Copy Link
Add to Bookmark
Report
AIList Digest Volume 2 Issue 081
AIList Digest Thursday, 28 Jun 1984 Volume 2 : Issue 81
Today's Topics:
AAAI - Instructions,
Standards - Maintaining High Quality in AI Products,
Business - Softwar,
Mathematics - Best fitting curve,
Knowledge Representation - Frames Question,
AI and Statistics - Bibliography,
AI Programming - Spelling Correctors,
Turing Test - Machines vs People
----------------------------------------------------------------------
Date: 26 June 1984 1651-EDT
From: Dave Touretzky at CMU-CS-A
Subject: AAAI paper presentations
Claudia Mazzetti, executive director of AAAI, warns that many of the
paper presentations at this year's conference will be in very large
concert or lecture halls. Ordinary transparencies done in 20-point
font will *not* be readable. AAAI very strongly recommends using
35mm slides for paper presentations. If you must use transparencies,
a 36-point font is recommended.
------------------------------
Date: 27 Jun 84 16:02:56-PDT (Wed)
From: hplabs!hao!seismo!brl-tgr!abc @ Ucb-Vax.arpa
Subject: Re: Maintaining High Quality in AI Products
Article-I.D.: brl-tgr.3065
I suggest that the ACM provides an appropriate umbrella under which such
an effort can at least be planned. It is sufficiently broad-based as to
be representative and not exclusive and its democratic procedures
provide protection from the types of abuses that could be possible. (I
do not mean to slight the AAAI; it's just that ACM seems to have more of
the "mechanisms" that such an efort will need.)
Also, I have felt for many years that ACM should, at least in the US,
provide the kind of accreditation of Computer Science curricula that the
engineering societies provide for theirs.
------------------------------
Date: 26 June 1984 07:04-EDT
From: Herb Lin <LIN @ MIT-MC>
Subject: Softwar
From: syming%B.CC at Berkeley
They sent us a tape with a fixed time (two months or so?)
payment notice and stated that the program would vanish after that time. Of
course, we paid in time and they sent us a 20(?)-digit long key word and
instruction to make our trial copy a one-year-life-time program, since the
service contract was year by year.
I'm a bit confused. How could this particular program make itself
vanish without some external reference to a date? It seems that a
simple routine to change the date to the date of original purhcase
whenever the routine was invoked would do the trick. Do you know if
anyone ever actually has their program vanish? Maybe the whole thing
was a bluff?
------------------------------
Date: 24 Jun 84 12:26:54-PDT (Sun)
From: hplabs!sdcrdcf!sdcsvax!sdccsu3!ee171bbr @ Ucb-Vax.arpa
Subject: Best fitting curve
Article-I.D.: sdccsu3.1970
Given three points, what is the equation
of the best fit curve. (how does one
go about solving this?)
also, what is knuth's cocubic equation
and would that solve my problem?
John F.
------------------------------
Date: 25 Jun 84 6:08:52-PDT (Mon)
From: ihnp4!houxm!mhuxl!ulysses!allegra!mouton!mwg @ Ucb-Vax.arpa
Subject: Re: Best fitting curve - 3 points
Article-I.D.: mouton.90
Since three points determine a parabola, just plug them into
y = Ax^2 + Bx + C and solve the system. If you are in more than two
dimensions, you can probably do a transformation somehow into the
plane determined by the three points and solve; then translate back.
-Mark
------------------------------
Date: 25 Jun 84 11:40:06-PDT (Mon)
From: decvax!yale-comix!leichter @ Ucb-Vax.arpa
Subject: Re: Best fitting curve
Article-I.D.: yale-com.4061
The notion of a "best fitting curve" through some points has no inherent
meaning. You have to specify what kinds of curves you are willing to allow,
what kind of constraints you want to put on them, and what kind of measurement
of "fit" you are interested in.
Given n points in the plane, in general there is a unique (n-1)st degree poly-
nomial that passes through those points. (Hence, there are infinitely many
nth degree polynomials, one for every other point in the plane that you can
consider to be an (n+1)st point.) Such a polynomial is "best" in the sense
that it has 0 error at every point specified. It is almost certainly not
what you would want for fitting data; it will generally oscillate between
your data points and will have uncontrollable behavior outside the range in
which your data points occur - i.e., the curve will not look at all "smooth"
to the eye. Even if you want a curve that "looks good", you can use a
cubic (or higher-order) spline curve. This is a curve defined by fitting
together polynomials; you take the first 4 points in order, pass a cubic
through them, take the last two and the next one, pick a cubic through those
that has the same derivative at the 4th point as the first cubic, etc...
(There are many other ways to choose spline curves. This particular method
passes through all the points; in some cases, "smoothness" of some sort may
be more important than actually touching the points, so some kinds of splines
don't even pass through the given data points. All spline curves are piece-
wise defined polynomials; there is no simple algebraic formula that defines
them; rather, there is a series of such formulas, one for each range of input
values.)
If the goal of "best fit" is to produce good interpolated values for a function,
rather than a curve that "looks" like it is determined by the points, all sorts
of other techniques exist. For example, a Chebyshev approximation will have
the least maximum error (assuming a model in which you are approximating some
known, complex function by choosing some representative points on it and
an approximating polynomial.) However, least maximum error is not the same
as least average absolute error, or least RMS error, or...
So, in summary: If you give me three points, I can write down pretty much
ANY function and find some way to defend it's being the "best fit" to the
three given points. You will have to specify your goals more precisely.
-- Jerry
decvax!yale-comix!leichter leichter@yale
------------------------------
Date: Tue 26 Jun 84 11:44:04-MDT
From: Stan Shebs <SHEBS@UTAH-20.ARPA>
Subject: Frames Question
As a relief from the insubstantial debates on insubstantial souls, I
have a question about frames.
From my studies, I have observed two fundamentally ways of viewing
slots in frames: as heads of predicates, or as instance variables of
objects.
In the first view, if a FIDO frame has an AGE slot with the value 2,
then that is equivalent to making the assertion AGE(FIDO,2). Thus the
name of the slot becomes the head of a predicate. The advantages of
this view are twofold: the inheritance mechanism of a frame system
then appears as an inference rule, and slots can be made into frames
themselves, thus making meta-level knowledge easy (for instance, one
could say DATATYPE(AGE,NONNEGATIVE_NUMBER) to assert that AGE could
only hold values of a certain type). This view of slots as
first-class concepts or frames is exemplified by RLL, and by simple
frame systems built on top of logic languages.
The second view is exemplified by FRL, its descendants, and any of a
number of object-oriented systems. Here, slots are in some sense
"local" to frames or classes of frames, and an AGE of FIDO may have a
completely different meaning than an AGE of PINOT_NOIR. Meta-level
knowledge generally resides in facets and other subparts of a slot, so
in a well-developed system, the "value" of a slot is often a rather
complex entity. Interestingly enough, the facets (such as $VALUE,
$IF-ADDED etc) are usually quite consistent in meaning (which no doubt
simplifies meta-knowledge; one then needs only a few frames named
$VALUE, $IF-ADDED, ... to express the meanings of facets).
Each view can be simulated using the other. To simulate the "slot as
frame view", the "objects view" can make all slots be defined for a
toplevel frame THING, and then have frames with the same names as the
slots; while the "slot as frame view" can have slots of slot frames
that point to many different ones (so for instance the AGE slot frame
has a slot VERSIONS that points to ANIMAL_AGE and WINE_AGE slots - all
the associated paperwork is handled automatically by the system). Of
course, such simulations may be extremely inefficient! but I just
mention them to show that neither method is inherently more capable
than the other.
Now for the question: which view is favored by practitioners, and why?
Do any existing KRLs allow the view of slots to be changed according
to the problem, or do the two views require such fundamentally
different implementations that it's just better to stick to one or the
other? Is it possible to do work using frames without being concerned
about the particular view imposed by the frame system? (my own
experience says no - converting an FRL-based program to an RLL-based
one is not easy!). Are there problem domains in which one view is
distinctly superior to the other? If so, what are they, and why is
that view superior?
Any answers or insights will be greatly appreciated...
stan shebs
------------------------------
Date: Wed 27 Jun 84 11:35:48-PDT
From: Michael Walker <WALKER@SUMEX-AIM.ARPA>
Subject: AI & statistics
Ken,
Thank you for mentioning our work on RADIX in the recent AILIST
response about AI and regression analysis. It prompted me to put together
a partial list of articles in AI and statistics, which I have been meaning
to do. I've left out a number of articles by these authors in more obscure
journals and proceedings. There is also work going on at Brunel University,
and at BBN, but I haven't seen any publications from them yet. If people
have additions to make, I would be happy to collect them and send them to
the list.
If readers would like reprints, the following addresses may be
useful. Daryl Pregibon and Bill Gale can be reached at:
Bell Laboratories
600 Mountain Avenue
Murray Hill, New Jersey
07974
For D. Rodbard, write:
D. Rodbard, M.D.
National Institute of Child Health and HUman Development
National Institutes of Health
Bethesda, Maryland
Our address here at the RADIX project is:
Robert L. Blum and Michael G. Walker
RADIX Project
Department of Computer Science
Margaret Jacks Hall
Stanford University
Stanford, California
94305
Mike Walker
WALKER@SUMEX-AIM.ARPA
[Blum 82a] Blum, R.L.
Discovery and Representation of Causal Relationships from a
Large Time-oriented Clincal Database: The RX Project.
Springer-Verlag, 1982.
Vol. 19 in the Medical Informatics series edited by D.A.B.
Lindberg and P.L. Reichertz.
[Blum 82b] Blum, R. L.
Discovery, Confirmation, and Incorporation of Causal
Relationships from a Large Time-Oriented Database: The RX
Project.
Computers and Biomedical Research 15(2):164-187, 1982.
[Blum 82c] Blum, R. L.
Induction of Causal Relationships from a Time-Oriented Clinical
Database: An Overview of the RX Project.
In Proceedings of the Symposium on Computer Applications in
Medical Care. IEEE Computer Society, 1982.
[Blum 84] Blum, R.L.
Two-Stage Regression: Application to a Time-Oriented Clinical
Database.
1984.
in preparation.
[Chambers 81] Chambers, J.M., Pregibon, D., and Zayas, E.
Expert Software for Data Analysis: An Initial Experiment.
In 43rd Session ISI. Buenos Aires, Argentina, 1981.
[Gale 83] Gale, W.A., and Pregibon, D.
Using Expert Systems for Developing Statistical Strategy.
In Joint Statistical Meetings. Toronto, 1983.
[Hajek 82] Hajek, P., and Ivanek, J.
Artificial Intelligence and Data Analysis.
In COMPSTAT 1982, pages 54-60. International Association for
Statistical Computing, Physics-Verlag, Vienna, 1982.
[Rodbard 83] Rodbard, D., Cole,B.R., and Munson,P.J.
Development of a Friendly, Self-Teaching, Interactive
Statistical Package for Analysis of Clinical Research Data:
The BRIGHT STAT-PACK.
In Seventh Annual Symposium on Computer Applications in Medical
Care, pages 701-704. IEEE Computer Society, 1983.
------------------------------
Date: Wed, 27 Jun 84 09:36:41 PDT
From: Michael Pazzani <pazzani@AEROSPACE>
Subject: Spelling Correctors = Geography test correctors?
Ignoring philosophical issues (after all, this is AILIST not a bad remake
of "My Dinner With Andre") I don't feel that the spelling correctors or
the geography test correctors are really that intelligent. The geography
corrector seems to be very similar to the programs which grade SAT tests.
Surely, one wouldn't want to call a SAT test correcting program AI
even though it does a better and faster job than I would.
I think its more important to discuss how to make these programs smarter.
What would it take to have a spelling corrector find the intended word
instead of all of the possibilities? A while ago, I worked on a program
to do word sense selection. I wrote a spelling corrector for that
program which treated a misspelled word as new word whose senses were
the senses of all the possible corrections. It worked well when
things like part of speech or selectional restrictions could
disambiguate. How could one make this program smarter? Is it possible
to try the "closer" possibilities first? Can you propagate the part of
speech or semantic constraints into the search for possibilities? How
would one store a large dictionary so it is efficient to find nouns,
which are vehicles which look like "planh"? How can you detect a
spelling error if the mistake is another word? (e.g. "I just typed
rm *. Can you restore my flies from backup tape?) How do people
do this anyway?
------------------------------
Date: 23 Jun 84 8:49:24-PDT (Sat)
From: hplabs!hao!seismo!rochester!rocksvax!sunybcs!gloria!colonel @
Ucb-Vax.arpa
Subject: Re: The Turing Test - machines vs people
Article-I.D.: gloria.255
[This followup was actually written by a very clever computer program.]
As you say, the Turing test is a _conversational_ test. Do you remember
Turing's original "conversation"? "...Count me out on this. I never
could write poetry."
The whole conversation is fatuous! But then, it has no bonafide purpose.
It was merely set up by a scientist to prove something. Nothing would
be easier, for that matter, than to program a computer to take part in
what Berne calls "8-stroke rituals":
Hi.
Hi.
How are you?
Fine. How are you?
Fine. Nice day, isn't it?
Yes.
Well, goodbye.
Goodbye.
But would you want to carry on such a conversation with a computer?
One converses socially only with conversers that one knows to be people.
Col. G. L. Sicherman
...seismo!rochester!rocksanne!rocksvax!sunybcs!gloria!colonel
------------------------------
End of AIList Digest
********************
