home *** CD-ROM | disk | FTP | other *** search
- Path: sparky!uunet!gatech!usenet.ins.cwru.edu!agate!nobozo.CS.Berkeley.EDU!forrest
- From: forrest@nobozo.CS.Berkeley.EDU (Jon Forrest)
- Newsgroups: comp.compression
- Subject: Is This Possible (or is it B.S)? [200:1 Lossless Compression]
- Date: 22 Jan 1993 00:08:26 GMT
- Organization: University of California, Berkeley
- Lines: 68
- Distribution: world
- Message-ID: <1jndtq$odg@agate.berkeley.edu>
- NNTP-Posting-Host: nobozo.cs.berkeley.edu
-
- I'm the first to admit that I don't know anything about compression.
- That's why I'm extremely puzzled by the following abstract of a talk that's
- going to be given here at Berkeley. Is this B.S. or should I raise
- my arms and cry hallelujah?
-
- ---
- CS 298-9
- Sequoia 2000 Seminar
-
- Joe Bugajski, Founder and President
- Triada Ltd.
-
- TRIADA'S 200:1 (LOSSLESS)
- DATA COMPRESSION ENGINE
-
- Friday, January 29, 1993
- 1:00 p.m. - 3:00 p.m.
- 608-7 Evans Hall
-
-
- Triada, a development stage company located in Michi-
- gan, has developed an encoding transform capable of up to
- 200:1 lossless data compression. The theoretical underpin-
- ning of the transform is believed to extend classical infor-
- mation theory. The new theory treats an information source
- that evolves from a null state to stationarity. A model
- based on the theory consists of an energetically conserva-
- tive field of uniformly distributed photons and an observer
- randomly walking through this field. Observations are made
- with a bandwidth-limited set of transducers (an "eye") that
- respond to events during the walk. Each transducer in the
- set is uniquely tuned to a sub-band and generates a signal
- when it receives a signal. The output signals from the
- input set of transducers are "observed" by a second set of
- transducers (a mid brain), except the bandwidth limit of the
- second set of transducers is a non-overlapping lower fre-
- quency range than the bandwidth of the first set of trans-
- ducers, and only events from the first transducer set are
- "seen" by the second. This procedure is repeated n times or
- through N levels or "time domains," td. These levels form a
- hierarchy of planes of "memories," (a visual cortex), and
- individual (associative) memories (transducers) are unique.
-
- The dynamics of the digital computer equivalent of this
- model were studied during the past four years. In particu-
- lar, the size of the memory sets was measured as a function
- of events "seen" at level td, where the input data were (not
- compressed) randomly collected, large samples of text,
- black/white images, eight-bit grey scale images, and copies
- of data bases. The number of memories formed were always
- Boltzman distributed, and sub-band frequencies at every
- memory level, or td, exhibited by the Self- Organized Criti-
- cality (SOC) theory [Bak et al. (1987) Phys Rev Lett: 59,
- 381-384). Because the memories self organize and the digi-
- tal version of the transform is invertible, a real-time
- digital implementation of the memory structure can be built.
- Additional tests are underway to confirm these results for a
- larger variety of data types using greater than 10 gigabyte
- data samples.
-
- Triada's founder and chief theoretician, Joe Bugajski,
- a mathematician, spent the first 13 years of his career
- doing information system design at Ford Motor Company's
- Research and Engineering Center. He has devoted the last
- seven years to data compression and related topics.
- --
- Anything you read here is my opinion and in no way represents the Univ. of Cal.
- Jon Forrest WB6EDM forrest@postgres.berkeley.edu 510-643-6764
-