Path: utzoo!attcan!uunet!mcsun!ukc!cam-eng!ajr
From: ajr@eng.cam.ac.uk (Tony Robinson)
Newsgroups: comp.ai.neural-nets
Subject: Re: d
Summary: T800 is old, parallel is good, I'd like a comp.speech
Message-ID: <15881@rasp.eng.cam.ac.uk>
Date: 24 Oct 90 18:57:46 GMT
References: <9279.2724bc2c@ul.ie>
Sender: ajr@eng.cam.ac.uk
Organization: Cambridge University Engineering Department, UK
Lines: 26

In article <9279.2724bc2c@ul.ie> oreillyj@ul.ie writes:
>* has anybody tried implementing neural networks using INMOS transputers?
>  I haven't started getting into it and am wondering whether it is worth
>  the effort.

I use an array of 64 transputers, each one gives me about 0.8 MFLOPS so you
need a few of them to get anywhere, and then you have lots of fun dealing
with the parallel programming problems.  My opinion is that the T800
transputer technology is now outdated, but parallel machines are good for
training artificial neural networks if you have the time and patience to do
the parallel programming.

>* are there many people out there working on neural network based speech
>  recognition?  if there is, I think there should really be another
>  newgroup set up specifically for this subject.

Speech recognition is one of the big application areas, however, I don't
think that there is enough traffic for a comp.ai.neural-nets.speech.  I made
a tentative suggestion in comp.dsp and news.groups a while back about a
comp.speech/comp.ai.speech newsgroup, which was to cover all aspects of
speech processing (e.g. coding, synthesis and recognition).  Only a few
people replied so I gave up the idea.  Maybe there is more interest from the
comp.ai.neural-nets readership?


Tony Robinson