psychoceramics: Re: psychoceramics-digest V1 #131

[Neil Harris <neil @ nharris.demon.co.uk>]

Has anyone been following the Jules Gilbert debate on comp.compression?

Jules asserts that he can compress 'random appearing data' (RAD), something
widely believed impossible, and furthermore that he can compress it by
quite startling amounts.

>From the point of view of orthodox computer science, this is like 
claiming to be able to square the circle or build a perpetual motion
machine.  Were such a thing possible, the inventor would be hailed as 
a genius.

The discussion has grown quite heated, with claims, counter-claims and
accusations flying in all directions.  Here is one such gem in which
Jules appears to claim

1   "First, neither I nor anyone else can compress random-appearing data files."
2   "But my program works every time on random-appearing input data"
3   "Instead I have a different 'compression' model.  One way my model is 
     different from other compressors is that it doesn't compress at all."
and, in <4un70u$a--@s--.ici.net>
4   "But I can compress RAD files 100:1 if allowed to re-process the output
     through the same process."

>From: c--@s--.ici.net (Jules Gilbert)
>Newsgroups: comp.compression
>Subject: Re: Jules Gilbert's New Compresssion Technology
>Date: 7 Aug 1996 11:31:49 -0400
>Organization: The Internet Connection, Inc.
>Lines: 43
>Message-ID: <4uact5$b--@s--.ici.net>
>References: <4t3rbv$1--@s--.ici.net> <DvHBsp.K8y@world.std.com> <4u80ga$9m9@sol.sun.csd.unb.ca> <DvqHEI.997@world.std.com>
>NNTP-Posting-Host: spock
>Keywords: random-appearing data ('RAD')
>
>I am amazed at the level of confusion (and other negative mental things)
>present on this news-group.  But I can repeat something that may help lower the
>confusion level.  (I've said this before.)
>
>First, neither I nor anyone else can compress random-appearing data files.  And
>even if one is fortunate enough to be given, say an older compressed file,
>which did not compress very well, surely AT BEST, a one-time 2:1 compression
>factor is the very best that one could hope for in such a situtation. (This
>would be a once in a lifetime gift!)
>
>But my program works every time on random-appearing input data, such as
>produced by the output of say, ARJ, or some other high quality conventional
>compressor.  The only limitation associated with my technique is a requirement
>for a minimum file or message size, of say 200k.
>
>I have a process which pretty much avoids the paradigm employed by all present
>day compressor tools, that of relying on prior observed CONTEXT and/or HISTORY
>as a guide to predicting/compressing the next character.
>
>Instead I have a different 'compression' model.  One way my model is different
>from other compressors is that it doesn't compress at all.  If given a large
>buffer, it emits a large buffer of pretty much the same size.  But the contents
>of the result buffer are readily compressible, whereas (if given random-
>appearing input), the source buffer is not compressible at all.
>
>Recently, I made a change which coupled a conventional compressor to back-end
>my compressor to improve the seamlessness of the program, but the basic under-
>pinnings did not change.
>
>Now, many students of computer science will (correctly) have a problem with my
>claim that I accept a 'source' buffer and produce a 'target' buffer, where the
>input is random-appearing and the output is readily compressible.  I realize
>that.  None-the-less, that is exactly what is happening.
>
>And, I do it with very conventional code.  Just loops, compares, and such. 
>Nothing fancy.  
>
>In the past, I have made jokes ("I use the space between the bits.") and many
>have thought my remarks simple fraud.  But every statement made here is
>absolutely accurate.
>
>
>Jules Gilbert ('c--@i--.net')

The debate has now been joined by 

>From: u--@a--.com (U137)
>Newsgroups: comp.compression
>Subject: Re: Jules Gilbert's Compression Technology
>Date: 18 Aug 1996 12:48:11 -0400
>Organization: America Online, Inc. (1-800-827-6364)
>Lines: 11
>Sender: r--@n--.news.aol.com
>Message-ID: <4v7hgb$s--@n--.news.aol.com>
>References: <4v60ao$n--@s--.ici.net>
>Reply-To: u--@a--.com (U137)
>NNTP-Posting-Host: newsbf02.mail.aol.com
>
>Wh have just been issued a patent (US. #5,533,051) and have several more
>pending on a new method for data compression. It will compess all types of
>data, including "random", and data containing a uniform distribution of
>"0's" and "1's".
>
>In a few days I will be posting an article discussing one of our methods.
>It will be entitled, Hyper Space(R) a method for data compression. This
>particular method is patent pending, but I am releasing the source code
>and the core encoding and decoding algorithms. 
>
>The code is written in BASIC and is about 5K.

comp.compression, folks.  Check it out.

-- Neil