture of human
vision"
Bart Anderson
University of Sydney
http://www.psych.usyd.edu.au/staff/barta/
A fundamental challenge in vision research is to determine the
organizational principles that give rise to the dimensions of our
experience. Insights into this process can be gained through a
variety of sources, including the physical processes underlying image
formation, physiology, and the structure of our phenomenal experience.
In this talk, I will explain how visual illusions can be used to
evaluate and develop theories of perceptual organization in a number
of visual domains, including the perception of surface reflectance,
and the construction of illusory contours and surfaces.
____________
EE380: COMPUTER SYSTEMS COLLOQUIUM
on Wednesday, 22 October 2008, 4:15pm to 5:30pm
Gates B01
http://ee380.stanford.edu/contents.html
"Scalable Privacy-Friendly Client Cloud Computing:
a gathering Perfect Disruption"
Carl Hewitt
MIT EECS (emeritus)
http://carlhewitt.info
Organizations of Restricted Generality (ORGs) is an architecture
providing foundations for the development of privacy-friendly Internet
applications by incorporating cloud computing into clients. Clients
range from single chip sensors, handhelds, notebooks, desktops, and
entertainment centers to huge data centers. ORGs are well suited for
the issues posed by, what I call, a "Perfect Disruption":
* New Hardware Paradigm: Multicore architecture
* New Software Paradigm: Privacy-friendly client cloud computing
* New Applications Paradigm: Scalable semantic integration (e.g.
search & discovery)
The "Perfect Disruption" is rapidly gathering strength and no one
knows the outcome. Clearly, it will significantly impact big players
like AMD, ARM, Cisco, Google, IBM, Intel, Microsoft, Oracle, SAP,
Yahoo!, and Via. How can basic research help them respond to the
disruption? By addressing fundamental questions like the following:
* How much of concurrent computation is reducible to deduction?
* How can large-scale systems be practically specified, analyzed,
and tested?
* Are the laws of thought consistent?
* Is "recovering rapidly" a more viable policy goal than "maintaining
consistency"?
About the speaker: Speaker Photo Carl Hewitt is Emeritus in the
Electrical Engineering and Computer Science department at the
Massachusetts Institute of Technology (MIT). He is known for his work
on concurrency theory, concurrent programming languages, automatic
storage reclamation and relocation (garbage collection), participatory
semantics, and strongly paraconsistent logic.
He can often be found at the Stanford Logic Group meetings at NOON on
Wednesdays in the Gates Bldg. 2A Open Area. You can look him up at
____________
CCRMA HEARING SEMINAR
on Thursday, 23 October 2008, 11:00am
CCRMA Seminar Room, The Knoll
http://ccrma-mail.stanford.edu/mailman/listinfo/hearing-seminar
I'm very pleased to introduce Daniel Abrams to the CCRMA community.
He'll be speaking about his work on temporal processing at this week's
Stanford Hearing Seminar.
Many years ago we had a couple of talks at Stanford that connected the
perception of speech to reading ability. It is pretty amazing. Let
people with reading disabilities listen to *slow* speech and their
reading scores improve! This is pretty fanciful, but a number of
studies have demonstrated the effect. Now the only problem is
explaining why.
Think about speech (and music). They contain dozens of events a
second. Sometimes there is a sudden change in the energy at one
frequency---that's the start of a "t". Other times the energy moves
from one frequency to another. The most interesting sounds change
continuously, and we must understand them to make sense of our world.
This week at the CCRMA Hearing Seminar we have Daniel Abrams, a recent
graduate of Bev Wright's lab at Northwestern, here to explain his
work. How do we perceive the temporal features of speech, and why
might this ability help or hinder our ability to read?
This talk will be interesting to anybody interested in the perception
of speech and music.
Bring your favorite temporal processor and I'm sure that Dan's talk
and the ensuing discussion will make your temporal processor work
better. --Malcolm Slaney
"Temporal Features of Speech in the Auditory System: Normal
and Dyslexic Children"
Daniel Abrams
Stanford
An amazing characteristic of the speech signal is that it contains a
variety of temporal features that occur simultaneously in the signal,
and each of these features provides unique and essential information
for speech perception. An equally astonishing fact is that, in most
cases, the human auditory system is able to efficiently extract these
temporal acoustic features from the speech signal as a precursor to
higher-order linguistic, cognitive and mnemonic processes associated
with speech reception. A clinical population that has shown abnormal
processing of rapid temporal features in speech is reading-impaired
individuals (RI), and it has been proposed that auditory-temporal
impairments preclude normal development of phonological systems
necessary for reading acquisition. The primary goals of this work are
to describe central mechanisms responsible for encoding temporal
features in speech in the unimpaired human auditory system, and to
examine the extent to which these mechanisms may be impaired in RI.
Results are the first to show right-hemisphere cortical asymmetry in
the representation of the speech envelope, the slow temporal cue that
provides syllable pattern information in speech. We also provide the
first neurophysiological evidence that reading-impaired (RI)
individuals have impaired speech envelope representation, a finding
that challenges an influential hypothesis stipulating that temporal
impairments are specific to rapid features of speech. Taken together,
we have made new discoveries of how the unimpaired human auditory
system encodes perceptually-important temporal features inherent to
the speech signal, and the abnormal function of these mechanisms in
RI.
____________
SRI AI SEMINAR SERIES
on Thursday, 23 October 2008, 4:00pm - 5:30pm
EJ228, SRI International
http://www.ai.sri.com/seminars/
"True Knowledge: Open-Domain Question Answering Using
Structured Knowledge and Inference"
William Tunstall-Pedoe
True Knowledge
True Knowledge is a Cambridge, UK based internet technology
start-up. We have developed a complete question answering system that
combines a knowledge base of structured "facts" about the world; a
query language that represents questions in language-independent form;
a query-answering system that answers these queries both with
knowledge stored directly and knowledge generated with a generalised
inference system and a natural language translation system that takes
questions in English (or potentially any other language) and
translates them to the queries. The user experience is thus getting a
direct accurate answer to a question.
The approach is non-statistical, meaning that the system reliably
knows when it doesn't understand the question or when it doesn't know
the answer and can then fail over to alternative methods. The
knowledge comes from a variety of sources including
crowd-sourcing. The company has been going for three years and the
system works with meaningful numbers of real-world questions.
The talk will explain how the system works, what the remaining
challenges are and will include a live demonstration.
About the Speaker: William Tunstall-Pedoe is the founder of True
Knowledge and the inventor of the technology. William is a Cambridge
University computer science graduate but his career has been outside
of academia, developing AI applications and marketing them through
businesses he has founded.
Previous products include a commercial chess playing program, the
first and only program which can solve and explain cryptic crossword
clues and an AI anagram-generating software which was used by Dan
Brown to create the anagrams that appeared in the Da Vinci Code book
and movie.
____________
UC BERKELEY CIS SEMINAR
on Thursday, 23 October 2008, 4:00pm - 5:00pm
Soda Hall 310 (UC Berkeley)
http://www.eecs.berkeley.edu/CIS/seminars/seminars.html
"On the Complexity of Linear Prediction:
Risk Bounds, Margin Bounds, and Regularization"
Ambuj Tewari
Toyota Technological Institute
Prediction using linearly parametrized function classes is a topic of
fundamental importance to machine learning. The study of the
generalization properties of linear prediction occupies an important
place within this topic. In this talk, I will present sharp
Rademacher and Gaussian complexity bounds for a large family of
linearly parametrized function classes. These complexity measures
play a key role in deriving generalization bounds. These bounds are
derived using conjugate duality, a key tool from convex analysis.
They make short work of providing a number of corollaries including:
risk bounds for linear prediction (including settings where the weight
vectors are constrained by either L_2 or L_1 constraints), margin
bounds (including both L_2 and L_1 margins, along with more general
notions based on relative entropy), a proof of the PAC-Bayes theorem,
and L_2 covering numbers (with L_p norm constraints and relative
entropy constraints). In addition to providing a unified analysis,
these results provide some of the sharpest risk and margin bounds
(improving upon a number of previous results by logarithmic factors).
Interestingly, our results show that the uniform convergence rates of
empirical risk minimization algorithms tightly match the regret bounds
of online learning algorithms for linear prediction (up to a constant
factor of 2).
(Joint work with Sham M. Kakade and Karthik Sridharan)
____________
SSP10: SYMBOLIC SYSTEMS FORUM
on Thursday, 23 October 2008, 4:15pm - 5:30pm
Bldg. 380:380C (Math Corner)
http://symsys.stanford.edu/ssp_events
"How Democracy Resolves Conflict in Difficult Games"
Steven J. Brams
New York University
Democracy resolves conflicts in difficult games like Prisoners'
Dilemma and Chicken by stabilizing their cooperative outcomes. It
does so by transforming these games into games in which voters are
presented with a choice between a cooperative outcome and a
Pareto-inferior noncooperative outcome. In the transformed game, it
is always rational for voters to vote for the cooperative outcome,
because cooperation is a weakly dominant strategy independent of the
decision rule and the number of voters who choose it. Such games are
illustrated by 2-person and n-person public-goods games, in which it
is optimal to be a free rider, and a biblical story from the book of
Exodus.
Note: For background reading, see items 4 and 5 on
http://ai.stanford.edu/~epacuit/lmh/voting-topics.html
(joint work with D. Marc Kilgour)
____________
CS547: HUMAN-COMPUTER INTERACTION SEMINAR
on Friday, 24 October 2008, 12:30-2:00pm
Gates B01
http://hci.stanford.edu/cs547/
"Information foraging theory"
Peter Pirolli
PARC
Information Foraging Theory is a theory of human-information
interaction that aims to explain and predict how people will best
shape themselves to their information environments, and how
information environments can best be shaped to people. The approach
involves a kind of reverse engineering in which the analyst asks (a)
what is the nature of the task and information environments, (b) why
is a given system a good solution to the problem, and (c) how is that
"ideal" solution realized (approximated) by mechanism. Typically, the
key steps in developing a model of information foraging involve: (a) a
rational analysis of the task and information environment (often
drawing on optimal foraging theory from biology) and (b) a
computational production system model of the cognitive structure of
task. This talk will provide a survey of models and applications
developed within the theory, including recent models of Web surfing,
exploratory search, and interaction with information visualizations,
as well as outlines of extension of the theory to social information
foraging.
About the Speaker: Peter Pirolli is a Research Fellow in the Augmented
Social Cognition Area at the Palo Alto Research Center (PARC), where
he has been pursuing studies of human information interaction since
1991. Prior to joining PARC, he was an Associate Professor in the
School of Education at UC Berkeley. Pirolli received his doctorate in
cognitive psychology from Carnegie Mellon University in 1985. He is
an elected Fellow of the American Association for the Advancement of
Science, the Association for Psychological Science, the National
Academy of Education, and the Association for Computing Machinery
Computer-Human Interaction Academy. His recent book is titled
"Information Foraging Theory: Adaptive Interaction with Information".
____________
BERKELEY INFORMATION ACCESS SEMINAR
on Friday, 24 October 2008, 3:00pm - 5:00pm
107 South Hall (Berkeley)
http://courses.ischool.berkeley.edu/i296a-1/f08/schedule.html
"Representing Indigenous Knowledge"
Geoffrey Bowker
University of Santa Clara
The new tools of the information society have largely been created by
and for the developed world. In this talk, I discuss the significance
of and difficulties with representing other ways of knowing. I will
conclude by describing two new projects in this area.
____________
BERKELEY INTERNATIONAL COMPUTER SCIENCE INSTITUTE
on Tuesday, 28 October 2008, 2:00pm - 3:00pm
Main Lecture Hall, ICSI, 1947 Center Street, Sixth Floor, Berkeley
http://www.icsi.berkeley.edu/
"Semantic Role Labeling and Kernel Methods for Question
Answering and Spoken Language Understanding"
Alessandro Moschitti
University of Trento
A deep approach to Natural Language Understanding has been proven to
be infeasible from a computational complexity point of view. This has
led to the development of shallow methods for semantic processing,
where Semantic Role Labeling (SRL) systems are the most recent example
of such research trend. However, the noise and errors in the data
automatically produced by SRL and other natural language parsers
prevent logic-based approaches from effectively carrying out semantic
reasoning.
Machine learning approaches have been shown to be robust to noise but
they require expertise, intuition and deep knowledge about the target
problem to convert semantic structures into attribute-value
representations. Kernel Methods are powerful techniques, which can
simplify the data representation modeling, by encoding semantic
information at a more abstract level than the usual attribute-value
method.
In this talk, after briefly introducing the theory of kernel methods
and Support Vector Machines, we will show the use of different kernels
for complex semantic processing such as Spoken Language Understanding
and Question Answering.
In particular, we will show (a) the use of kernels for improving
standard spoken language understanding, (b) the design of a FrameNet
system for spoken dialog data, which demonstrates that SRL annotation
can be automatically generated for spoken dialog systems and (c) the
effective use of SRL for Question/Answer Classification.
About the Speaker: Alessandro Moschitti is an Assistant Professor at
the Information Engineering and Computer Science Department of the
University of Trento. In 1998, he graduated from the University of
Rome "La Sapienza" with a Master Degree in Computer Science, and then
in 2003 he obtained his PhD in Computer Science at the University of
Rome "Tor Vergata". Between 2002 and 2004, he worked as an associate
researcher in the University of Texas at Dallas for two years.
His expertise concerns machine learning approaches to Natural Language
Processing, Information Retrieval and Data Mining. In particular, he
has designed applications of supervised and unsupervised learning for
Text Categorization, Named Entity Recognition, Co-Reference
Resolution, Text Summarization, Textual Entailment Recognition,
Question Answering, Semantic Role Labeling and Spoken Dialog
Systems. He has recently devised innovative kernels within Support
Vector and other kernel-based machines for advanced syntactic/semantic
processing. His work has been published in the major conferences of
different research communities, e.g. ACL, ICML, CIKM and ICDM.
____________
COGNITIVE SYSTEMS SEMINAR
Nora Suppes Hall
on Tuesday, 28 October 2008, 4:15pm
http://cll.stanford.edu/css
"Adaptive Control of Autonomous Underwater Vehicles"
Kanna Rajan
Monterey Bay Aquarium Research Institute (MBARI)
http://cll.stanford.edu/css
Ocean Science the world over is at a cusp, with a move from the
Expeditionary to the Observatory mode of doing science. With the
advent of ocean observatories, a number of key technologies have
proven to be promising for sustained ocean presence. Mobile robots
routinely map the benthic environment and sample the water-column up
to depths of 6000 meters while tele-operated vehicles navigate remote
depths, performing scientific experiments in-situ relating to
biogeochemical processes. Such platforms, however, have inherent
limitations with how they are commanded and operated; pre-defined
sequences of commands are currently used to determine what actions the
robot will perform and when, irrespective of the contextual
environment in which it operates. As a consequence, not only can
robots not recover from unforeseen failure conditions, they are unable
to significantly leverage their substantial onboard assets to do
opportunistic science.
To mitigate such shortcomings, we are developing new techniques to
dynamically command Autonomous Underwater Vehicles (AUV). Our efforts
use a blend of generative and deliberative Artificial Intelligence
Planning and Execution techniques to shed goals, introspectively
analyze onboard resources and recover from failures. In addition, we
are working on clustering techniques to adaptively trigger science
instruments that will contextually sample the seas as driven by
scientific intent. The end goal is to enable unstructured exploration
of subsea environments that are a rich trove of problems for
autonomous systems. Our work builds on research efforts at NASA to
command deep space probes and Mars rovers, and transfers lessons into
the oceanic domain. In this talk I will articulate the challenges of
working in the hostile underwater domain, lay out the differences
relative to space applications, and motivate our architecture for
goal- driven autonomy on AUV's.
About the Speaker: Kanna is the Principal Researcher in Autonomy at
the Monterey Bay Aquarium Research Institute (http://www.mbari.org) a
privately funded non-profit Oceanographic institute which he joined in
October 2005. Prior to that he was a Senior Research Scientist and a
member of the management team of the 95 member Autonomous Systems and
Robotics Area at NASA Ames Research Center Moffett Field, California.
At Ames, Kanna was the Principal Investigator of the MAPGEN Mixed-
Initiative Planning effort to command and control the Spirit and
Opportunity rovers on the surface of the Red Planet. He was one of the
six principals of the Remote Agent Experiment (RAX) team, which
designed, built, tested and flew the first closed-loop AI based
control system on a spacecraft. The RA was the co-winner of NASA's
1999 Software of the Year, the agency's highest technical award
<http://ic.arc.nasa.gov/projects/remote-agent/>. Kanna's interests are
in automated Planning/Scheduling, modeling and representation for real
world planners and agent architectures for Distributed Control
applications.
____________
MATHEMATICAL LOGIC SEMINAR
on Tuesday, 28 October 2008, 4:15pm
Bldg. 80:115
http://www-logic.stanford.edu/logic-seminar.html
"The Consistency Strength of NFU*"
Robert Solovay
Berkeley
New Foundations (NF) is a rather strange variant of set theory proposed
by Quine. Its consistency is still an open problem. Jensen proposed a
variant, NFU, which is close in spirit to NF but which can be proved
consistent.
I have computed the consistency strength of several variants of NFU
proposed by Randall Holmes. The topic of this talk is the system
NFU*. It turns out to have precisely the consistency strength of ZC
(Zermelo Set Theory with choice) plus Sigma_2-replacement. The proof
exploits a variant of Barwise compactness which holds for suitable
models of "ZC + Sigma_2 Replacement + V=L".
____________
SRI AI SEMINAR SERIES
on Thursday, 30 October 2008, 4:00pm - 5:30pm
EJ228, SRI International
http://www.ai.sri.com/seminars/
"Managing Personal Tasks with Time Constraints and Preferences"
Ioannis Refanidis
University of Macedonia
This talk presents work on solving the problem of managing personal
tasks on top of an electronic calendar application. To solve this
problem, we adopted the Squeaky Wheel Optimization (SWO) framework,
enhanced with powerful heuristics and full constraint propagation. The
scheduling problem involves preemptive and non-preemptive tasks, with
extra constraints imposed on the sizes of and the distances between
the parts of each preemptive task. Travelling times are imposed by
the alternative localization possibilities of each task. Ordering
constraints are imposed by the producer-consumer relations between
tasks. The user may have preferences regarding scheduling options of
single tasks or pairs of tasks. Higher degree time constraints and
preferences could be implemented as well. SWO allows for fast
scheduling and rescheduling. Several heuristics are proposed to
estimate the difficulty to schedule each task and to compensate with
the degree of the user's satisfaction. A prototype system, called
SelfPlanner, has been implemented using a client-server three tier
architecture and embedding Google Calendar and Google Maps
applications.
About the Speaker: Ioannis Refanidis is an assistant professor at the
Department of Applied Informatics, University of Macedonia. He
received his Ph.D. in "Heuristic Planning Systems" from Aristotle
University in 2001. His research interests include planning,
scheduling and constraint satisfaction. His work can be divided in two
areas: The first one concerns heuristic planning systems and resulted
in the GRT planner and several variations of it, whereas the second
one (and more recent) concerns personal time management, which
resulted in the SelfPlanner system. He has published in
ECP/AIPS/ICAPS, ECAI, JAIR and AIJ. In the past he has been involved
in the organization of several ICAPS conferences, as well as the 6th
International Planning Competition, whereas he was PC member at
several main conferences. Currently he serves as conference chair of
ICAPS-2009.
____________
UC BERKELEY CIS SEMINAR
on Thursday, 31 October 2008, 4:00pm-5:30pm
Soda Hall 310 (UC Berkeley)
http://www.eecs.berkeley.edu/CIS/seminars/seminars.html
"Discriminative Methods with Structure"
Simon Lacoste-Julien
UC Berkeley
Real world problems such as machine translation involve complex
dependencies. Generative models have provided an elegant and flexible
framework to model those dependencies, but they appear to lack
robustness to model misspecification compared to discriminative models
for classification. In this talk, we present methods for leveraging
the advantages of generative models in the discriminative framework.
In the first part of the talk, we tackle the word alignment problem
from natural language processing. We formulate it as a weighted
bipartite matching problem and show how to learn the weights by using
a large-margin approach for structured prediction. By providing a
flexible discriminative modeling framework, we were able to cut the
Alignment Error Rate in half compared to the previous best performing
generative models for word alignment.
In the second part of the talk, we study probabilistic topic models
which have been popular for modeling latent structures in text
documents (as bag of words) or images (as bag of visual words). They
are usually trained as generative models with maximum likelihood
estimation, though this could be suboptimal if one is interested in
doing classification. In contrast, we present a discriminative version
of the Latent Dirichlet Allocation (LDA) model which attempts to
uncover the latent structure in the documents while optimizing its
predictive power for the task of classification. We show positive
results on the 20 Newsgroup dataset for document classification.
(joint work with Fei Sha, Ben Taskar, Dan Klein and Michael I. Jordan)
____________
SSP10: SYMBOLIC SYSTEMS FORUM
on Thursday, 30 October 2008, 4:15pm
Bldg. 380:380C (Math Corner)
http://symsys.stanford.edu/ssp_events
"Web Access to Voting Records: Motivations and Issues"
Todd Davies
Symbolic Systems Program
I will describe the thinking behind Who Voted?, a new website where
voter histories that are legally available to the public can be
uploaded and viewed by anyone with an Internet connection. The Who
Voted site complements advocacy aimed at ensuring auditable paper
ballots, addressing the aspect of secret ballot election integrity
that relies on verification of the list of those who voted. It also
aims to promote voting, by making the list of those who voted (but not
who they voted for) easier to see. The road to producing this site
was long, and provoked criticism from some privacy advocates, which in
turn affected the design of the site and what information it makes
available. I will review the reasoning behind the site, and describe
the issues it raises as well as how it addresses them. One issue is
that voter histories for some but not all states can be posted legally
on the Internet, but even those states where the information is public
generally do not post it on the Web. I will discuss this why this is
so, and describe the case for Web access to voting records. (Joint
work with Jeffrey Gerard, Gordon Lyon, and Reid Chandler)
____________
BERKELEY INFORMATION ACCESS SEMINAR
on Friday, 31 October 2008, 3:00pm - 5:00pm
107 South Hall (Berkeley)
http://courses.ischool.berkeley.edu/i296a-1/f08/schedule.html
"The Extent of Geographic Resources Available on the Web"
Robert Pasley
Sheffield Univ., U.K.
We describe a methodology to estimate the extent of geographic
resources available on the web without the need for secondary
knowledge or complex geo-tagging. This is achieved by randomly
selecting toponyms from the Ordnance Survey 50K gazetteer to create
search queries and thus gather document counts from various web
sources for Great Britain. The same gazetteer is then used to geo-code
the results and enable mapping.
____________
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
sympa@lists-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). You will be asked to confirm
the subscription in either case. To unsubscribe use the word
unsubscribe instead of subscribe. Problems with subscribing or
unsubscribing should be sent to incalendar@csli.stanford.edu
The full current issue is at
http://cslicalendar.stanford.edu/current.shtml
and the archives at
http://cslicalendar.stanford.edu/Archive/
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://cslicalendar.stanford.edu/locations.shtml