Copy Link
Add to Bookmark
Report
AIList Digest Volume 4 Issue 068
AIList Digest Tuesday, 8 Apr 1986 Volume 4 : Issue 68
Today's Topics:
Seminars - Tek Tools and Technology (Ames) &
Machine Learning, Clustering and Polymorphy (Rutgers) &
Feedback During Skill Acquisition (CMU) &
Growing Min-Max Game Trees (MIT) &
State, Models, and Qualitative Reasoning (MIT) &
Functional Computations in Logic Programs (UPenn)
----------------------------------------------------------------------
Date: Tue, 1 Apr 86 08:32:09 pst
From: eugene@AMES-NAS.ARPA (Eugene Miya)
Subject: Seminar - Tek Tools and Technology (Ames)
From: MER::ANDREWS
National Aeronautics and Space Administration
Ames Research Center
AMES AI FORUM
SEMINAR ANNOUNCEMENT
Tektronix AI Tools and Technology
Tektronix Representatives:
Steve Levine - AI Specialist
Brad Martinson - Systems Analyst
Tamarah Day - Sales Engineer
Tuesday, April 8, 1986 10:30 - 11:30 am
B 239 rm B39 (Life Sciences Basement Auditorium)
NASA Ames Research Center
Agenda:
10:30 - 11:00 Slide presentation - AI history overview
Question & answer period
11:00 - 11:30 or Product demonstrations on the Tektronix 4404 and 4406
12:00 Artificial Intelligence Workstations. Demonstrations
will include a Preliminary Expert Ground Analysis
Scheduler developed by Harris Corporation for Kennedy
Space Center to assist in the scheduling of ground
processing activities. Also presented will be an
electronic circuit board diagnostic expert system and
applications of software prototyping and user
interfacing.
point of contact: Alison Andrews (415)694-6741
mer.andrews@ames-vmsb.ARPA
N.B. For those of you who cannot make it to this Ames AI Forum, Tektronix
is having a similar presentation and demo on April 3, with the
following agenda:
8:30-9:00 Coffee and doughnuts
9:00-10:30 Presentations (AI Overview, AI at TEK Labs, Managing
the Knowledge Engineering Process)
10:30-11:15 Demonstrations
11:15-11:30 Summary
11:30-12:00 Questions and Answers
12:30-4:00 Afternoon Schedule
R.S.V.P. Mary Clement (408)496-0800
Tektronix is located at 3003 Bunker Hill Lane (just off Great
America Parkway, near cross street Betsy Ross), Santa Clara.
Attendees of the April 3 demo will not be shown the Kennedy Space
Center expert system, so do try to make it to the Ames AI Forum,
despite the lack of doughnuts!
------------------------------
Date: 3 Apr 86 16:56:24 EST
From: PRASAD@RED.RUTGERS.EDU
Subject: Seminar - Machine Learning, Clustering and Polymorphy (Rutgers)
MACHINE LEARNING COLLOQUIUM
Machine Learning, Clustering and Polymorphy
Stephen Jose Hanson
and
Malcolm Bauer
Bell Communications Research
and
Princeton University Cognitive Science Laboratory
April 8, Tuesday
#423, Hill Center
I will describe a conceptual clustering program (WITT) that
attempts to model human categorization. Experiments will
also be described in which the output of WITT and other
Conceptual clustering programs will be compared to the
performance of human subjects using the same stimuli.
Properties of categories to which human subjects are
sensitive includes best or prototypical members, relative
contrasts between putative categories, and polymorphy
(neither necessary or sufficient features). Polymorphy (m
out of N, m < N) represents a weakening of conjunction
predicates which still seem to be of an order that is
learnable to humans. Wittengentein refers to polymorphy as
a basis for a category theory in which category "criteria"
determine the nature of the membership rule.
This approach represents an alternative to usual
Artificial Intelligence approaches to generalization,
conceptual clustering and semantic analysis which tend to
focus on common feature rules, impoverished category
structure, and simple search and match schemes. WITT uses
feature inter-correlations, category structure (prototypes,
basic levels, etc..) and a conservative search strategy in
order to construct a set of categories given objects defined
on a multi-valued feature list. Information retrieval was
used for a test domain for WITT in order to discover
reasonable categories from the psychological abstracts,
which were subsequently compared to psychologists from
Princeton psychology department sorting the same abstracts.
Another test domain involved constructing meta-level
categories for nations of the world, where semantic features
were extracted from a machine readable version of the 1985
World Almanac. WITT discovered concepts like "third world
countries" and "european countries" and "technologically
advanced countries".
** If you wish to host the speakers or meet with them, please send
a message to PRASAD@RUTGERS.ARPA
------------------------------
Date: 4 April 1986 1433-EST
From: Cathy Hill@A.CS.CMU.EDU
Subject: Seminar - Feedback During Skill Acquisition (CMU)
Impact of Feedback Content during Initial
Skill Acquisition
Jean McKendree
Wednesday, April 9 12:00-1:30 pm
****** BH 340A ******
Most theories of learning and skill acquisition acknowledge the
importance of feedback, particularly after errors. However, none of
them are explicit about the content of this information. I will
present hypotheses about the efficacy of different sorts of feedback
content and relate them briefly to current information processing
theories. I will then present the results from experiments
which vary information content after errors and which begin to look at
differences in experience level. The proposed experiment will use
verbal protocols as well as quantitative data to better
understand the usefulness of different sorts of information for
error correction. A simulation model will attempt to compare the
impact of these different types of information assuming an identical
starting point.
------------------------------
Date: Mon, 7 Apr 1986 18:12 EST
From: JHC%OZ.AI.MIT.EDU@XX.LCS.MIT.EDU
Subject: Seminar - Growing Min-Max Game Trees (MIT)
[Forwarded from the MIT bboard by SASW@MIT-MC.]
Thursday , April 10 4:00pm Room: NE43 8th floor Playroom
The Artificial Intelligence Lab
Revolving Seminar Series
A New Procedure for Growing Min-Max Game Trees
David McAllester
AI Lab, MIT
In games such as chess decisions must be based on incomplete search
trees. A new tree-growth procedure is presented which is based on
"conspiracy numbers" as a formal measure of the accuracy of the root
minimax value of an incomplete tree. Trees can be grown with the goal
of maximizing the accuracy of the root value. Trees grown in this way
are often deeper and narrower than alpha-beta optimal trees with the
same number of nodes. On the other hand, if all nodes have the same
static value then the new procedure reduces to d-ply search with
alpha-beta pruning. Unlike B* search, non-uniform growth is achieved
without any modification of the static board evaluator.
------------------------------
Date: Mon, 31 Mar 1986 17:21 EST
From: JHC%OZ.AI.MIT.EDU@XX.LCS.MIT.EDU
Subject: Seminar - State, Models, and Qualitative Reasoning (MIT)
[Forwarded from the MIT bboard by SASW@MIT-MC.]
The Artificial Intelligence Lab
Revolving Seminar Series
State, Models, and Qualitative Reasoning
Jerry Roylance
AI Lab, MIT
Qualitative reasoning, modeling, and representations of state are
important issues in AI. Machines need interesting models of their
task and methods that enable them to reason with those models.
Without models machines can offer little help in relieving the
programmer's or system builder's workload.
A conventional program is a literal description of what to do. By
investing the program with a model of what it is doing and some methods,
we can use code that is both simpler and more believable. Numerical
subroutines, for example, have several unifying ideas about search,
approximation, and transformation. Using these ideas directly (rather
than the results of the ideas) eliminates a lot of ugly code.
While qualitative reasoners gain their power in the simplicity of their
algebra, they pay a price in resolving the ambiguity that that
simplicity produces. We look at the simplifications that qualitative
reasoners do in light of the mathematical properties of the original
equations, the choice of distinguished values, and traditional
simulation methods.
Modeling a world is a difficult problem. State is a part of modeling
that is not described very well; the best descriptions that we have are
Moore machine descriptions that the current state and the inputs give us
the next state. Better, goal-oriented, descriptions that do more than
just simulation are needed.
Thursday, April 3 4:00pm Room: NE43 8th floor Playroom
Refreshments at 3:30pm
------------------------------
Date: Mon, 31 Mar 86 11:09 EST
From: Tim Finin <Tim%upenn.csnet@CSNET-RELAY.ARPA>
Subject: Seminar - Functional Computations in Logic Programs (UPenn)
Forwarded From: Glenda Kent <Glenda@UPenn> on Mon 31 Mar 1986 at 10:42
FUNCTIONAL COMPUTATIONS IN LOGIC PROGRAMS
Saumya K. Debray
SUNY at Stony Brook
Tuesday, April 1, 1986
Room 216 - Moore School
3:00 - 4:30 p.m.
While the ability to simulate nondeterminism and return multiple outputs for a
single input is a powerful and attractive feature of logic programming
languages, it is expensive in both time and space. This overhead is especialy
undesirable because programs are very often functional, i.e. do not return more
than one output for any given input, and so do not use this feature of these
languages. This talk describes how programs may be analyzed statically to
determine which literals and predicates are functional, and how the program may
then be optimized using this information. Our notion of "functionality"
subsumes the notion of "determinacy" that has been considered by various
researchers. The algorithm we describe is less reliant on features such as
cut, and thus extends more easily to parallel evaluation strategies, than
others that have been proposed.
------------------------------
End of AIList Digest
********************
