[vorbis] New decode chip

[Shawn]

A functionality must have a certain level of specificity in order to
evolve a solution.

You would need to carefully design inputs and outputs so that it would
be unlikely for something evolved to be able to go wrong in some
unforseen case.

It's definitely possible, but VERY easy to screw up really bad.

On 08/06, Alejandro G. Belluscio said something like:
> Hello Shawn,
>   I didn't had time to look at the specific method. But genetic
> algorithms are widely used and even though it's not so well understood
> they usually "just work" (but it can be shown why if you like
> difference math). The fact is that if you start from a good starting
> point (let's say a reference implementation) and you separate it a
> lots of stages (or let them evolve independently, which is the same)
> you can get very fast convergence to the optimum.
> 
> Monday, August 05, 2002, 9:14:11 PM, you wrote:
> 
> S> I remember it.  Specifically, it is genetic algorithms as they relate to
> S> FPGAs.
> S> http://slashdot.org/article.pl?sid=01/12/29/007258&mode=thread&tid=137
> 
> S> At issue is whether you could feed it the source (an ogg) and a dest
> S> (name your output fmt) and expect an ogg decoder to evolve. I think then
> S> you'd get a specific this file to that file changer.
> 
> S> You'd need to break the ogg encoding functions into their atomic pieces
> S> and then evolve routines to duplicate the functionality. It's
> S> non-trivial just to plan such an undertaking for ogg vorbis.
> 
> S> Even then, there's a question about how deterministic the behavior of
> S> these magic programs would be, especially on different hardware.
> 
> S> Just wait for clockless, analog and FPGA based systems, then /maybe/
> S> quantum computing. They'll probably each find a niche in very different
> S> hardware. (FPGA's have potential in PDAs on up, but qbits will always
> S> likely be the realm of larger systems)
> 
> S> On 08/05, Oliver D. Jones said something like:
> >> Does anyone remember that experiment done in New Scientist, where they
> >> had a FPGA "mutate" until it did a specific job well enough? I wonder if
> >> something similar could be done to encode Ogg Vorbis. It might come up
> >> with some improvements that even the cleverest hand-tuners may miss.
> >> 
> >> Oliver.
> >> 
> >> -----Original Message-----
> >> From: owner-vorbis at xiph.org [mailto:owner-vorbis at xiph.org] On Behalf Of
> >> Anthony Frazier
> >> Sent: Monday, August 05, 2002 9:53 PM
> >> To: vorbis at xiph.org
> >> Subject: Re: [vorbis] New decode chip
> >> 
> >> 
> >> On 5 Aug 2002 at 22:28, Olaf Marzocchi wrote:
> >> 
> >> > Newbie question: every time we reach a better compression we have to 
> >> > use
> >> > more power (rar compared to zip, mp3 to one of the first compression 
> >> > schemes, divx compared to cinepak,...). Ogg compress more than mp3 but
> >> at 
> >> > about the same speed.
> >> > Is there an explanation to a such silly question?
> >> 
> >> The easy answer: "Vorbis works smarter, not harder."
> >> 
> >> Actually, Vorbis is slower at encoding than some of the better tuned mp3
> >> 
> >> encoders, but that's mostly because it hasn't had the time to be tuned
> >> as 
> >> much.
> >> 
> >> I imagine that if someone wanted to fork the Vorbis encoder and give it
> >> lots 
> >> of hand tuning and assembly code, you could probably get some pretty
> >> hefty 
> >> improvements.  (For example, like GoGo derived from Lame.)
> >> 
> >> Pax,
> >> 
> >> Anthony Frazier
> >> 
> >> 
> >> --- >8 ----
> >> List archives:  http://www.xiph.org/archives/
> >> Ogg project homepage: http://www.xiph.org/ogg/
> >> To unsubscribe from this list, send a message to
> >> 'vorbis-request at xiph.org' containing only the word 'unsubscribe' in the
> >> body.  No subject is needed. Unsubscribe messages sent to the list will
> >> be ignored/filtered.
> >> 
> >> 
> >> 
> >> --- >8 ----
> >> List archives:  http://www.xiph.org/archives/
> >> Ogg project homepage: http://www.xiph.org/ogg/
> >> To unsubscribe from this list, send a message to 'vorbis-request at xiph.org'
> >> containing only the word 'unsubscribe' in the body.  No subject is needed.
> >> Unsubscribe messages sent to the list will be ignored/filtered.
> S> --
> S> Shawn Leas
> S> core at enodev.com
> 
> S> On the other hand, you have different fingers...
> S>                                                 -- Stephen Wright
> 
> 
> --- >>8 ----
> S> List archives:  http://www.xiph.org/archives/
> S> Ogg project homepage: http://www.xiph.org/ogg/
> S> To unsubscribe from this list, send a message to 'vorbis-request at xiph.org'
> S> containing only the word 'unsubscribe' in the body.  No subject is needed.
> S> Unsubscribe messages sent to the list will be ignored/filtered.
> 
> 
> 
> -- 
> Best regards,
>  Alejandro                            mailto:baldusi at uol.com.ar
> 
> 
> 
> 
> --- >8 ----
> List archives:  http://www.xiph.org/archives/
> Ogg project homepage: http://www.xiph.org/ogg/
> To unsubscribe from this list, send a message to 'vorbis-request at xiph.org'
> containing only the word 'unsubscribe' in the body.  No subject is needed.
> Unsubscribe messages sent to the list will be ignored/filtered.

--
Shawn Leas
core at enodev.com

The other day, I was walking my dog around my building...  on
the ledge.  Some people are afraid of heights.  Not me, I'm
afraid of widths.
						-- Stephen Wright

<p>--- >8 ----
List archives:  http://www.xiph.org/archives/
Ogg project homepage: http://www.xiph.org/ogg/
To unsubscribe from this list, send a message to 'vorbis-request at xiph.org'
containing only the word 'unsubscribe' in the body.  No subject is needed.
Unsubscribe messages sent to the list will be ignored/filtered.