Re: Genetic programming and code evolution

henry@spsystems.net (Henry Spencer)
24 Dec 2005 21:21:09 -0500

          From comp.compilers
Related articles
Genetic programming and code evolution gael@tele2adsl.dk (=?ISO-8859-1?Q?Ga=EBl_Rosset?=) (2005-12-23)
Re: Genetic programming and code evolution mailbox@dmitry-kazakov.de (Dmitry A. Kazakov) (2005-12-24)
Re: Genetic programming and code evolution henry@spsystems.net (2005-12-24)
| List of all articles for this month |
From: henry@spsystems.net (Henry Spencer)
Newsgroups: comp.compilers
Date: 24 Dec 2005 21:21:09 -0500
Organization: SP Systems, Toronto, Canada
References: 05-12-065 05-12-074
Keywords: optimize
Posted-Date: 24 Dec 2005 21:21:09 EST

Dmitry A. Kazakov <mailbox@dmitry-kazakov.de> wrote:
>From the machine learning perspective the way a genetic algorithm
>works is selection. If you want to optimize for some criterion (like
>speed) you need to measure it to be able to select the most promising
>mutations. Measuring program speed is a way too expensive for the
>incredible number of mutations you would need.


People *have* used genetic algorithms to do things like finding better
image-processing techniques, evaluating results by measuring program
speed. However, Dmitry is basically correct -- it's appallingly slow and
expensive, verging on impractical, even with carefully-selected program
representations.
--
spsystems.net is temporarily off the air; | Henry Spencer
mail to henry at zoo.utoronto.ca instead. | henry@spsystems.net

Post a followup to this message

Return to the comp.compilers page.
Search the comp.compilers archives again.