[Techtalk] Prolog compilers, neural nets (was: Running Matlab & Nural
Networking (Re: Remote X applications )
David Cohen
qohen at post.harvard.edu
Tue Mar 16 20:19:09 EST 2004
>Finally there is one off the planet suggestion.
>Prolog would be great to write neural nets in but being an interpreted
>language it would take "400 years to run" (to quote my lecturer). I don't
>suppose there is any such strange best as a prolog compiler?
There have been Prolog compilers for at least 1.5 to 2 decades. As with
Lisp, I'd assume that most Prologs have been doing incremental compilation
or the like for a long time, so you get the feel of working with an
interpreted language while actually getting things compiled.
As far as getting a Prolog system to work with, right now, the freebie that
people are using appears to be SWI-Prolog, for which binaries are available
for Windows, Mac and Linux:
http://www.swi-prolog.org
There's an IDE available too, called XPCE--look on the site and you'll see it.
Now, as far as doing neural nets with Prolog, I have to say the idea sounds
a bit dodgy for the simple reason that the two approaches, logic and neural
net, were typically viewed back in the heyday of commercial AI as being at
odds with each other--the rule-based approach that attempts to mimic
reasoning vs. the mindless numerical/statistical approach. Unlike some
applications, I don't see any advantage that Prolog would give you here.
And if you go Googling around, I don't think you'll see much, if anything,
involving implementing neural nets in Prolog. Of course, SWI-Prolog has a
foreign-function interface, so you could call out to a neural net written
in C/C++ interface but in that case, it'd make more sense to just stick
with C/C++.
Mind you, if you want to work with this type of high-level language, Common
Lisp has excellent numerical facilities, as this discussion between Richard
O'Keefe (a Prolog legend) and Daniel Lakeland
mentions:
http://www.swi.psy.uva.nl/projects/SWI-Prolog/mailinglist/archive/old/2389.html
The money quote (from Lakeland, I believe):
> > Sadly as much as I enjoyed working on computer algebra and
> > solving numerical computations I have to admit that I haven't
> > done it for far too long, nor in anything more advanced than C++
> > (shockingly, as I can't abide C style languages, my first choice
> > these days if I were to start working on that stuff again would
> > be Common Lisp)
Hope this helps.
___
DC
More information about the Techtalk
mailing list