Re: Re: 1-Wire network wiring

["Rob Iles" <rob.iles@xxxxxxxxx>]

or the neural net designed to "tag" russian tanks with
mines/explosives of
somesort - - proven very reliable in test - - but when let loose on the
field, went rampant on the US tanks! .......if memory serves (and if this
is
a recollection of a bad film plot, I apologise in advance!) .......all the
pictures of Russian tanks had been taken in wet/damp/dark conditions - -
all
the US tanks were photographed/filmed in sunlight. .....Whilst it seemed
the
system had learned to accurately distinguish russian from american tanks,
what it had actually learned was light vs dark (or wet vs dry) ......and
when presented with a field full of tanks on a wet/dark day - it tagged the
lot!!!

I either need lots more coffee, or lots of sleep - can't decide which!!

R.


On 10/26/07, Mal Lansell <mal@xxxxxxx> wrote:
>
>   Neural nets can be very difficult to set up and train, because it's
> hard to tell if it is learning the right thing.
>
> A classic example is a camera watching a train platform to see if it
> is getting too crowded. It seemed to be working but would fail
> occasionally - it turned out it had "learned" not whether
the
> platform was full, but whether it could see a bin in the corner
> (which was always blocked from view by crowds, but was also very
> occasionally blocked by a lone passenger).
>
> Mal
>
> --- In ukha_d@xxxxxxx <ukha_d%40yahoogroups.com>, "Chris
Hunter"
> <cjhunter@...> wrote:
> >
> > The obvious way to (try to) get it to work is using neural net
> software (using it to recognise pattern, programming it via a
> teaching & learning process, through giving it lots of examples)
...
> using other sensors as well (micropones, PIRs, etc) ... wonder if
> anyone has tried something like this ?
> >
> > Chris
> >
> > ('E61 / T-Mobile W'n'W)
> >
> >
> > - original message -
> > Subject: [ukha_d] Re: 1-Wire network wiring
> > From: "JonS" <semi.jon@...>
> > Date: 26/10/2007 10:27 am
> >
> > --- In ukha_d@xxxxxxx <ukha_d%40yahoogroups.com>, Chris
Hunter
> <cjhunter@> wrote:
> > >
> > > sorry, I was really just exploring the idea of lots of temp'
> > > sensors ... . Did you see Nigel's post ... interesting ! ?
> > >
> >
> > Yep, and I agree interesting things can be gleaned by
observation,
> the
> > open door to the kitchen correlates well with a dip in kitchen
> temp
> > here. But if you have door sensors and relays on the lights the
> temp
> > data is interesting but I think very hard to use as an indicator
> of
> > other information in an automated fashion. Not wishing to put you
> off
> > if you can see a way to make it work but it seems a hard way of
> going
> > about it, but then I have not played with the fuzzy logic module
> in
> > Cortex - still getting to grips with the basics.
> >
> > Jon
> >
>
>
>


[Non-text portions of this message have been removed]