X-Message-Number: 4338
From: Eli Brandt <>
Subject: NNs' capabilities
Date: Wed, 3 May 1995 02:32:39 -0700 (PDT)

Joseph Strout quoting the comp.ai.nn FAQ:
>    In principle, NNs can compute any computable function, i.e. they
>    can do everything a normal digital computer can do.

Feedforward networks, which are guaranteed to halt, cannot be as
powerful as a TM; they can calculate the recursive functions, but not
the recursively enumerable functions.

Recurrent networks are a different matter.  I suspect that that when
people talk about ideal recurrent networks, they're allowing them to
use arbitrary-precision arithmetic.  This involves unbounded memory
use, making them more than finite automata.

-- 
   Eli   


Rate This Message: http://www.cryonet.org/cgi-bin/rate.cgi?msg=4338