| View previous topic :: View next topic |
| Author |
Message |
SuperFly
Guest
|
 Posted: Sun Aug 31, 2003 12:28 am Post subject: genetic algorithm for compression .. would it work ? |
 |
|
Hi,
I just read some articles on genetic algorithms and started to wonder
if it would be a good idea to make a genetic algorithm program to
brute force compress a file. Or is the result of something like an
arithmetic compressor the optimal way. And would a good genetic
program give the same result at best?
And is it something like finding the way home from work with your eyes
closed. If you wait long enough you will find your home but it will
never beat the shortest 'eyes open' route.
But I can also imagine that investing a huge amount of cpu/memory
power can give a new more optimal compression result. Maybe even way
better.
Any ideas on this? |
|
| |
|
 Back to top |
Jesper Nordenberg
Guest
|
| Posted: Mon Sep 01, 2003 1:30 pm Post subject: Re: genetic algorithm for compression .. would it work ? |
|
|
fake@email.com (SuperFly) wrote in message news:<3f50f487.797390@news.euronet.nl>...
[quote]I just read some articles on genetic algorithms and started to wonder
if it would be a good idea to make a genetic algorithm program to
brute force compress a file. Or is the result of something like an
arithmetic compressor the optimal way.
[/quote]
Arithmetic compression is close to optimal for a given a model, but
the data model is the most important aspect of a compression
algorithm. If you can>t find a good model, the data will not compress.
Here>s where genetic algorihtms might be useful. A well designed
genetic algorithm could find a model in an input which a normal
compressor can>t compress. One disadvantage with a genetic algorithm
is that you need to store a description of the model in the compressed
data, something that you don>t have to do when using hard coded models
like PPM or BWT.
[quote]And would a good genetic
program give the same result at best?
[/quote]
It could. But remember that PPM and BWT are pretty complicated models
that would be hard for a genetic algorithm to "invent". I doubt you
will get as good compression for text, executables, images etc. as the
best algorithms, but for more unusual data it could very well beat
common compression algorithms.
/Jesper Nordenberg |
|
| |
|
| Back to top |
SuperFly
Guest
|
| Posted: Mon Sep 01, 2003 3:01 pm Post subject: Re: genetic algorithm for compression .. would it work ? |
|
|
On 1 Sep 2003 01:30:43 -0700, megagurka@yahoo.com (Jesper Nordenberg)
wrote:
[snip]
[quote]And would a good genetic
program give the same result at best?
It could. But remember that PPM and BWT are pretty complicated models
that would be hard for a genetic algorithm to "invent". I doubt you
will get as good compression for text, executables, images etc. as the
best algorithms, but for more unusual data it could very well beat
common compression algorithms.
[/quote]
I created some files with obvious patterns for a human and ran them
against some regular compression schemes. And most of the time the
results weren>t that good.
Which made me wonder if there has been research in this area. And
maybe there already is some good (genetic) 1-D pattern recognition
software/source code out there that i>m not aware of. If so , I wonder
what to expect of it (results), and where to get it.
Anybody has links & information on this? |
|
| |
|
| Back to top |
|
