How NOT to choose a neural network

Hi,

A neural network is a calculator, with an extra function. To
determine how ridiculous (or reasonable) a given statement about neural
nets is, replace 'neural' by 'multiplication' and see how it sounds.

"Everyone knows multiplication doesn't work."

"Everyone knows multiplication works"

"We offer multiplication solutions"

"Why doesn't your calculator have buy/sell signals next to the
multiplication button?"

"Our predictions are based on pure multiplication"

----------

Or, think about how ridiculous it would be that a website offers
***automatic*** stock
predictions based only on multiplication.

-----------

Or, another claiming that multiplication can predict the lottery.

-----------

It is clear,

*if a website claims their neural network can predict the lottery, OR
*if there is no place for any user input, OR
* if it generates automatic buy/sell signals,

SOMETHING IS OBVIOUSLY VERY WRONG WITH THAT WEBSITE.

-----------

There are admittedly issues associated with adjusting and getting a
neural calculation working in one setting or another, having to do with
scaling, normalization, metrics, etc.
But apart from this, there is really one standard calculation, and it
is a calculation which a user makes for a particular purpose of his
own.

If a website makes claims or statements that are obviously false, one
has to begin to question where they found a programmer, what steps they
took to verify their program is working correctly, and what is the
likelihood they have implemented the program correctly with respect to
these subsidiary -- but necessary --- issues.


john moody
www.goldengem.co.uk

.

Quantcast