Copy Link
Add to Bookmark
Report
AIList Digest Volume 4 Issue 204
AIList Digest Monday, 6 Oct 1986 Volume 4 : Issue 204
Today's Topics:
Seminars - Connectionist Networks (UPenn) &
Automatic Class Formation (SRI) &
Computers are not Omnipotent (CMU) &
Automating Diagnosis (CMU) &
Temporal Logic (MIT) &
Program Transformations and Parallel Lisp (SU) &
Temporal Theorem Proving (SU) &
Efficient Unification of Quantified Terms (MIT) &
Planning Simultaneous Actions (BBN) &
Cognitive Architecture (UPenn)
----------------------------------------------------------------------
Date: Mon, 29 Sep 86 14:52 EDT
From: Tim Finin <Tim@cis.upenn.edu>
Subject: Seminar - Connectionist Networks (UPenn)
CONNECTIONIST NETWORKS
Jerome A. Feldman
Computer Science Department
University of Rochester
There is a growing interest in highly interconnected networks of very simple
processing elements within artificial intelligence circles. These networks are
referred to as Connectionist Networks and are playing an increasingly important
role in artificial intelligence and cognitive science. This talk briefly
discusses the motivation behind pursuing the the connectionist approach, and
discusses a connectionist model of how mammals are able to deal with visual
objects and environments. The problems addressed include perceptual
constancies, eye movements and the stable visual world, object descriptions,
perceptual generalizations, and the representation of extrapersonal space.
The development is based on an action-oriented notion of perception. The
observer is assumed to be continuously sampling the ambient light for
information of current value. The central problem of vision is taken to be
categorizing and locating objects in the environment. The critical step in
this process is the linking of visual information to symbolic object
descriptions, i.e., indexing. The treatment focuses on the different
representations of information used in the visual system. The model employs
four representation frames that capture information in the following forms:
retinotopic, head-based, symbolic, and allocentric.
The talk ends with a discussion of how connectionist models are being realized
on existing architectures such as large multiprocessors.
Thursday, October 2, 1986
Room 216 - Moore School
3:00 - 4:30 p.m.
Refreshments Available
Faculty Lounge - 2:00 - 3:00 p.m.
------------------------------
Date: Wed 1 Oct 86 11:46:40-PDT
From: Amy Lansky <LANSKY@SRI-VENICE.ARPA>
Subject: Seminar - Automatic Class Formation (SRI)
PROBABILISTIC PREDICTION THROUGH AUTOMATIC CLASS FORMATION
Peter Cheeseman (CHEESEMAN@AMES-PLUTO)
NASA Ames Research Center
11:00 AM, MONDAY, October 6
SRI International, Building E, Room EJ228
A probabilistic expert system is a set of probabilistic connections
(e.g. conditional or joint probabilities) between the known variables.
These connections can be used to make (conditional) probabilistic
predictions for variables of interest given any combination of known
variable values. Such systems suffer a major computational problem---
once the probabilstic connections form a complex inter-connected
network, the cost of performing the necessary probability calculations
becomes excessive. One approach to reducing the computational
complexity is to introduce new "variables" (hidden causes or dummy
nodes) that decouple the interactions between the variables. Judea
Pearl has described an algorithm for introducing sufficient dummy
nodes to create a tree structure, provided the probabilistic
connections satisfy certain (strong) restrictions. This talk will
describe a procedure for finding only the significant "hidden causes",
that not only lead to a computationally simple procedure, but subsume
all the significant interactions between the variables.
VISITORS: Please arrive 5 minutes early so that you can be escorted up
from the E-building receptionist's desk. Thanks!
------------------------------
Date: 28 Sep 1986 1228-EDT
From: David A. Evans <DAE@C.CS.CMU.EDU>
Subject: Seminar - Computers are not Omnipotent (CMU)
PHILOSOPHY COLLOQUIUM ANNOUNCEMENT:
COMPUTERS ARE NOT OMNIPOTENT
David Harel
Weizmann Institute
and
Carnegie Mellon University
Monday, October 6 4:00 p.m.
Porter Hall 223D
In April, 1984, TIME magazine quoted a computer professional as saying:
"Put the right kind of sofware into a computer and it will do
whatever you want it to. There may be limits on what you can
do with the machines themselves, but there are no limits on
what you can do with the software."
In the talk we shall disprove this contention outright, by exhibiting a
wide array of results obtained by mathematicians and computer scientists
between 1935 and 1983. Since the results point to inherent limitations of
any kind of computing device, even with unlimited resources, they appear
to have interesting philosophical implications concerning our own
limitations as entities with finite mass.
------------------------------
Date: 29 September 1986 2247-EDT
From: Masaru Tomita@A.CS.CMU.EDU
Subject: Seminar - Automating Diagnosis (CMU)
Date: 10/7 (Tuesday)
Time: 3:30
Place: WeH 5409
Some AI Applications at Digital
Automating Diagnosis: A case study
Neil Pundit
Kamesh Ramakrishna
Artificial Intelligence Applications Group
Digital Equipment Corporation
77 Reed Road (HLO2-3/M10)
Hudson, Massachusetts, 01749
The Artificial Intelligence Applications Group at Digital is engaged in the
development of expert systems technology in the context of many real-life
problems drawn from within the corporation and those of customers. In
addition, the group fosters basic research in AI by arrangements with
leading universities. We plan to briefly describe some interesting
applications. However, to satisfy your appetite for technical content, we
will describe in some detail our progress on Beta, a tool for automating
diagnosis.
The communication structure level is a knowledge level at which certain
kinds of diagnostic reasoning can occur. It is an intermediate level between
the level at which current expert systems are designed (using knowledge
acquired from experts) and the level at which ``deep reasoning'' systems
perform (based on knowledge of structure, function, and behavior of the
system being diagnosed). We present an example of an expert system that was
designed the old-fashioned way and the heuristics that were the basis for
recognizing the existence of the communication structure level.
Beta is a language for specifying the communication structure of a system so
that these heuristics can be compiled into a partially automatically
generated program for diagnosing system problems. The current version of
Beta can handle a specific class of communication structure that we call a
``control hierarchy'' and can analyze historical usage and error data
maintained as a log file. The compiler embeds the heuristics in a generated
mix of OPS5 and C code. We believe that Beta is a better way for designers
and programmers who are not AI experts to express their knowledge of a
system than the current rule-based or frame-based formalisms.
------------------------------
Date: Thu, 2 Oct 86 15:12:49 EDT
From: "Elisha P. Sacks" <elisha%MX.LCS.MIT.EDU@MC.LCS.MIT.EDU>
Subject: Seminar - Temporal Logic (MIT)
E. Taatnoon
"The Third Cybernetics and Temporal Logic"
I aim to link up the concepts of system bifurcation and system
catastrophe with temporal logic in order to show the applicability of
dialectical reasoning to metamorphic system transformations. A system
catastrophe is an innovation resulting from reorganization resulting
from a switch from positive to negative feedback or vice versa. The
subsystems would then be oscillators and the truth of any descriptive
statement is then distributive. Such oscillations would produce an
uncertainty in the temporal trajectory of the system which would
increase both towards the past and the future. This means that time
is not a scalar dimension, but a quadratic paraboloid distribution of
converging and diverging transition probabilities. A social system
composed of such oscillators would be heterarchical rather than
hierarchical.
Refreshments.
Hosts: Dennis Fogg and Boaz Ben-Zvi
Place: 8th Floor Playroom
Time: Noon
------------------------------
Date: 30 Sep 86 0947 PDT
From: Carolyn Talcott <CLT@SAIL.STANFORD.EDU>
Subject: Seminar - Program Transformations and Parallel Lisp (SU)
Speaker: James M. Boyle, Argonne National Laboratory
Time: Monday, October 6, 4pm
Place: 252 Margaret Jacks (Stanford Computer Science Dept)
Deriving Parallel Programs
from Pure LISP Specifications
by Program Transformation
Dr. James M. Boyle
Mathematics and Computer Science Division
Argonne National Laboratory
Argonne, IL 60439-4844
boyle@anl-mcs.arpa
How can one implement a "dusty deck" pure Lisp program on global-
memory parallel computers? Fortunately, pure Lisp programs have a declara-
tive interpretation, which protects their decks from becoming too dusty!
This declarative interpretation means that a pure Lisp program is not
over-specified in the direction of sequential execution. Thus there is
hope to detect parallelism automatically in pure Lisp programs.
In this talk I shall describe a stepwise refinement of pure Lisp pro-
grams that leads to a parallel implementation. From this point of view,
the pure Lisp program is an abstract specification, which program transfor-
mations can refine in several steps to a parallel program. I shall
describe some of the transformations--correctness preserving rewrite rules
--used to carry out the implementation.
An important property of a parallel program is whether it can
deadlock. I shall discuss a number of the design decisions involved in the
refinement and their role in preserving the correctness of the transformed
program, especially with regard to deadlock.
Implementing a transformational refinement often leads to interesting
insights about programming. I shall discuss some of these insights,
including one about the compilation of recursive programs, and some that
suggest ways to systematically relax the "purity" requirement on the Lisp
program being implemented.
We have used this approach to implement a moderately large pure Lisp
program (1300 lines, 42 functions) on several parallel machines, including
the Denelcor HEP (r.i.p.), the Encore Multimax, the Sequent Balance 8000,
and the Alliant FX/8. I shall discuss some measurements of the performance
of this program, which has achieved a speedup of 12.5 for 16 processors on
realistic data, and some of the optimizations used to obtain this perfor-
mance.
Oh, yes, and by the way, the transformations produce a parallel pro-
gram in FORTRAN!
------------------------------
Date: 01 Oct 86 1134 PDT
From: Martin Abadi <MA@SAIL.STANFORD.EDU>
Subject: Seminar - Temporal Theorem Proving (SU)
PhD Oral Examination
Wednesday, October 8, 2:15 PM
Margaret Jacks Hall 146
Temporal Theorem Proving
Martin Abadi
Computer Science Department
In the last few years, temporal logic has been applied in the
specification, verification, and synthesis of concurrent programs, as
well as in the synthesis of robot plans and in the verification of
hardware devices. Nevertheless, proof techniques for temporal logic
have been quite limited up to now.
This talk presents a novel proof system R for temporal logic. Proofs are
generally short and natural. The system is based on nonclausal resolution,
an attractive classical logic method, and involves a special treatment of
quantifiers and modal operators.
Unfortunately, no effective proof system for temporal logic is
complete. We examine soundness and completeness issues for R and other
systems. For example, a simple extension of our resolution system is
as powerful as Peano Arithmetic. (Fortunately, refreshments will
follow the talk.)
Like classical resolution, temporal resolution suggests an approach to
logic programming. We explore temporal logic as a programming language
and a temporal resolution theorem prover as an interpreter for
programs in this language.
Other modal logics have found a variety of uses in artificial
intelligence and in the analysis of distributed systems. We discuss
resolution systems analogous to R for the modal logics K, T, K4, S4,
S5, D, D4, and G.
------------------------------
Date: Sat, 4 Oct 86 12:30:41 EDT
From: "Steven A. Swernofsky" <SASW%MX.LCS.MIT.EDU@MC.LCS.MIT.EDU>
Subject: Seminar - Efficient Unification of Quantified Terms (MIT)
From: Susan Hardy <SH at XX.LCS.MIT.EDU>
JOHN STAPLES
University of Queensland
Efficient unification of quantified terms
DATE: Tuesday, October 7, l986
TIME: 2:45 pm. - Refreshments
3:00 pm. - Talk
PLACE: 2nd Floor Lounge
Quantifiers such as for-all, integral signs, block headings would be
a valuable enrichment of the vocabulary of a logic programming language
or other computational logic. The basic technical prerequisite is a
suitable unification algorithm. A programme is sketched for the
development of data structures and algorithms which efficiently
support the use of quantified terms. Progress in carrying out this
programme is reviewed. Both structure sharing and non structure
sharing representations of quantified terms are described, together
with a unification algorithm for each case. The efficiency of the
approach results from the techniques used to represent terms,
which enable naive substitution to implement correct substitution
for quantified terms. The work is joint with Peter J. Robinson.
HOST: Arvind
------------------------------
Date: Sat, 4 Oct 86 13:04:33 EDT
From: "Steven A. Swernofsky" <SASW%MX.LCS.MIT.EDU@MC.LCS.MIT.EDU>
Subject: Seminar - Planning Simultaneous Actions (BBN)
From: Brad Goodman <BGOODMAN at BBNG.ARPA>
BBN Laboratories
Science Development Program
AI/Education Seminar
Speaker: Professor James Allen
University of Rochester
(james@rochester)
Title: Planning Simultaneous Actions in Temporally Rich Worlds
Date: 10:30a.m., Monday, October 6th
Location: 3rd floor large conference room,
BBN Labs, 10 Moulton Street, Cambridge
This talk describes work done with Richard Pelavin over the last few
years. We have developed a formal logic of action that allows us to
represent knowledge and reason about the interactions between events
that occur simultaneously or overlap in time. This includes interactions
between two (or more) actions that a single agent might perform
simultaneously, as well as interactions between an agent's actions and
events occuring in the external world. The logic is built upon an
interval-based temporal logic extended with modal operators similar to
temporal necessity and a counterfactual operator. Using this formalism,
we can represent a wide range of possible ways in which actions may
interact.
------------------------------
Date: Thu, 2 Oct 86 11:34 EDT
From: Tim Finin <Tim@cis.upenn.edu>
Subject: Seminar - Cognitive Architecture (UPenn)
WHAT IS THE SHAPE OF THE COGNITIVE ARCHITECTURE?
Allen Newell
Computer Science Department
Carnegie Mellon University
12:00 noon, October 17
Alumni Hall, Towne Building
University of Pennsylvania
The architecture plays a critical role in computational systems, defining
the separation between structure and content, and hence the capability of
being programmed. All architectures have much in common. However, important
characteristics depend on which mechanisms occur in the architecture (rather
than in software) and what shape they take. There has been much research
recently on architectures throughout computer and cognitive science. Within
computer science the main drivers have been new hardware technologies (VLSI)
and the felt need for parallelism. Within cognitive science the main drivers
have been the hope of comprehensive psychological models (ACT*), the urge to
ground the architecture in neurophysiological mechanisms (the
connectionists) and the proposal of modularity as a general architectural
principle (from linguistics). The talk will be on human cognitive
architecture, but considerations will be brought to bear from everywhere.
------------------------------
End of AIList Digest
********************
