Copy Link
Add to Bookmark
Report
AIList Digest Volume 5 Issue 225
AIList Digest Friday, 2 Oct 1987 Volume 5 : Issue 225
Today's Topics:
Queries - Expert/AI Work in Communication Networks &
Annual Review of Computer Science & Neural Hardware &
Using the ATMS to Scan Homeric Verse,
Bindings - CPSR,
Representation - Time,
Neural Networks - Boltzmann Machines,
Obituary - Slava Prazdny
----------------------------------------------------------------------
Date: 29 Sep 87 14:33:39 GMT
From: cbosgd!cblpf!dtm@ucbvax.Berkeley.EDU (Dattaram Mirvke)
Subject: Need references on Expert/AI work in Communication Networks.
Need references/pointers to work done in the Expert systems/AI in the
Network domain. I am specially interested in simulations/modeling/
behavior analysis and fault diagnosis.
Please e-mail the responses to me unless the information is of
general interest. If I get sufficient responses I will summarise to the
net.Thanks in advance.
- Datta Miruke
cbosgd!cblpf!dtm
cbosgd!ncpe!drm
------------------------------
Date: 30 Sep 87 19:51:42 GMT
From: wucs1!grs@uunet.UU.NET (Guillermo Ricardo Simari)
Subject: Annual Review of Computer Science
The book "Logical Foundations of Artificial Intelligence"
by M. R. Genesereth, N. J. Nilsson contains the following reference,
Levesque, H., "Knowledge Representation and Reasoning",
Annual Review of Computer Science, 1986
Can anyone give me information about the above journal? I cannot find it
anywhere.
Guillermo Simari Washington University, Department of Computer Science
St. Louis, MO, 63130, U.S.A.
UUCP: grs@wucs1.UUCP or ...!{ihnp4,uunet}!wucs1!grs
------------------------------
Date: 1 Oct 87 02:02:46 GMT
From: munnari!mulga.oz!jayen@uunet.UU.NET (Jayen Vaghani)
Subject: Want information on Neural Hardware in use
I am preparing a talk for one of my honours subjects and part of the talk
centres on comparing neural hardware to other architectures.
I need information on what neural hardware is in use (possibly commercially
available) and what it is being used for. I would also like to know why this
direction was chosen as against using a more general purpose parallel
architecture and modelling the neural network on that. Perhaps some feelings
about whether the approach has any future would also be nice and what problems
were encountered in using the system.
Responses can be to the net or mailed to me. If people are interested I will
summarise any personal responses to the net. Possibly someone else has already
asked this question so I would be happy to hear what responses they got.
Thanks in advance,
Jayen.
-------
UUCP: {seismo,ukc,ubc-vision,mcvax}!mulga.oz!jayen
ARPA: jayen%mulga.oz@seismo.css.gov
CSNET: jayen%mulga.oz@australia
------------------------------
Date: 29 Sep 87 15:18:35 GMT
From: eagle!icdoc!qmc-cs!flash@ucbvax.Berkeley.EDU (Flash Sheridan)
Subject: Using the ATMS to Scan Homeric Verse
As a toy demo, I'm trying to use Johan deKleer's Assumption Based
Truth Maintenance System to scan Homer. I'd appreciate comments.
I'd also appreciate it if somebody could email me a hundred or so
lines, so I don't have to type in any more.
------------------------------
Date: 1 Oct 87 22:58:08 GMT
From: acornrc!rbbb@ames.arpa (David Chase)
Subject: Re: Using the ATMS to Scan Homeric Verse
In article <297@sequent.cs.qmc.ac.uk>, flash@ee.qmc.ac.uk (Flash Sheridan)
writes:
> ... to scan Homer.
> I'd also appreciate it if somebody could email me a hundred or so
> lines, so I don't have to type in any more.
[this doesn't really belong on this list, but this is a mighty stale
pointer. I am hoping that this will jog the memory of someone else on the
list with more recent information.]
Sometime around about 1975 (in high school) I went to a seminar at the Nat.
Junior Classical League convention where someone from Dartmouth talked
about feeding the Aeneid to a computer program, doing the meter, counting
"et"s, etc. I was under the impression that they had other classics on
line or on tape.
By the way, how should I type in ancient Greek on my U.S.A. keyboard? I
can use ` and ' and ~ for accents, but what about the breath marks?
David Chase, Olivetti Research Center
------------------------------
Date: 29 Sep 87 22:10:56 GMT
From: sdcrdcf!ism780c!jimh@hplabs.hp.com (Jim Hori)
Subject: Re: Is Computer Science Science? (Funding)
In article <2868@ames.arpa> eugene@pioneer.UUCP (Eugene Miya N.) writes:
>be read in the latest CPSR [Computer Professionals for Social
>Responsibility] Newsletter. It appears in the halls of places like
can you, or anyone, post the address of this
newletter?
jimh ...yeah you right
........................
------------------------------
Date: 30 Sep 87 16:28:47 GMT
From: pioneer!eugene@ames.arpa (Eugene Miya N.)
Subject: Re: Is Computer Science Science? (Funding)
Computer Professionals for Social Responsibility (National Office)
646 Emerson St.
Palo Alto, CA 94301
------------------------------
Date: 1 Oct 87 12:31:56 GMT
From: ihnp4!homxb!homxc!del@ucbvax.Berkeley.EDU (D.LEASURE)
Subject: Re: Is Computer Science Science? (Funding)
In article <7397@ism780c.UUCP>, jimh@ism780c.UUCP (Jim Hori) writes:
> can you, or anyone, post the address of this
> newletter? [CPSR]
CPSR, Inc. PO Box 717, Palo Alto CA 94301 415/322-3778
$30/yr $10/yr for student
--
David E. Leasure - AT&T Bell Laboratories - (201) 615-5307
------------------------------
Date: Tue, 29 Sep 87 09:46:05 -0400
From: koomen@cs.rochester.edu
Subject: Representation of Time
>From: mcvax!unido!uklirb!noekel@uunet.uu.net
>Subject: J.F.Allen's work on time - (nf)
Reference intervals, automatic interval hierarchy structuring, duration
logic, etc, have indeed been implemented, in support of my PhD research
project. For a description, watch for a UofR Tech Report and my
dissertation, both expected to appear within the next year.
-- Hans
EMail: Koomen@CS.Rochester.Edu Paper: Johannes A. G. M. Koomen
Dept. of Computer Science
Phone: (716) 275-9499 [work] University of Rochester
(716) 442-4836 [home] Rochester, NY 14627
------------------------------
Date: Tue, 29 Sep 87 13:13:25 PDT
From: ladkin@kestrel.ARPA (Peter Ladkin)
Subject: Re: J.F. Allen's work on time
There has been much subsequent work done on this topic. I'll try to
summarise what I know about. James Allen's system can be formulated as
a relation algebra in the sense of Tarski, and also as a complete,
countably categorical first-order theory. The theory is thus
decidable, and admits of quantifier elimination. This means that
arbitrary first-order constraints expressed in the system are
computably equivalent to a set of constraints without quantifiers.
Such collections of Boolean constraints may be checked for consistency
either by an extension of Allen's method, or in other ways. Marc
Vilain and Henry Kautz showed that the general constraint satisfaction
problem for this system is NP-complete. Allen and Pat Hayes have
formulated an alternative theory of time intervals as a collection of
first-order axioms. They want to allow the collection of pairs of
integers as a model, and thus the axioms are weaker than the original
system. These axioms have as models exactly sets of pairs from an
unbounded linear order. Johan Van Benthem has investigated interval
theories in his book `The Logic of Time'. All the theories are
comparable, it turns out, since the collections of primitives are
interdefinable.
For a reading list, the AAAI-86, AAAI-87, IJCAI-85 and IJCAI-87
conference proceedings contain papers on interval systems for time
representation, Allen and Hayes have a technical report (University of
Rochester) due out any day, I have also technical reports not in the
above sources (Kestrel Institute), and Edward Tsang (University
of Essex) has some also. Tom Dean (Brown University) is using interval
representations in his planner, and has investigated the most commonly
occurring constraint satisfaction problems in detail. Richard Pelavin
has incorporated Allen's interval system into the design for a
planner, and Henry Kautz has also investigated the use of interval
specifications in general planning problems. (Both are former
students of James Allen). There is van Benthem's book, and a review
of it by Steven Kuhn in the September 1987 Journal of Symbolic Logic
(of the open problems mentioned by Kuhn, the first and last were
solved already, by Roger Maddux and I, and I'm sure some others). Joe
Halpern and Yoav Shoham formulated a modal logic of time with interval
modalities, in the First Logic in Computer Science conference (1986,
Proceedings published by IEEE). Klaus Schultz at Tubingen has a
technical report comparing Allen's approach with Kamp's event theory,
and Austin Tate and Colin Bell have investigated the use of interval
constraints in the O-Plan planner at Edinburgh (Bell is at the
University of Iowa). Other references may be found by taking the
transitive closure of the `references' relation on these sources.
There is very closely related work being done by Robert Kowalski's
group on event structures (Imperial College).
Apologies to those whose work I've missed or are unaware of (please
let me know). Things are progressing fast, so we all ought to be on
each other's mailing lists.
peter ladkin
ladkin@kestrel.arpa
------------------------------
Date: Tue, 29 Sep 1987 11:09 EDT
From: "Scott E. Fahlman" <Fahlman@C.CS.CMU.EDU>
Subject: Boltzmann Machines
To answer your question about Boltzmann machines:
In the original Boltzmann Machine formulation, a pattern (think of this
as both inputs and outputs) is clamped into the visible units during the
teaching phase; the network is allowed to free-run, with nothing
clamped, during the normalization phase. The update of each weight is a
function of the difference between co-occurrence statistics measured
across that connection during the two phases.
The result (if all goes well) is a trained network that has no concept
of input and output: clamp a partial pattern into the visible units, and
the network will try to complete it in a way that is consistent with the
training examples. Clamp nothing, and the network should settle into
states whose distribution approximates the distribution of examples in
the training set.
Later, someone (Geoff Hinton, I think), realized that if the network was
really being trained to produce a certain input-to-output mapping, it
was wasteful of links and training effort to train the network to
reproduce the distribution of input vectors; an input will always be
supplied when the network is performing. If the visible units are
divided into an input set and an output set, if the teaching phase is
done as before, and if the inputs (only) are clamped during the
normalization phase, the network will "concentrate" on learning to
produce the desired outputs, given the inputs, and will not develop the
capability of reproducing the input distribution.
Some papers refer to the "completion" model, others to the "Input/Ouput"
model. The distinction is not always emphasized. The learning
procedure is essentially the same in either case.
Note that, unlike Boltzmann, the back-propagation model is inherently an
I/O model, though it is possible to do completion tasks with some added
work. For example, one might train a backprop network to map each of a
set of patterns into itself, and then feed it partial patterns at the
inputs.
-- Scott Fahlman, CMU
------------------------------
Date: 30 Sep 87 21:21:50 GMT
From: giraffe..arpa!krulwich@uunet.uu.net (Bruce Krulwich)
Reply-to: yale.ARPA!krulwich@uunet.uu.net (Bruce Krulwich)
Subject: Re: Boltzmann Machine
> Since the expression for dG/dWij is the same in both cases, the
> definitions of Pij- must be equivalent. The only explanation I could
> think of was that "clamping" the inputs ONLY was the same thing as letting
> the environment have a free run of them, so the case being described is
> the free-running one.
The point is that for any given inputs learning is done by comparing
the desired outputs with the outputs computed by the machine. This
called monitored learning, and is similar in this sense to back
propogation learning. This is used for networks that perform a
computation based on some input being clamped in the input units.
When the output units are clamped, the P values are something like
what they "should" be, so comparing these to the P values for
unclamped output units lets you approximate the error between the
units in qestion and learn from it.
Bruce Krulwich
ARPA: krulwich@yale.arpa If you're right 95% of the time,
or krulwich@cs.yale.edu why worry about the other 3% ??
Bitnet: krulwich@yalecs.bitnet
UUCP: {harvard, seismo, ihnp4}!yale!krulwich
------------------------------
Date: 29 Sep 87 11:04:18 PDT (Tue)
From: baird@cel.fmc.com (Michael Baird)
Subject: Slava Prazdny (Bindings)
As many of you know by now, Slava Prazdny died Saturday, September 19th, in a
hang-gliding accident, high in the California mountains. He is survived by
his wife, Dagmar Dolan, their as yet unborn child, and his 15 year old
daughter Bronja Prazdny. Slava was 38. He was with FMC's Santa Clara AI
Center during the past two years, and before that at Schlumberger's Palo Alto
Research Center / Fairchild Laboratory for Artificial Intelligence Research.
Memorial Services will be held at 3:15 p.m., Wednesday October 7th, 1987,
outdoors in Foothills Park, operated by the City of Palo Alto, just a few
miles up Page Mill Road "west" of I-280. Flowers may be brought to the
memorial. Dagmar invites members of the AI community to attend.
It is suggested that you enter the park gate (it says for Palo Alto residents
only -- but tell the guard that you are attending the memorial) around 3 p.m.
From the parking area find the "Lee" grove (two large redwoods) beyond the
picnic tables. Services will be informal, as Slava would have wanted them to
be.
Slava had published over 60 refereed papers, and was recognized
internationally as an expert in many aspects of human and machine perception.
His latest works in stereo vision and adaptive "neural" networks were deemed
scientific breakthroughs.
A beautiful Redwood tree in Big Basin State Park will be dedicated in Slava's
name. This will be a pleasant place we can go to remember Slava. The family
has asked that donations be sent in his name to The Sempirvirens Fund, 2483
Old Middlefield Way, Mountain View, CA 94043.
Mike Baird
baird@cel.fmc.com
(408) 289-4932
------------------------------
End of AIList Digest
********************