Copy Link
Add to Bookmark
Report
AIList Digest Volume 4 Issue 194
AIList Digest Sunday, 21 Sep 1986 Volume 4 : Issue 194
Today's Topics:
Seminars (Past) - Rule Induction in Computer Chess (ACM LA Chapter) &
Mechanization of Geometry (SU) &
Automatic Algorithm Designer (CMU) &
Representations and Checkerboards (CMU) &
Deriving Problem Reduction Operators (Rutgers) &
Evolution of Automata (SRI) &
Active Reduction of Uncertainty (UPenn) &
Rational Conservatism and the Will to Believe (CMU) &
BiggerTalk: An Object-Oriented Extension to Prolog (UTexas)
----------------------------------------------------------------------
Date: 21 Aug 86 12:01:50 PDT (Thu)
From: ledoux@aerospace.ARPA
Subject: Seminar - Rule Induction in Computer Chess (ACM LA Chapter)
ACM LOS ANGELES CHAPTER DINNER MEETING
WEDNESDAY, 3 SEPTEMBER 1986
STRUCTURED EXPERT RULE INDUCTION
Expert Systems and Computer Chess
Speaker: Dr. Alen Shapiro
One of the major problems with expert systems is "the knowledge
engineering bottleneck." This occurs when development is delayed
because specifications are unavailable and either the expert system
developers need time to learn the problem, or else the domain experts
who already know the problem need time to learn how to use the often
opaque expert system development languages. A promising approach to
overcoming the bottleneck is to build tools that automatically extract
knowledge from the domain experts. This talk presents an overview of
inductive knowledge acquisition and the results of experiments in
inductive rule generation in the domain of chess endgames. The system
that will be described was able to generate humanly-understandable rules
and to play correct chess endgames. This research has significant
implications for the design of expert system languages and rule
induction programs. The talk is also an interesting look into the world
of computer chess.
Dr. Shapiro, a Fellow of the Turing Institute since its inception in
1983, received his Ph.D. in Machine Intelligence from the University of
Edinburgh in 1983. From 1979 to 1986 he was associated with Intelligent
Terminals, Ltd., and a member of the Rulemaster and Expert-Ease design
teams. He has served as Visiting Professor at the University of
Illinois on two occasions. His publications include articles on pattern
recognition, automatic induction of chess classification rules, and
(with David Michie), "A Self-Commenting Facility for Inductively
Synthesized Endgame Expertise."
In 1986 Dr. Shapiro joined the New Technology Department at Citicorp-TTI
in Santa Monica as a Computer Scientist concerned with the development
of inductive knowledge engineering tools for the banking industry.
PLACE
Amfac Hotel
8601 Lincoln Blvd.
corner of Lincoln & Manchester
Westchester, California
8:00 p.m.
------------------------------
Date: Mon, 18 Aug 86 11:25:38 PDT
From: coraki!pratt@Sun.COM (Vaughan Pratt)
Subject: Seminar - Mechanization of Geometry (SU)
SPEAKER Professor Wu Wen-tsun
TITLE Mechanization of Geometry
DATE Thursday, August 21
TIME 2:00 pm
PLACE Margaret Jacks Hall, room 352
A mechanical method of geometry based on Ritt's characteristic set
theory will be described which has a variety of applications including
mechanical geometry theorem proving in particular. The method has been
implemented on computers by several researchers and turns out to be
efficient for many applications.
BACKGROUND
Professor Wu received his doctorate in France in the 1950's, and was a
member of the Bourbaki group. In the first National Science and
Technology Awards in China in 1956, Professor Wu was one of three
people awarded a first prize for their contributions to science and
technology. He is currently the president of the Chinese Mathematical
Society.
In 1977, Wu extended classical algebraic geometry work of Ritt to an
algorithm for proving theorems of elementary geometry. The method has
recently become well-known in the Automated Theorem Proving community;
at the University of Texas it has been applied it to the machine proof
of more than 300 theorems of Euclidean and non-Euclidean geometry.
------------------------------
Date: 5 September 1986 1527-EDT
From: Betsy Herk@A.CS.CMU.EDU
Subject: Seminar - Automatic Algorithm Designer (CMU)
Speaker: David Steier
Date: Friday, Sept. 12
Place: 5409 Wean Hall
Time: 3:30 p.m.
Title: Integrating multiple sources of knowledge in an
automatic algorithm designer
One of the reasons that designing algorithms is so difficult is the
large amount of knowledge needed to guide the design process. In this
proposal, I identify nine sources of such knowledge within four
general areas: general problem-solving, algorithm design and
implementation techniques, knowledge of the application domain,
and methods for learning from experience. To understand how
knowledge from these sources can be represented and integrated, I
propose to build a system that automatically designs algorithms.
An implementation of the system, Designer-Soar, uses several
of the knowledge sources described in the proposal to design several
very simple algorithms. The goal of the thesis is to extend
Designer-Soar to design moderately complex algorithms in a domain
such as graph theory or computational geometry.
------------------------------
Date: 10 September 1986 1019-EDT
From: Elaine Atkinson@A.CS.CMU.EDU
Subject: Seminar - Representations and Checkerboards (CMU)
SPEAKER: Craig Kaplan, CMU, Psychology Department
TITLE: "Representations and Checkerboards"
DATE: Thursday, September 11
TIME: 4:00 p.m.
PLACE: Adamson Wing, BH
Given the right representation, tricky "insight" problems
often become trivial to solve. How do people arrive at the right
representations? What factors affect people's ability to shift
representations, and how can understanding these factors help us
understand why insight problems are so difficult?
Evidence from studies using the Mutilated Checkerboard
Problem points to Heuristic Search as a powerful way of addressing
these questions. Specifically, it suggest that the quality of
the match between people's readily available search heuristics
and problem characteristics is a major determinant of problem
difficulty for some problems.
------------------------------
Date: 11 Sep 86 20:01:20 EDT
From: RIDDLE@RED.RUTGERS.EDU
Subject: Seminar - Deriving Problem Reduction Operators (Rutgers)
I am giving a practice talk of a talk I will be giving in a few weeks.
It is at 1 pm in 423 on Monday the 15th.
Everyone is invited and all comments are welcome.
The abstract follows.
This research deals with automatically shifting from one problem
representation to another representation which is more efficient, with
respect to a given problem solving method, for this problem class. I
attempt to discover general purpose primitive representation shifts
and techniques for automating them. To achieve this goal, I am
defining and automating the primitive representation shifts explored
by Amarel in the Missionaries & Cannibals problem @cite(amarel1).
The techniques for shifting representations which I have already
defined are: compiling constraints, removing irrelevant information,
removing redundant information, deriving macro-operators, deriving
problem reduction operators, and deriving macro-objects. In this
paper, I will concentrate on the technique for deriving problem
reduction operators (i.e., critical reduction) and a method for
automating this technique (i.e., invariant reduction). A set of
sufficient conditions for the applicability of this technique over a
problem class is discussed; the proofs appear in a forthcoming
Rutgers technical report.
------------------------------
Date: Wed 10 Sep 86 15:00:22-PDT
From: Amy Lansky <LANSKY@SRI-WARBUCKS.ARPA>
Subject: Seminar - Evolution of Automata (SRI)
THE EVOLUTION OF COMPUTATIONAL CAPABILITIES
IN POPULATIONS OF COMPETING AUTOMATA
Aviv Bergman (BERGMAN@SRI-AI)
SRI International
and
Michel Kerszberg
IFF der KFA Julich, W.-Germany
10:30 AM, MONDAY, September 15
SRI International, Building E, Room EJ228
The diversity of the living world has been shaped, it is believed, by
Darwinian selection acting on random mutations. In the present work,
we study the emergence of nontrivial computational capabilities in
automata competing against each other in an environment where
possession of such capabilities is an advantage. The automata are
simple cellular computers with a certain number of parameters -
characterizing the "Statistical Distribution" of the connections -
initially set at random. Each generation of machines is subjected to a
test necessitating some computational task to be performed, e.g
recognize whether two patterns presented are or are not translated
versions of each other. "Adaptive Selection" is used during the task
in order to "Eliminate" redundant connections. According to its
grade, each machine either dies or "reproduces", i.e. it creates an
additional machine with parameters almost similar to its own. The
population, it turns out, quickly learns to perform certain tests.
When the successful automata are "autopsied", it appears that they do
not all complete the task in the same way; certain groups of cells are
more active then others, and certain connections have grown or decayed
preferentially, but these features may vary from individual to
individual. We try to draw some general conclusions regarding the
design of artificial intelligence systems, and the understanding of
biological computation. We also contrast this approach with the usual
Monte-Carlo procedure.
------------------------------
Date: Wed, 13 Aug 86 08:51 EDT
From: Tim Finin <Tim@cis.upenn.edu>
Subject: Seminar - Active Reduction of Uncertainty (UPenn)
Active Reduction of Uncertainty in Multi-sensor Systems
Ph.D. Thesis Proposal
Greg Hager
(greg@upenn-grasp)
General Robots and Active Sensory Perception Laboratory
University of Pennsylvania
Department of Computer and Information Sciences
Philadelphia, PA 19104
10:00 AM, August 15, 1986
Room 554 Moore
If robots are to perform tasks in unconstrained environments, they will have
to rely on sensor information to make decisions. In general, sensor
information has some uncertainty associated with it. The uncertainty can be
conceptually divided into two types: statistical uncertainty due to signal
noise, and incompleteness of information due to limitations of sensor scope.
Inevitably, the information needed for proper action will be uncertain. In
these cases, the robot will need to take action explicitly devoted to
reducing uncertainty.
The problem of reducing uncertainty can be studied within the theoretical
framework of team decision theory. Team decision theory considers a number
of decision makers observing the world via information structures, and
taking action dictated by decision rules. Decision rules are evaluated
relative to team and individual utility considerations. In this vocabulary,
sensors are considered as controllable information structures whose behavior
is determined by individual and group utilities. For the problem of
reducing uncertainty, these utilities are based on the information expected
as the result of taking action.
In general, a robot does not only consider direct sensor observations, but
also evaluates and combines that data over time relative to some model of
the observed environment. In this proposal, information aggregation is
modeled via belief systems as studied in philosophy. Reducing uncertainty
corresponds to driving the belief system into one of a set of information
states. Within this context, the issues that will be addressed are the
specification of utilities in terms of belief states, the organization of a
sensor system, and the evaluation of decision rules. These questions will
first be studied through theory and simulation, and finally applied to an
existing multi-sensor system.
Advisor: Dr. Max Mintz
Committee: Dr. Ruzena Bajcsy (Chairperson)
Dr. Zolton Domotor (Philosophy Dept.)
Dr. Richard Paul
Dr. Stanley Rosenschein (SRI International and CSLI)
------------------------------
Date: 10 Sep 1986 0848-EDT
From: Lydia Defilippo <DEFILIPPO@C.CS.CMU.EDU>
Subject: Seminar - Rational Conservatism and the Will to Believe (CMU)
CMU
PHILOSOPHY COLLOQUIUM
JON DOYLE
RATIONAL CONSERVATISM AND THE WILL TO BELIEVE
DATE: MONDAY SEPTEMBER 15
TIME: 4:OO P.M.
PLACE: PORTER HALL, RM 223d
* Much of the reasoning automated in artificial intelligence is either
mindless deductive inference or is intentionally non-deductive. The common
explanations of these techniques, when given, are not very satisfactory, for
the real explanations involve the notion of bounded rationality, while over
time the notion of rationality has been largely dropped from the vocabulary of
artificial intelligence. We present the notion of rational self-government, in
which the agent rationally guides its own limited reasoning to whatever degree
is possible, via the examples of rational conservatism and rationally adopted
assumptions. These ideas offer improvements on the practice of mindless
deductive inference and explantions of some of the usual non-deductive
inferences.
------------------------------
Date: Mon 15 Sep 86 10:35:02-CDT
From: ICS.BROWNE@R20.UTEXAS.EDU
Subject: Seminar - BiggerTalk: An Object-Oriented Extension to Prolog (UTexas)
Object-Oriented Programming Meeting
Friday, September 19
2:00-3:00 p.m.
Taylor 3.128
BiggerTalk:
An Object-Oriented Extension to Prolog
Speaker: Eric Gullichsen
MCC Software Technology Program
BiggerTalk is a system of Prolog routines which provide a capability for
object-oriented programming in Prolog. When compiled into a standard
Prolog environment, the BiggerTalk system permits programming in the
object-oriented style of message passing between objects, themselves
defined as components of a poset (the 'inheritance structure')
created through other BiggerTalk commands. Multiple inheritance of
methods and instance variables is provided dynamically. The full functional
capability of Prolog is retained, and Prolog predicates can be invoked
from within BiggerTalk methods.
A provision exists for storage of BiggerTalk objects in the MCC-STP
Object Server, a shared permanent object repository. The common external
form for objects in the Server permits (restricted) sharing of objects
between BiggerTalk and Zetalisp Flavors, the two languages currently
supported by the Server. Concurrent access to permanent objects is
mediated by the server.
This talk will discuss a number of theoretical and pragmatic issues of
concern to BiggerTalk and its interface to the Object Server. Some
acquaintance with the concepts of logic programming and object-oriented
programming will be assumed.
------------------------------
End of AIList Digest
********************
