Copy Link
Add to Bookmark
Report
AIList Digest Volume 4 Issue 149
AIList Digest Monday, 16 Jun 1986 Volume 4 : Issue 149
Today's Topics:
Seminars - Possible Worlds Planning (SRI) &
Automatic Expert System Induction (NASA Ames) &
Learning by Selection (CMU) &
Connectionist Knowledge Representation System (CMU) &
Object Recognition using Category Models (UPenn) &
CODER Information Retrieval (VPI),
New Society - Bay Area AI and Education Meeting
----------------------------------------------------------------------
Date: Wed 11 Jun 86 11:27:30-PDT
From: Amy Lansky <LANSKY@SRI-WARBUCKS.ARPA>
Subject: Seminar - Possible Worlds Planning (SRI)
POSSIBLE WORLDS PLANNING
Matt Ginsberg (SJG@SAIL)
Stanford University
11:00 AM, MONDAY, June 16
SRI International, Building E, Room EJ228 (new conference room)
The size of the search space is perhaps the most intractable of all of
the problems facing a general-purpose planner. Some planning methods
(means-ends analysis being typical) address this problem by
encouraging the system designer to give the planner domain-specific
information (perhaps in the form of a difference table) to help govern
this search.
This paper presents a domain-independent approach to this problem
based on the examination of possible worlds in which the planning goal
has been achieved. Although a weak method, the ideas presented lead
to considerable savings in many examples; in addition, the natural
implementation of this approach has the attractive property that
incremental efforts in controlling the search provide incremental
improvements in performance. This is in contrast to many other
approaches to the control of search or inference, which may require
large expenditures of effort before any benefits are realized.
VISITORS: Please arrive 5 minutes early so that you can be escorted up
from the E-building receptionist's desk. Thanks!
------------------------------
Date: Thu, 12 Jun 86 00:18:33 pdt
From: eugene@AMES-NAS.ARPA (Eugene Miya)
Subject: Seminar - Automatic Expert System Induction (NASA Ames)
Subject: June 17, 1986, NASA Ames AI Forum, Automatic Induction
National Aeronautics and Space Administration
Ames Research Center
AMES AI FORUM
SEMINAR ANNOUNCEMENT
SPEAKER: Dr. Peter Cheeseman
Information Sciences Office
NASA Ames Research Center
TOPIC: Automatic Induction of Probabilistic Expert Systems
Many have realized that expert systems that make decisions under uncertainty
must represent this uncertainty and manipulate it correctly. This cannot be
done in general by "symbolic" (i.e. non-numeric) methods or by sprinkling
numbers over logical inference, as advocated by many authors in AI. Probability
has been proved to be the only consistent inference scheme if uncertainty is
represented by a real number. Probabilistic inference requires assessing the
effect of ALL the relevant evidence on the hypothesis of interest through ALL
the possible chains of inference (rather than establishing a single path from
axioms to theorem, as in logic). However, some methods used in probabilistic
inference in AI (e.g. Prospector) impose strong constraints on the structure of
the information (e.g. conditional independence) or require large amounts of
information. The solution to this problem is to use Maximum Entropy to spread
the uncertainty over the set of possibilities as evenly as possible consistent
with the known information. A computationally efficient method for performing
the maximum entropy calculation will be presented as well as a method for
extracting the necessary probabilistic information directly from data. The
result is a complete probabilistic expert system without using an expert.
DATE: Tuesday, TIME: 10:30-11:30 am BLDG. 239 Room B39
June 17, 1986 (Basement Conf. Room)
POINT OF CONTACT: Alison Andrews PHONE NUMBER: (415)694-6741
NET ADDRESS: mer.andrews@ames-vmsb.ARPA
VISITORS ARE WELCOME: Register and obtain vehicle pass at Ames Visitor
Reception Building (N-253) or the Security Station near Gate 18. Do not
use the Navy Main Gate.
Non-citizens (except Permanent Residents) must have prior approval from the
Director's Office one week in advance. Submit requests to the point of
contact indicated above. Non-citizens must register at the Visitor
Reception Building. Permanent Residents are required to show Alien
Registration Card at the time of registration.
------------------------------
Date: 12 June 1986 1156-EDT
From: Richard Wallstein@A.CS.CMU.EDU
Subject: Seminar - Learning by Selection (CMU)
The CMU Summer Research Seminar Series continues this Friday, June 13 at
2:30 PM, 7500 WeH with a talk by Geoffrey Hinton on his new research:
A New Algorithm for Learning by Selection
Imagine a complicated non-linear process that contains specific steps that are
controlled by switches which can be on or off. Each switch has a particular
stored probability of being on. Using these probabilities, we generate a
random combination of switch settings and then run the process and decide
whether the result is good or bad. I shall describe a new learning algorithm
that uses information about the goodness of the outcomes to revise the stored
probabilities associated with the switches. The algorithm is guaranteed to
change the switch probabilites in such a way that future random combinations of
switch settings are more likely to produce good outcomes. It can be applied to
stochastic processes of arbitrary complexity. If each switch is a synapse, it
suggests a new model of learning in the cortex. If each switch is an enzyme
and its stored probability is the relative frequency of the relevant gene in
the gene pool, the learning algorithm is an efficient way of using the
information provided by survival to optimize gene frequencies. The extension to
optimizing frequencies of gene combinations appears to be feasible.
------------------------------
Date: 11 Jun 86 01:17:06 EDT
From: Mark.Derthick@g.cs.cmu.edu
Subject: Seminar - Connectionist Knowledge Representation System (CMU)
I will present my thesis proposal, "A Connectionist Knowledge Representation
System," 2pm Wednesday, June 18, in 5409.
I propose to develop a knowledge representation system that is functionally
similar to KL2, but implemented on a parallel, non-symbolic architecture.
Answering queries is carried out by a Boltzmann Machine
network in which concepts, roles, and individuals are represented by
patterns of activity of very simple processing units. By choosing good
representations, a small network suffices to capture the knowledge as
pairwise interactions among the units in the network. A single parallel
constraint satisfaction search accomplishes the answering process. I will
prove that for any definable knowledge base, the network constructed will
answer queries as specified by the formal knowledge level semantics.
------------------------------
Date: Wed, 11 Jun 86 14:05 EDT
From: Tim Finin <Tim%upenn.csnet@CSNET-RELAY.ARPA>
Subject: Seminar - Object Recognition using Category Models (UPenn)
OBJECT RECOGNITION USING FUNCTION BASED CATEGORY MODELS
Ph. D. Thesis Proposal
Franc Solina
GRASP Laboratory
UNIVERSITY of PENNSYLVANIA
Department of Computer and Information Sciences
Philadelphia, PA 19104-6389
Phone (215) 898 8298
Net address: franc@upenn
We propose a modeling system for recognition of generic
objects. Based on the observation that fulfilling of the
same function results in similar shapes we will consider
object categories that are formed around the principle of
functionality. The representation consists of a prototypi-
cal object represented by prototypical parts and relations
between these parts. Parts are modeled by superquadric
volumetric primitives which are combined via boolean opera-
tions to form objects. Variations between objects within a
category are described by allowable changes in structure and
shape deformations of prototypical parts. Each prototypical
part and relation has a set of associated features that can
be recognized in the images. The recognition process
proceeds as follows; the input is a pair of stereo reflec-
tance images. The closed contours and sparse 3-D points,
the result of low level vision, are analyzed to find domain
specific features. These features are used for indexing the
model data base to make hypotheses. The selected hypotheses
are then verified on the geometric level by deforming the
prototype in allowable way to match the data. We base our
design of the modeling system upon the current psychological
theories of the human visual perception.
advisor: R. Bajcsy
commitee: N. Badler, H. ElGindy, J. Kender (Columbia University).
Time: Monday, June 16, 11 PM, room 216
------------------------------
Date: Tue, 27 May 86 10:31:37 edt
From: vtcs1::fox
Subject: Seminar - CODER Information Retrieval (VPI)
[Forwarded from IRList Digest V2#26 by Laws@SRI-AI.]
The M.S. defense of Robert K. France will be held at 10am Monday June 2 in
Norris 301. The title of his thesis is "An Artificial Intelligence Environment
for Information Retrieval Research."
The CODER (COmposite Document Expert/extended/effective Retrieval)
project is a multi-year effort to investigate how best to apply
artificial intelligence methods to increase the effectiveness of
information retrieval systems. Particular attention is being given to
analysis and representation of heterogeneous documents, such as
electronic mail digests or messages, which vary widely in style,
length, topic, and structure. In order to ensure system adaptability
and to allow reconfiguration for controlled experimentation, the
project has been designed as a moderated expert system. This thesis
covers the design problems involved in providing a unified
architecture and knowledge representation scheme for such a system,
and the solutions chosen for CODER. An overall object-oriented
environment is constructed using a set of message-passing primitives
based on a modified Prolog call paradigm. Within this environment is
embedded the skeleton of a flexible expert system, where task
decomposition is performed in a knowledge-oriented fashion and where
subtask managers are implemented as members of a community of experts.
A three-level knowledge representation formalism of elementary data
types, frames, and relations is provided, and can be used to construct
knowledge structures such as terms, meaning structures, and document
interpretations. The use of individually tailored specialist experts
coupled with standardized blackboard modules for communication and
control and external knowledge bases for maintenance of factual world
knowledge allows for rapid prototyping, incremental development, and
flexibility under change. The system as a whole is structured as a
set of communicating modules, defined functionally and imple- mented
under UNIX using sockets and the TCP/IP protocol for communication.
Inferential modules are being coded in MU-Prolog; non-inferential
modules are being prototyped in MU-Prolog and will be re-implemented
as needed in C++.
Host: Dr. Edward A. Fox, Dept. of Computer Science
------------------------------
Date: Fri 13 Jun 86 11:45:06-PDT
From: Mark Richer <RICHER@SUMEX-AIM.ARPA>
Subject: New Society - Bay Area AI and Education Meeting
Date: 13 Jun 86 10:27 PDT
From: dmrussell.pa@Xerox.COM
Subject: Bay Area AI and Education Meeting: June 23rd, 6PM, PARC
What: Bay Area AI and Education Group holding its first meeting.
Where: Xerox Palo Alto Research Center (PARC)
3333 Coyote Hill Rd.
Palo Alto, CA
(send for detailed directions)
When: June 23rd, 6PM
Who:
Speakers: Jim Greeno and Peter Pirolli
"Some New Directions in the Science of
Instructional Design"
Math Science and Technology
Education Dept.
University of Calif. Berkeley
Host: Daniel Russell
Intelligent Systems Lab
PARC
Amplification:
BARRET (Bay ARea Research in Educational Technology) is an
attempt to bring together many of the local people working in the area
of applying AI to education. There are significant efforts at
Berkeley, Stanford, UCSF, SRI, PARC and so on. BARRET is a way of
establishing some communication between the various groups, by hosting
technical talks on this topic and setting aside time for informal
discussion.
To do this, BARRET will be implemented as a moving sequence of talks
circulating throughout the Bay Area on a (roughly) monthly basis. We
hope to have high quality talks on areas of mutual interest to be
followed by an equally high-quality dinner that will allow us to meet
and discuss topics further.
This first meeting of BARRET will be followed by dinner at Chef Chu's,
assuming that we can get a reasonable headcount. (With enough warning,
non-MSG-ers and veggies can be accomodated.)
So, if you are interested in attending, please message (or call) me and
let me know of your intentions. That will allow us to do some planning
for our first meeting.
-- Dan Russell --
ArpaNet: DMRussell.PA@XEROX.COM
Phone: (415)-494-4308
Mail: Dan Russell
ISL
3333 Coyote Hill Rd.
Palo Alto, CA 94304
------------------------------
End of AIList Digest
********************
