[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: Rocket Science



PureBytes Links

Trading Reference Links

One of my favorite stories about NNs comes from when the military was trying
to develop a system to recognize tanks from photographs.  They went out and
took lots of pictures of scenery with tanks and scenery without tanks and
went back to the lab and trained the neural nets.  When they finished
training them, they took some more pictures of scenery with and without
tanks and ran them thru the system and it was a total failure.  They puzzled
over the training and testing images a while before they finally discovered
the problem: in the training images, all of the images without tanks had
shadows and all of the images with tanks did not.  So the system they had
trained was looking for images with shadows, not tanks.  I guess shadows
were easier for the neural net to lock on to.

Kent


-----Original Message-----
From: Marlowe Cassetti <marlowec@xxxxxxx>
To: IndTrader2@xxxxxxx <IndTrader2@xxxxxxx>; omega-list@xxxxxxxxxx
<omega-list@xxxxxxxxxx>
Date: Thursday, February 03, 2000 3:22 PM
Subject: Re: Rocket Science


No first hand knowledge on the French SW but on the topic of fuzzy logic
there are a lot of practical applications of FL out there.  Most video
cameras have FL anti-shake control logic.  FL has been embedded in the
control systems of many industrial systems such as HVAC, cement reduction
furnaces, etc.  For those interested there are many books and magazine
articles on this topic.

FL is a tool that has merit in the decision making as well as control
systems.  I routinely use a Fuzzy Logic decision method to evaluate trading
results.  I also developed a FL trading system for SuperCharts.  It gives
good entry points but the exits give back too much.  It needs more work and
is far from tradable.

Neural networks have been successfully used for speech recognition, cancer
cell detection, and other pattern recognition applications.  Again there is
a lot of literature out there on the topic.  Probably the "killer app" for
NN is in intelligent robotics.

I suspect that in another generation, FL and NN will be routinely accepted
by the general populace.

----- Original Message -----
From: <IndTrader2@xxxxxxx>
To: <omega-list@xxxxxxxxxx>
Sent: Thursday, February 03, 2000 7:08 AM
Subject: Re: Rocket Science


> I'll bet the conclusion is that most of those approaches can NOT beat the
> > market.  The moving average folks have nothing to fear from the rocket
> > scientists.  Speaking of rockets - What ever became of that last one
that
> > went to Mars?  Ooops!  Must have been using Chaos theory or was it
neural
> > nets.
>
> No, it was powered by a french fuzzied software
>
> Has ANYONE purchased said french fuzzied software...........they keep
raising
> the price (interesting sales tactic) yet, product seems to have no
followers.
> I do know a guy in the military who uses fuzzies for tank (image)
recognition
> who showed me some results......amazing. I guess it is easier to find
tanks
> than market cycles.
>
> brent
>