Copy Link
Add to Bookmark
Report
AIList Digest Volume 4 Issue 079
AIList Digest Friday, 11 Apr 1986 Volume 4 : Issue 79
Today's Topics:
Bibliography - Technical Reports #3
----------------------------------------------------------------------
Date: WED, 10 JAN 84 17:02:23 CDT
From: E1AR0002%SMUVM1.BITNET@WISCVM.WISC.EDU
Subject: Technical Reports #3
%A N. S. Sridharan
%T Representational Facilities of AIMDS: A Sampling
%R CBM-TM-86
%D 1/82
%I Rutgers University, Department of Computer Science
%K AI01
%X The quest for fundamental and general mechanisms of intelligence,
especially problem solving and heuristic search techniques, that
guided early research in Artificial Intelligence has given way in the
last decade to the search for equally fundamental and general methods
for structuring and representing knowledge. This is the result of the
realization that a duality exists between knowledge and search:
Knowledge of the task domain can abbreviate search and search thru a
problem space can yield new knowledge. AIMDS is one of the recently
developed systems which permits experimentation with knowledge
representation in the course of building an AI program.
%A C. F. Schmidt
%T The Role of Object Knowledge in Human Planning
%R CBM-TM-87
%I Rutgers University Department of Computer Science
%K AI08 AI09
%D 1/82
%X AI research on planning provides an important reference point from
which the cognitive psychologist can build an understanding of human
planning. It is argued that the human planning context differs from
this reference point due to the incomplete knowledge that persons
typically possess about the situation within which the plan will be
executed. Various types of general functional knowledge about objects
are then defined. This knowledge serves as a source of default
assumptions for use in the planning process, and thus allows planning
to continue despite the absence of complete knowledge of the planning
situation. However, such assumption-based expectations must be
tested. From this point of view, planning must also include a process
for a kind of hypothesis testing and plan revision. The implications
of this claim are briefly discussed.
%A S. Amarel
%T Initial Thoughts on Characterization of Expert Systems
%R CBM-TM-88
%I Rutgers University Department of Computer Science
%D 1/82
%K AI01
%X Expertise in a given domain is commonly characterized by skillful,
high performance, problem solving activity in the domain. An expert
solves problems in a domain more rapidly, more accurately, and with
less conscious deliberation about his plan of attack than a novice
does. An excellent discussion of general characteristics of expert
behavior appears in a recent article in @u(Science) by Larkin et al.
[1].
.sp 1
Expert behavior is equivalent to high performance problem solving
behavior in a specific domain. It requires: knowledge of the domain,
knowledge of problem solving schemas and methods, knowledge/experience
about solution of specific problems in the domain with given methods,
knowledge about special properties and regularities in the problem
space, and highly effective ways of @u(using) all these bodies of
knowledge in approaching the solution of new problems in the domain.
Essentially, expert problem solving requires the
conceptualization/formulation of a given problem within a framework
wherein knowledge is embodied in definitions of states, moves,
constraints, evaluation functions, etc. in such a way that solutions
are attained with very little search. In other words, an expert
problem solver works within a highly 'appropriate' problem
representation: he describes situations and problem types within
'appropriate' conceptual frameworks, he specifies problem
decompositions that minimize subproblem interactions, he often uses
hierarchies of abstractions in his planning, he uses 'macromoves'
where a novice would painstakingly have to piece together elementary
moves, and he has rules for early recognition of unpromising as well
as of promising developments. An expert problem solver behaves as if
the great variety of knowledge sources needed for his
solution-construction activities are available to him in a
@u(compiled) procedural form.
.sp 1
Usually, expertise in a domain requires @u(problem solving experience)
in the domain. One can be scholar in a domain, and not an expert--if
he does not know how to effectively @u(apply) domain knowledge to a
variety of specific situations. Also, expertise implies a certain
amount of robustness in performance-- which means that it is not
sufficient to know how to handle a few 'textbook' cases; it is
important to be able to handle a broad range of variations.
%A S. Amarel
%T Review of Characteristics of Current Expert Systems
%R CBM-TM-89
%I Rutgers University Department of Computer Science
%D 3/81
%K AI01
%X This report does not cover all current work in the area of Expert
systems. It is intended to introduce a set of dimensions for
characterizing Expert systems and to describe some of the important
Expert systems that are now in existence (or are under active
development) in terms of these dimensions.
.sp 1
We have a dual purpose: (a) to illustrate via concrete examples the
dimensions that are being introduced, and (b) to show what is the
current state of the field from the perspective of this system of
dimensions.
.sp 1
We are using here ten main dimensions, and an optional eleventh called
@ux(Special Features), which provides added flexibility for the
presentation of relevant information about a system. Two of the main
dimensions, @ux(Performance) and @ux(Utility), are concerned with the
quality of the system's behavior and the impact of the system on the
domain of application and on AI. Another two dimensions are concerned
with the system's scope, its ability to handle situations that are
outside its area of major expertise, and its ability to improve: they
are called @u(Breadth, Intelligence, Robustness) and @u(Expertise
improvement ability). The remaining six dimensions are concerned with
the type of tasks performed by the system, its structure and its means
of interacting with users: they are called @u(Task type, Main Method,
Mode of Knowledge Representation, User Interface for main task,
Explanation facilities) and @u(Reasoning under Uncertainty).
.sp 1
The systems considered are DENDRAL, CASNET/GLAUCOMA, MACSYMA, MYCIN,
INTERNIST, PROSPECTOR and CRYSALIS.
.sp 1
This report covers material which was prepared for inclusion in the
Chapter 'What are Expert Systems' (co-authored with Ron Brachman, Carl
Engelman, Robert Engelmore, Edward Feigenbaum and David Wilkins) of a
book on Expert Systems which is currently under preparation; the book
is based on the Rand Workshop on Expert Systems which took place in
San Diego, California on August 25-28, 1980.
%A John Kastner
%A Sholom M. Weiss
%T A Precedence Scheme for Selection and Explanation of Therapies
%R CB-TM-90
%I Rutgers University Department of Computer Science
%D 3/81
%K AA01 AI01
%X A general scheme to aid in the selection of therapies is described. A
topological sorting procedure within a general production rule
representation is introduced. The procedure is used to choose among
competing therapies on the basis of precedence rules. This approach
has a degree of naturalness that lends itself to automatic explanation
of the choices made. A system has been implemented using this
approach to develop an expert system for planning therapies for
patients diagnosed as having ocular herpes simples. An abstracted
example of the system's output on an actual case is given.
%A P. Politakis
%A S. M. weiss
%T A System for Empherical Experimentation with Expert Knwoledge
%R CBM-TM-91
%I Rutgers University, Department of Computer Science
%D 1/82
%K AI01 AA01 rheumatology
%X An approach to the acquisition of expert knowledge is presented based
on the comparison of dual sources of knowledge: expert-modeled rules
and cases with known conclusions. A system called SEEK has been
implemented to give to the expert interactive advice about rule
refinement. SEEK uses a simple frame model for expressing
expert-modeled rules. The advice takes the form of suggestions of
possible experiments in generalizing or specializing rules in the
model. This approach has proven particularly valuable in assisting
the expert in domains where two diagnoses are difficult to
distinguish. Examples are given from an expert consultation system
being developed for rheumatology.
%A G. Drastal
%A C. Kulikowski
%T Knowledge Based Acquisition of Rules for Medical Diagnosis
%R CBM-TM-92
%I Rutgers University, Department of Computer Science
%D 10/81
%K AA01 AI01
%X Medical consultation systems in the EXPERT framework contain rules
written under the guidance of expert physicians. We present a
methodology and preliminary implementation of a system which learns
compiled rule chains from positive case examples of a diagnostic class
and negative examples of alternative diagnostic classes. Rule
acquisition is guided by the constraints of physiological process
models represented in the system. Evaluation of the system is
proceeding in the area of glaucoma diagnosis, and an example of an
experiment in this domain is included.
%A N. S. Sridharan
%T AIMDS: Applications and Performance Enhancements
%R CBM-TM-93
%I Rutgers University, Department of Computer Science
%D 1/82
%K AI01 AA24
%X AIMDS is a programming environment (language, editors, display drivers, file
system) in which several programs are being constructed for modeling
commonsense reasoning and legal argumentation. The main obstacle to realistic
applications in these and other areas is system performance when the knowledge
bases used are scaled up one or two orders of magnitude. The other obstacle
is user performance resulting from the complexity of constructing and debugging
large scale knowledge bases. This proposal argues that performance enhancement
of AIMDS as a system is needed and that the usual solutions of software tuning
have been exhausted and that new hardware ideas fitted to the characteristics
of the task need to be experimented with. We adopt as important constraints:
the requirement that existing programs should receive graded enhancement of
performance, maintaining continuity of application programs; that user programs
should not reflect changing machine configurations or architectures. Redesign
and recoding of AIMDS should provide the necessary opacity to the user.
With these constraints in mind, we suggest interim solutions and long-term
solutions. The interim solutions include: converting large-address space
personal Lisp machines with bit-mapped graphics; fast coding of low-level
functionalities via microprogramming. The long-term solutions include the
building and testing of multiprocessors. The long-term solutions open up
a number of rather difficult software and hardware research problems whose
solutions depend upon having good facilities to experiment in the search for
answers.
%A B. Lantz
%T The AIMDS Interactive Command Parser
%R CBM-TM-94
%I Rutgers University, Department of Computer Science
%D 9/82
%K AA24 AI01 T03
%X Characters entered by the user are parsed immediately in order to provide
interactive services to the user while he is entering commands. Services
provided to the user include immediate verification of syntax, supplying the
user with information about the correct syntax and semantics of a command,
completion of long descriptive atom names, pretty printing the entered
command, and defining special functions for selected characters. The parser
accepts user defined grammars, thus providing a useful command parser for a
great variety of applications.
%A B. Lantz
%T The AIMDS On-Line Documentation Facility
%R CBM-TM-95
%I Rutgers University, Department of Computer Science
%D 9/82
%K AA24 AI01 T03
%X The documentation system for the AIMDS language is designed to be suitable
for both beginning and expert users, and to be capable of serving the needs
of a changing system such as AIMDS. The documentation must be quickly and
easily updatable, and the updated information should be available to, and
easily used by, a wide variety of users.
%X This paper is a short description of the documentation system for the
AIMDS language. It includes a discussion of the considerations taken during
the design of the documentation system, a description of the implemented
system, and instructions for using the system for other documentation tasks.
%A J. Roach
%A N. S. Sridharan
%T Implementing AIMDS on a Multiprocessor Machine. Some Considerations
%R CBM-TM-96
%D 4/83
%I Rutgers University, Department of Computer Science
%K AA24 T03 H03 AI01
%X As a possible long term solution for performance enhancement of AIMDS,
a Lisp based multiprocessor system was proposed. Converting an
existing AI knowledge based system from the current uniprocessor
environment into a multiprocessor based regime is a largely unexplored
research question. This report discusses some of the issues raised by
such a proposal and attempts to evaluate some of the current models of
parallel processing in regards to implementing an AIMDS based system.
An extensive bibliography with commentary is included.
%A George A. Drastal
%A Casimir A. Kulikowski
%T Knowledge-Based Acquisition of Rule for Medical Diagnosis
%R CBM-TM-97
%I Rutgers University, Department of Computer Science
%D 11/82
%K AI01 AA01 T03
%X Medical consultation systems in the EXPERT framework contain rules written
under the guidance of expert physicians. We present a methodology and
preliminary implementation of a system that learns compiled rule chains
from positive case examples of a diagnostic class and negative examples
of alternative diagnostic classes. Rule acquisition is guided by the
constraints of physiological process models represented in the system.
Evaluation of the system is proceeding in the area of glaucoma diagnosis,
and an example of an experiment in this domain is included.
%A S. Weiss
%A K. Kern
%A C. Kulikowski
%A M. Uschold
%T A Guide to the Use of the EXPERT Consultation System
%R CBM-TR-94
%I Rutgers University, Department of Computer Science
%D 1/82
%K T03 AI01
%X EXPERT is a system for designing and applying consultation models.
An EXPERT model consists of hypotheses (conclusions), findings
(observations), and rules for logically relating findings to
hypotheses. Three phases of model development are outlined for users
of the system. These include: the design of a decision-making model,
compilation of the model, and consultation using the model. The
facilities of the system are described, and examples of models and
consultation sessions are presented.
%A R. Banerji
%A T. Mitchell
%T Description Languages and Learning Algorithms: A Paradigm for Comparison
%R CBM-TR-107
%D 1/82
%I Rutgers University, Department of Computer Science
%K AI04 Inductive inference, learning, generalization, description languages.
%X We propose and apply a framework for comparing various methods for
learning descriptions of classes of objects given a set of training
exemplars. Such systems may be usefully characterized in terms of
their descriptive languages, and the learning algorithms they employ.
The basis for our characterization and comparison is a
general-to-specific partial ordering over the description language,
which allows characterizing learning algorithms independent of the
description language with which they are associated. Two existing
learning systems are characterized within this framework, and
correspondences between them made clear.
------------------------------
End of AIList Digest
********************
