| |
 |
|
|
Science Forum Index » Compression Forum » genetic based algorithms
Page 1 of 1
|
| Author |
Message |
| SuperFly |
 Posted: Wed Sep 17, 2003 1:34 am |
|
|
|
Guest
|
Hi,
I want to write a genetic/self learning program that does a reversible
transformation on a piece of data and then compares the result to the
previous results and adjusts according to it, to learn something about
genetic coding.
It would be cool if I could use a method that has the potential to
grow in regular algorithms but also in new ones. I've got some ideas
to make this work. But before I reinvent the wheel 3 questions.
1) Does anybody know about similar projects
2) What basic transformation(s) would be best to use.
3) Which algorithm(s) would be best to test the results.
Any ideas on this?
(If you think this will never work, and have a motivation, i would
also like to hear it) |
|
|
| Back to top |
|
| SuperFly |
| Posted: Wed Sep 17, 2003 1:39 am |
|
|
|
Guest
|
On Wed, 17 Sep 2003 09:34:57 +0200, SuperFly <fake@email.com> wrote:
Quote: Hi,
I want to write a genetic/self learning program that does a reversible
transformation on a piece of data and then compares the result to the
previous results and adjusts according to it, to learn something about
genetic coding.
It would be cool if I could use a method that has the potential to
grow in regular algorithms but also in new ones. I've got some ideas
to make this work. But before I reinvent the wheel 3 questions.
1) Does anybody know about similar projects
2) What basic transformation(s) would be best to use.
3) Which algorithm(s) would be best to test the results.
Any ideas on this?
(If you think this will never work, and have a motivation, i would
also like to hear it)
To be complete : the learning goal is gaining compression ..  |
|
|
| Back to top |
|
| David A. Scott |
| Posted: Wed Sep 17, 2003 6:38 am |
|
|
|
Guest
|
SuperFly <fake@email.com> wrote in
news:3s3gmvs82v37d9t1ovbn29tn8qnl2gndve@4ax.com:
Quote:
To be complete : the learning goal is gaining compression .. :)
Most people think self modifing code is not mantainable.
However say you wrote a simple bijective huffman or arithmetic
coder or better yet something along the lines of my bijective
LZW. Whenever you use it you can store the table somewhere and
use the old table as starting point for next file to compress.
Its messy and you have to keep track of what was compressed and in
what order but hopefully if your using this over and over again
for same types of files it would learn to compress better over
time.
David A. Scott
--
My Crypto code
http://cryptography.org/cgi-bin/crypto.cgi/Misc/scott19u.zip
http://cryptography.org/cgi-bin/crypto.cgi/Misc/scott16u.zip
http://www.jim.com/jamesd/Kong/scott19u.zip old version
My Compression code http://bijective.dogma.net/
**TO EMAIL ME drop the roman "five" **
Disclaimer:I am in no way responsible for any of the statements
made in the above text. For all I know I might be drugged.
As a famous person once said "any cryptograhic
system is only as strong as its weakest link" |
|
|
| Back to top |
|
| nicobi |
| Posted: Fri Sep 19, 2003 8:56 am |
|
|
|
Guest
|
Hello !
I guess, this could be a pretty good idea indeed. In fact, lot's of
compression sofware nowadays are taking this direction. Well not in
the metaheuristic point of view (the genetic algorithm being one of
them), but more in a simple heuristic fashion.
Example for these are the famous cabac, cavlc, or even grasp.
cabac or cavlc are patented algos that can be found in the last H264
codec (you could have a look at
http://bs.hhi.de/~suehring/tml/download/ to see the implementation).
In fact, cabac is one of the best compression algorithm nowadays. bac
meaning binary algorithm coding, ca : context adaptatif. idem for
cavlc (a classical vlc but with context adaptation). Grasp is a
further step into cabac : the adaptation being further proceeded
thanks to a kind of tree structure (cabac + a touch of memory of some
forgotten directions token from cabac). CABAC is arround 10-20% better
regarding to classical UVLC. Grasp should be arround 10% better than
cabac (dixit : Thomas Wiegand, Co chair man of H264). But in all those
adaptatives algorithm, the adaptative step is heuristic based.
So, I do personnaly think that doing the adaptation step with a kind
of genetic algorithm, simulated annealing, or even (why not !) ants
algorithm (i.e. meta heuristic) is possible, and could nicely improve
the compression. The only issue being : how fast can you be then ??
(but anyway doing a quick genetic research could only improve the
result, isn't it)
Another way, could be to mix up several compression algorithm, the
sequences of datas being chosen thanks to a meta heuristic... Er...
well... if necessary ?
So anyway, good candidates are in my opinion an inspired and adpatated
cabac.
nicobi
Quote: Hi,
I want to write a genetic/self learning program that does a reversible
transformation on a piece of data and then compares the result to the
previous results and adjusts according to it, to learn something about
genetic coding.
It would be cool if I could use a method that has the potential to
grow in regular algorithms but also in new ones. I've got some ideas
to make this work. But before I reinvent the wheel 3 questions.
1) Does anybody know about similar projects
2) What basic transformation(s) would be best to use.
3) Which algorithm(s) would be best to test the results.
Any ideas on this?
(If you think this will never work, and have a motivation, i would
also like to hear it) |
|
|
| Back to top |
|
| |
|
Page 1 of 1
All times are GMT - 5 Hours
The time now is Sat Nov 22, 2008 7:53 pm
|
|