Copy Link
Add to Bookmark
Report
AIList Digest Volume 8 Issue 053
AIList Digest Tuesday, 16 Aug 1988 Volume 8 : Issue 53
Responses:
Sigmoid transfer function
Feigenbaum's citation
----------------------------------------------------------------------
Date: 14 Aug 88 04:51:18 GMT
From: hubcap!shorne@gatech.edu (Scott Horne)
Subject: Re: Sigmoid transfer function
>From article <517@eeg.UUCP>, by marcus@eeg.UUCP (Mark Levin):
>> >Try this one : f(x) = x / (1 + |x|)
>
> If the function is graphed with the x-axis in log coordinates the
> function becomes your favorite Sigmoid Function.
/ x \
Then why not use f(x) = exp | ------- | ?
\ 1 + |x| /
--Scott Horne
uucp: ....!gatech!hubcap!scarlett!{hazel,citron,amber}!shorne
(If that doesn't work, change "scarlett" to "scarle")
(If *that* doesn't work, send to cchang@hubcap.clemson.edu)
(If *that* doesn't work, wait until January & write me at Yale)
SnailMail: Scott Horne, 812 Eleanor Dr., Florence, SC 29501
VoiceNet: 803 667-9848 (home); 803 669-1912 (office)
------------------------------
Date: 14 Aug 88 16:13:15 GMT
From: buengc!bph@bu-cs.bu.edu (Blair P. Houghton)
Subject: Re: Sigmoid transfer function
In article <2628@hubcap.UUCP> shorne@citron writes:
>From article <517@eeg.UUCP>, by marcus@eeg.UUCP (Mark Levin):
>>> >Try this one : f(x) = x / (1 + |x|)
>
> / x \
>Then why not use f(x) = exp | ------- | ?
> \ 1 + |x| /
I always thought the real sigmoid function was erf(x), which
results as the integral of a Normal distribution, which is
the all-time fave among scientists for describing natural
occurrences.
2 2
It's not much fun to calculate... but then, neither is exp(-x /sigma ).
Ya gotta run into Taylor's theorem somewhere...
It (erf(x)) is available in most math-library packages, though.
--Blair
"MacLaurin and Taylor and
Dirac are sitting in a leaking
balloon over the ocean, discussing
which will be the one to leap
into the sea..."
------------------------------
Date: 13 Aug 88 22:43:25 GMT
From: uccba!uceng!dmocsny@ohio-state.arpa (daniel mocsny)
Subject: Re: Feigenbaum's citation
In a previous article, David Nelson writes:
> To be replaced by the optimistic hope that dumping lots of examples into a
> dumb neural net will produce something profound :-)
>
> daven (Dave Nelson)
> arpa: daven @ lll-crg.llnl.gov
> uucp: ...{seismo,mordor,sun,lll-lcc}!lll-crg!daven
I for one would be happy to forego the profound to realize the usefully
mundane. I can scribble a diagram in one minute that I need thirty minutes
to write a LaTeX picture description for. If my computer was not so
severely retarded, I might be able to make my book deadline.
Neural nets may emerge as quick-and-dirty translators to aid us in slicing
through the Babel out there (and in here :-) ). Give me a computer that
has some idea of what I am saying, and I will handle the profundities. (Or
I will at least try...)
While we are dreaming, how close are you net.AI types to realizing a device
that can obey commands like this: `Examine all the applied math papers
relating to subjects x, y, ..., z, and find the ones bearing on my
current problem q.' I realize that's pretty vague, but I'm not talking about
a straight database query here. What we scientists and engineers really need
is some rational way for us to post our work so our peers can quickly and
transparently access it. How many times have each of us worked for months
on something, only to find that Dr. So-and-so already published in some
obscure conference proceedings not in your local library? Or worse, we did
_not_ find out? If a solution exists, I see it including (1) hardware
advances (terabytes of storage to handle all the scientific literature),
(2) political reform (intellectual property laws need to serve the
scientific community, not the science publishers), (3) standards (before
a computer can get at the technical literature, we need to agree on how
we plan to specify our pages), and (4) of course, the AI (the system must
understand NL queries from techno-types, and retrieve references by
content).
I know I'm probably out to lunch here, but I can't help thinking about
those research $$$ evaporating while I play librarian (and not well, at
that). Not to mention my altogether-too-brief lifespan allotment...
Dan Mocsny, u. of cincinnati *** standard disclaimer ***
------------------------------
Date: 12 Aug 88 08:12:12 GMT
From: kddlab!atr-la!geddis@uunet.uu.net (Donald F. Geddis)
Subject: Re: Feigenbaum's citation
In a previous article, Earl H. Kinmonth writes:
> As I remember, Feigenbaum achieved notoriety for his (probably
> largely ghosted book) on the Japanese "Fifth Generation
> Project." Did anything ever come out of the Fifth Generation
> project other than big lecture fees for Feigenbaum to go around
> warning about the Japanese peril?
>
> Is he really a pompous twit (the impression given by the book) or
> is that due to the scatter-brained ghost writer?
This kind of posting really bothers me. Given the current silence surrounding
results from the Fifth Generation Project, it may be fair to argue that the
"Japanese peril" was overblown. But it is crucial to remember that things are
always much clearer in hindsight, and we have a great advantage today over
Feigenbaum when the book was written. (I leave out whether Feigenbaum actually
wrote the book himself, as I have no information on the matter. But then,
obviously, neither does Earl.)
Even given current results of the Project, it is even more difficult to be
sure that the book's predictions were poor for its time. Japan was beginning
a massive, well-funded AI effort, and the U.S. lacked (still lacks?) any
similar project. It is almost always impossible to predict what will happen
to research on the cutting edge of technology, and there is a very real
danger that ignoring such a "threat" can significantly worsen the U.S.'s
position in AI.
And of course, none of the proceeding is any excuse for suggesting that
Feigenbaum is a "pompous twit". Earl, you really ought to show more
maturity than that. From personal contact (he has given me informal
advice over the last three months), I can say that he has been very helpful
and informative to at least one graduate student peon. He has a deep interest
in the Japanese, and is firmly convinced that the efforts of Japan are
largely ignored in the United States, much to the future regret of the U.S.
He has given me pages of unsolicited travel advice (I'm spending the summer
in Japan), and offered tours of his research lab at the Knowledge Systems
Lab at Stanford. At no time have I sensed anything that might be termed
"pompous".
Now while I know that my limited and personal experience tells little about
Feigenbaum's overall character, it seems that unless Earl has very good
backing, comments such as he made are better left unsaid.
-- Don
(Dis-Disclaimer: I am a Ph.D. student at Stanford in AI, so my opinions are
biased and probably to be disregarded on this subject.)
--
"You lock the door, and throw away the key
There's someone in my head, but it's not me." -- Pink Floyd
Internet: Geddis@Score.Stanford.Edu (which is forwarded to Japan...)
USnail: P.O. Box 4647, Stanford, CA 94309 USA
------------------------------
Date: 15 Aug 88 23:00:52 GMT
From: cck@deneb.ucdavis.edu (Earl H. Kinmonth)
Subject: Re: Feigenbaum's citation
In a previous article, Donald F. Geddis writes:
>In a previous article, Earl H. Kinmonth writes:
>> As I remember, Feigenbaum achieved notoriety for his (probably
>> largely ghosted book) on the Japanese "Fifth Generation
>> Project." Did anything ever come out of the Fifth Generation
>> project other than big lecture fees for Feigenbaum to go around
>> warning about the Japanese peril?
>>
>> Is he really a pompous twit (the impression given by the book) or
>> is that due to the scatter-brained ghost writer?
>
>This kind of posting really bothers me. Given the current silence surrounding
It should bother you more that a silly book like the Fifth Generation
Project would get taken seriously, testimony to the low level of
knowledge about Japan prevailing in this country.
>results from the Fifth Generation Project, it may be fair to argue that the
>"Japanese peril" was overblown. But it is crucial to remember that things are
That appears to be putting it mildly. My impression is that the gap
between hype and delivery is such that in almost any other context,
people would be screaming fraud and calling their lawyers.
>always much clearer in hindsight, and we have a great advantage today over
>Feigenbaum when the book was written. (I leave out whether Feigenbaum actually
>wrote the book himself, as I have no information on the matter. But then,
>obviously, neither does Earl.)
Not quite; I've read it. Have you? I've also heard (second
hand) that Feigenbaum has blamed the hype on McCorduck (his "joint
author").
>Even given current results of the Project, it is even more difficult to be
>sure that the book's predictions were poor for its time. Japan was beginning
>a massive, well-funded AI effort, and the U.S. lacked (still lacks?) any
Is that bad? The whole course of postwar Japanese economic development
and that of the US prior to World War II shows that there is a great
advantage to letting others do the big ticket, high risk research, and
concentrating instead on commercialization and mass production.
>similar project. It is almost always impossible to predict what will happen
>to research on the cutting edge of technology, and there is a very real
>danger that ignoring such a "threat" can significantly worsen the U.S.'s
>position in AI.
So what? Name the areas in which the Japanese have been at the cutting
edge of science (science, not technology). Name the areas in which the
US has been at the cutting edge. Compare the size of the lists. Now
compare the growth in real GNP, growth in real income, etc. for the two
countries.
It has been reported that the Fifth Generation project was so oversold
that the gap between the hype for it and the results has soured the
climate for funding other basic research in Japan....
There is also an even greater danger that blowing up a trivial or even
totally non-existent threat will divert scarce resources into
irrelevant or counter-productive areas. The US-USSR arms race provides
many examples of this.
I vaguely remember reading that the Fifth Generation Project (the book)
was in part responsible for the Pentagon asking for a 650 megabuck
boondoogle to meet the Japanese challenge in AI.
>And of course, none of the proceeding is any excuse for suggesting that
>Feigenbaum is a "pompous twit". Earl, you really ought to show more
I agree, but read what I originally wrote. I did not use the hype
concerning the Fifth Generation as the basis for deciding that he was a
"pompous twit," and did not in fact say that. I said that's the way he
comes over in the book! Read the book. I'm willing to bet that "modest"
or "self-effacing" are not the adjectives you would apply to the
personality depicted there.
>maturity than that. From personal contact (he has given me informal
READ THE BOOK. He comes over as a "legend in his own mind." As
I indicated, that may be due to the ghost-writer ("joint
author"), but I find it hard to take anyone seriously who'd allow
their name on such a puff piece unless they're running for public
office....
>advice over the last three months), I can say that he has been very helpful
>and informative to at least one graduate student peon. He has a deep interest
>in the Japanese, and is firmly convinced that the efforts of Japan are
But little real knowledge, again judging from the book.
>largely ignored in the United States, much to the future regret of the U.S.
>He has given me pages of unsolicited travel advice (I'm spending the summer
>in Japan), and offered tours of his research lab at the Knowledge Systems
>Lab at Stanford. At no time have I sensed anything that might be termed
>"pompous".
That may well be the case, and if it is, that is precisely what I
asked - specifically, was Feigenbaum in the flesh, the "pompous
twit" presented in the book. An adequate response in this context
would have been "No."
>Now while I know that my limited and personal experience tells little about
>Feigenbaum's overall character, it seems that unless Earl has very good
>backing, comments such as he made are better left unsaid.
My knowledge is inherently limited to Feigenbaum, the public
personality. Aside from the book, I've read reports of a couple
of his talks. Frankly, he comes over like a silicon snake oil
salesman. I get the same feeling from his pronouncements that I
get from each Pentagon report on Soviet military superiority: it
may be true, but I'd much rather hear the message from someone
who does not stand to gain from drum beating.
PS:
If you'd like some unartificial intelligence about Japan, I may
be able to help. I've lived there a total of six years including
three years as a graduate researcher (modern Japanese social and
economic history) at the University of Tokyo.
> -- Don
>
>(Dis-Disclaimer: I am a Ph.D. student at Stanford in AI, so my opinions are
>biased and probably to be disregarded on this subject.)
>--
I don't know about Stanford, but my graduate school experience
was such that any faculty person who even acknowledged the
existence of students appeared kind and loving....
>Internet: Geddis@Score.Stanford.Edu (which is forwarded to Japan...)
>USnail: P.O. Box 4647, Stanford, CA 94309 USA
E H. Kinmonth Hist. Dept. Univ. of Ca., Davis Davis, Ca. 95616
916-752-1636/0776
Internet: ehkinmonth@ucdavis.edu
cck@deneb.ucdavis.edu
BITNET: ehkinmonth@ucdavis
UUCP: {ucbvax, lll-crg}!ucdavis!ehkinmonth
{ucbvax, lll-crg}!ucdavis!deneb!cck
------------------------------
End of AIList Digest
********************
