Xref: utzoo comp.ai:9102 can.ai:144
Path: utzoo!unixhub!linac!pacific.mps.ohio-state.edu!zaphod.mps.ohio-state.edu!swrinde!cs.utexas.edu!rice!uw-beaver!cutthroat.cs.washington.edu!mikew
From: mikew@cutthroat.cs.washington.edu (Mike Williamson)
Newsgroups: comp.ai.neural-net,comp.ai,can.ai
Subject: Re: Adaptive Logic Networks
Message-ID: <1991Apr24.033222.16343@beaver.cs.washington.edu>
Date: 24 Apr 91 03:32:22 GMT
References: <1991Apr23.211210.29372@cs.UAlberta.CA>
Sender: news@beaver.cs.washington.edu (USENET News System)
Reply-To: mikew@cs.washington.edu (Mike Williamson)
Organization: Computer Science & Engineering, U. of Washington, Seattle
Lines: 18

In article <1991Apr23.211210.29372@cs.UAlberta.CA> dwelly@saddle-lk.cs.UAlberta.CA (Andrew Dwelly) writes:
>The algorithm is a radical departure from normal neural-net techniques
>because it is based on the synthesis of an Adaptive Logic Network (ALN).
>In the process of learning, the system constructs a boolean digital circuit
>to perform the task. Training is rapid, and the execution of the resulting
>network is extremely fast (and could easily be turned into hardware for
>even more speed).

Wait.  If the algorithm constructs a boolean digital circuit, how is
it related to neural nets?  Many traditional inductive learning
algorithms produce a "concept", which is often expressed as, e.g., a
restricted conjunctive-normal-form boolean expression.  No real trick
to make a circuit from that.

No wonder training is rapid, if you have departed so radically from
neural nets as to have a standard inductive learning algorithm.

-Mike