[Prev][Index]
CSLI Calendar, 8 June 1995, vol.10:31
-
To: friends
-
Subject: CSLI Calendar, 8 June 1995, vol.10:31
-
From: Tom Burke <burke>
-
Date: Wed, 7 Jun 1995 13:38:01 -0700
C S L I C A L E N D A R O F P U B L I C E V E N T S
______________________________________________________________________________
8 June 1995 Stanford Vol. 10, No. 31
______________________________________________________________________________
A weekly publication of the Center for the Study of Language and
Information (CSLI), Stanford University, Ventura Hall, Stanford, CA 94305-4115
____________
CSLI ACTIVITIES DURING 7 -- 16 JUNE 1995
WEDNESDAY, 7 JUNE
4:00 - Seminar on Computational Learning and Adaptation
Cordura Hall, Room 100
Speechreading by Computers
Marcus E. Hennecke, Ricoh California Research Center
Abstract below
THURSDAY, 8 JUNE
12:15 - CSLI Seminar
Series on Intelligent Agents
Cordura Hall, Room 100
Intelligent Agents for Interactive Simulation Environments:
Constraints and Design Choices
Milind Tambe, USC Information Sciences Institute
Abstract below
2:00 - SSP Forum
Building 60, Room 61-F
1994-95 Senior Honors Projects
Clay Kunz, Kjetil Larsen, Jon Lindsay
Abstracts below
2:15 - CSLI Seminar
Movement of Technology from University to Industry
Cordura Hall, Room 100
Artificial Intelligence, Agents, and Industry
Peter Norvig, Harlequin Inc.
Abstract below
4:00 - Stanford Humanities Review Open-House
Humanities Center Annex, 579 Alvarado Row
Publication of _Constructions of the Mind: Artificial
Intelligence and the Humanities_
Hosts: Wanda Corn and John Etchemendy, Stanford Humanities
FRIDAY, 9 JUNE
12:00 - Linguistics Seminar
Building 110, Room 111-A
The History of Counterfactual Conditionals in English
Rafal Molencki, Stanford Linguistics
12:00 - Logic Lunch
Building 380, Room 383-N
Formal, Informal, and Mechanical Proofs
Natarajan Shankar, SRI International
Abstract below
MONDAY, 12 JUNE
12:00 - CSLI Seminar
Movement of Technology from University to Industry
Cordura Hall, Room 100
Importance of Students versus Research in Private Industry
James F. Gibbons, Stanford Engineering
Abstract below
WEDNESDAY, 14 JUNE
4:00 - Seminar on Computational Learning and Adaptation
Cordura Hall, Room 100
Frameworks For Visual Learning
John Weng, Michigan State Computer Science
Abstract below
THURSDAY, 15 JUNE
12:15 - CSLI Seminar
Series on Intelligent Agents
Cordura Hall, Room 100
Network-based Information Brokers
Adam Farquhar, Stanford Knowledge Systems Laboratory
Abstract below
2:15 - CSLI Seminar
Movement of Technology from University to Industry
Cordura Hall, Room 100
Title to be announced
David Fuchs, Frame Technology
____________
The CSLI Calendar appears on Thursday of each week throughout the academic
year. Announcements, abstracts, and other information to appear in the
Calendar on a given Thursday should be submitted by e-mail to
<incalendar@csli.stanford.edu> by 5:00 p.m. on the previous Tuesday.
Past issues of the CSLI Calendar, a quarterly schedule of upcoming CSLI
events, and other information about CSLI are available at URL
<http://www-csli.stanford.edu/>. The Calendar, with available abstracts, is
also posted each week to the csli.bboard newsgroup.
____________
NOTE: This is the final issue of the CSLI Calendar for the 1994-95 academic
year. The first issue for 1995-96 will appear later in September 1995.
____________
SEMINAR ON COMPUTATIONAL LEARNING AND ADAPTATION
on Wednesday, 7 June
4:00 p.m., Cordura Hall, Room 100
Speechreading by Computers
Marcus E. Hennecke
Ricoh California Research Center
The visual image of a talker provides information complementary to the
acoustic speech waveform, and enables improved recognition accuracy,
especially in environments corrupted by high acoustic noise or multiple
talkers. In this talk I will present a complete speechreading system, which
is able to record an utterance using a standard color video camera, preprocess
both the audio and video signal, and perform speech recognition. This system
is based on new algorithms for finding the speaker's face and mouth and an
improved template algorithm for tracking the lips.
____________
CSLI SEMINAR
SERIES ON INTELLIGENT AGENTS
on Thursday, 8 June
12:15 p.m., Cordura Hall, Room 100
Intelligent Agents for Interactive Simulation Environments:
Constraints and Design Choices
Milind Tambe
USC Information Sciences Institute
<tambe@isi.edu>
<http://www.isi.edu/soar/tambe/>
Interactive simulation (virtual reality) environments are becoming ubiquitous,
finding applications in areas such as education, training, entertainment, and
manufacturing. Our current research effort is aimed at developing human-like,
intelligent agents (virtual humans) that can interact with each other, as well
as with humans in such virtual environments. These environments have
sufficiently high fidelity and realism that constructing intelligent agents
requires us to face hard research challenges in a variety of specialized agent
capabilities, as well as in their integration; but without the costs and
demands of low-level perceptual processing or robotic control.
To begin our effort, we have focused on the development of intelligent pilot
agents for real-world battlefield simulation environments (as part of a
collaborative effort involving researchers at Information Sciences Institute,
University of Southern California, and at University of Michigan). These
agents are based on the Soar/IFOR system, constructed using the Soar
integrated architecture. They have already participated in simulated combat
against human pilots -- most recently in November 1994, in a large-scale
simulated military exercise.
In this talk, I will provide an overview of the Soar/IFOR project (with video
tapes of pilot agents in simulated helicopters and fighter jets). I will also
discuss the pilot agent design in terms of the two sets of constraints that
influence it, specifically, (i) "top-down" constraints, i.e., capabilities
that pilot agents must possess if they are to succeed in this environment;
(ii) "bottom-up" constraints, i.e., the discipline enforced by the Soar
architecture. If time permits, I will also discuss research on specialized
agent capabilities, such as real-time agent modeling.
MILIND TAMBE is currently a research computer scientist at the Information
Sciences Institute at University of Southern California (USC) and a research
assistant professor with the computer science department at USC. He completed
his undergraduate education in computer science from the Birla Institute of
Technology and Science, Pilani, India, in 1986. He received his Ph.D. in 1991
from the School of Computer Science at Carnegie Mellon University.
____________
SYMBOLIC SYSTEMS FORUM
(*NOTE THE TIME CHANGE*)
on Thursday, 8 June
2:00 p.m., Building 60, Room 61-F
1994-95 Senior Honors Projects
Clay Kunz, Kjetil Larsen, Jon Lindsay
(1) Clay Kunz: "Learning Together -- Tools for Choosing and Training
Neural Networks in Parallel"
Artificial Neural Networks, like their biological counterparts, are highly
distributed entities. This project addresses the distributed nature of neural
network training at two levels. The first is at the level of individual
neural net training, where the training set can be split between several
processors on a parallel computer, providing gross speed increases. The
second level of distribution lies in the choice of neural network
architecture, and in the choice of parameters given to the training program.
At this level, a parallel computer can be made to train several networks
simultaneously on the same training set, selectively focusing on the networks
that perform the best (using a fitness-based approach inspired from the field
of genetic algorithms), and removing difficult and often uninformed choices
about neural net topology and parameter choice from the user. My thesis
includes an implementation of the first algorithm on a computer at the
Lawrence Livermore National Lab, and a simulation of the second on a
Sparcstation at Stanford.
(2) Kjetil Larsen: "Cooperative Interfaces"
Human computer interaction today is like spoon feeding. Computers are not very
cooperative or understanding -- they require the users to break up their task
into small pieces corresponding to the tools available and frequently force
the users to care about details too specific for their goals. Similarly, the
feedback from the computer is often irrelevant or intelligible.
My thesis is an attempt to provide interface designers with guidelines for a
kind of interaction that is more cooperative, inspired by human- human
interaction -- conversation. It is not an implementation of such an interface,
but a handbook for design. My goal is to outline what such an interface is,
why it is useful, and how it is possible.
(3) Jon Lindsay: "The Biology of Being-in-the-World"
The modern information processing paradigm for cognitive science presupposes a
strong division between subject and object, between the representational
homunculus "in here" and the real, objective world of objects, features and
events "out there". In the early 20th century, philosopher Martin Heidegger
challenged this subject/object metaphysics, arguing that humans were more
fundamentally "being-in-the-world", an active embodied engagement with the
world that was "before" the division between subject and object, or agent and
world. His philosophical criticism has been largely unheeded by cognitive
science, however, perhaps because it seems too "fuzzy", or information
processing just seems obvious.
In this project, I took Heidegger's description of being-in-the-world as a
sort of heuristic that a good scientific account of the human being should
satisfy. Then I argued that a paradigm of self-organization, rather than
representation/information processing, could accommodate being-in-the-world.
Specifically, I looked at Humberto Maturana and Francisco Varela's notion of
autopoiesis as a good biological description of the type of machine we would
have to be in order to be being-in-the-world, rather than a representing
subject knowing objects.
____________
CSLI SEMINAR
MOVEMENT OF TECHNOLOGY FROM UNIVERSITY TO INDUSTRY
on Thursday, 8 June
2:15 p.m., Cordura Hall, Room 100
Artificial Intelligence, Agents, and Industry
Peter Norvig
Harlequin, Inc.
<norvig@harlequin.com>
In this talk, I discuss a definition of and approach to AI (as set forth in
the book "Artificial Intelligence: A Modern Approach"). I will say why this
seemingly academic exercise is of interest to industry, and why it is of more
interest (to some) than the dominant approach of the 1980s. Examples from
several companies in industries such as manufacturing, telecommunication, and
software will be given, with technologies taken from machine learning, natural
language processing, and decision theory.
____________
STANFORD HUMANITIES REVIEW OPEN-HOUSE
on Thursday, 8 June
4:00 p.m., Humanities Center Annex, 579 Alvarado Row
Wanda Corn, Director of Stanford Humanities Center, and John Etchemendy,
Senior Associate Dean of Humanities and Sciences, invite you to an open-house
at the Stanford Humanities Center Annex to celebrate the publication of the
latest issue of the _Stanford Humanities Review_:
Constructions of the Mind: Artificial Intelligence and the Humanities
and to present ongoing work for prospective issues. A presentation of _SEHR_,
the new cyber-incarnation of the Stanford Humanities Review, will follow.
Wine & cheese and other refreshments will be served.
____________
LINGUISTICS SEMINAR
on Friday, 9 June
12:00 noon, Building 110, Room 111-A
The History of Counterfactual Conditionals in English
Rafal Molencki
Stanford Linguistics
____________
LOGIC LUNCH
on Friday, 9 June
12:00 noon, Building 380, Room 383-N
Formal, Informal, and Mechanical Proofs
Natarajan Shankar
SRI International Computer Science Lab
<shankar@csl.sri.com>
<http:://www.csl.sri.com/shankar.html>
Careful scrutiny will frequently reveal serious flaws even in a seemingly
convincing and rigorous informal argument. Formalization is one way to
scrutinize and verify an informal argument but it can be tedious,
labor-intensive, and itself prone to error. A mechanical proof is a computer
representation of an argument that has been constructed and checked with the
aid of a computer program. The point of mechanization is to bridge the gap
between formal and informal proofs so that the existence of a correct formal
proof can be easily demonstrated without sacrificing the cogency of the
original informal argument.
In this (informal) talk, we will discuss the extent to which mechanical proofs
have succeeded in bridging this gap. In particular, we will examine the case
of the arithmetic-geometric mean problem (communicated by Alan Bundy) and its
proof using the PVS proof checker. The problem is to show that given N
non-negative real numbers, their arithmetic mean (the sum of the N numbers
divided by N) is never smaller than their geometric mean (the Nth root of the
product of the N numbers). Attendees are urged to solve this problem prior to
the talk, and to keep a record of the mental processes involved.
____________
CSLI SEMINAR
MOVEMENT OF TECHNOLOGY FROM UNIVERSITY TO INDUSTRY
on Monday, 12 June
12:00 noon, Cordura Hall, Room 100
Importance of Students versus Research in Private Industry
James F. Gibbons
Stanford Engineering
<gibbons@soe.stanford.edu>
A large sample of high tech companies in Silicon Valley have been analyzed to
elucidate the basic requirements for a successful venture, the time required
to make a judgment about success, and the approach that a venture capitalist
would take in judging the investment. These elements will be presented
together with some economic histories of firms that were started by graduates
of the School of Engineering over the last 50 years.
JAMES F. GIBBONS is Dean of the School of Engineering at Stanford.
____________
SEMINAR ON COMPUTATIONAL LEARNING AND ADAPTATION
on Wednesday, 7 June
4:00 p.m., Cordura Hall, Room 100
Frameworks For Visual Learning
John Weng
Michigan State Computer Science
<weng@cps.msu.edu>
This talk discusses some major difficulties encountered by currently
prevailing approaches to computer vision and describes the approach of
comprehensive visual learning, which involves two key ideas: that learning
must comprehensively cover the visual world; and that it must comprehensively
cover the vision system. Some critical issues to be discussed include: the
problem of high dimensionality; the automatic selection of features; learning
invariance to size, position, and orientation; segmentation of objects from
cluttered background; system self-organization through learning; indexing into
a large visual knowledge base; decision making; and the coordination of
learning with sensing and control. A new framework called SHOSLIF will be
described, along with preliminary results in applying it to the problems of
face recognition, hand-sign recognition, and autonomous navigation. Its
predecessor, the Cresceptron, will also be briefly discussed.
____________
CSLI SEMINAR
SERIES ON INTELLIGENT AGENTS
on Thursday, 15 June
12:15 p.m., Cordura Hall, Room 100
Network-based Information Brokers
Adam Farquhar
Stanford Knowledge Systems Lab
<axf@ksl.stanford.edu>
We describe a new project whose objective is to develop key technologies that
will enable vendors and buyers to build and maintain network-based information
brokers capable of retrieving information about services and products via the
Internet from multiple vendor catalogs and data bases for both human and
computer-based clients.
The ability to obtain relevant information in a timely and cost efficient
manner is central to the effective performance of most tasks in our society.
The widespread availability of computer-based information brokers will provide
that ability to large communities by significantly facilitating effective
access to the broad range of information that is rapidly becoming available on
the Internet. The general availability of the technology to build and
maintain information brokers will enable an industry to be established whose
primary products are computer-based network-accessible brokering services.
In order to accomplish this goal, it is essential to reduce the cost of
integrating information sources and to provide a path that allows for
incremental integration that can be responsive to users' demands. We present
an approach to integrating disparate heterogeneous information sources that
uses context logic. Our use of context logic reduces the up-front cost of
integration, provides an incremental integration path, and allows semantic
conflicts within a single information sources or between information sources
to be expressed and resolved.
____________
CSLI SEMINAR
MOVEMENT OF TECHNOLOGY FROM UNIVERSITY TO INDUSTRY
on Thursday, 15 June
2:15 p.m., Cordura Hall, Room 100
TeXno Babel
David Fuchs
Frame Technology
<drf@frame.com>
Don Knuth's ubiquitious TeX and Metafont systems, written in the mid-70's to
mid-80's, are the product of very different hardware and software environments
than what are available today. For instance, laser printers didn't yet exist,
nor workstations or PCs. Most development was on SAIL, a DEC 10 machine with
a process address space limit of only 256K 36-bit words. It did have custom
bitmap graphics terminals, which were unusual for the time.
This talk discusses that environment, how it changed over time, and how many
of the bits and pieces represented have evolved into multi-billion dollar
industries.
____________