*** WHAT COULD I DISCOVER ? ***
Ok, you will not go into history as the discoverer of the
first Artificial Intelligence (AI) of some new kind. What
you - or actually your computer - could discover are more
optimal settings for a type of neural network as described
in the paper "Mismatch and the Construction of Internal
Representations".
*** HOW COULD I WIN $1000 ? ***
All you need is CPU time and a java enabled browser. A java
applet will vary settings to find a more optimal learning
neural network. It will have to find a network that out of
ten cycles of 30.000 time steps gets six times (or more)
below the 'magical' benchmark. This benchmark is the point
at which the neural network creates predictions without
receiving any input at certain times. This is the feature
that sets this network apart from other approaches.
However, a fluke doesn't count. You (and I) must be able to
enter the found settings and reproduce 6 hits per 10 cycles!
And note that only the first email will be rewarded! Apart
from this, there's no catch, though you could be wasting
your time :-(
*** How do I participate? ***
Simply visit the link at the bottom of this introduction
which will take you to the java applet and more
information. Simply click the GO button on the applet and
see if it finds setting that meet the demands stated above.
If these turn out to be reproducible, mail the settings and
your full address to mervyn at xs4all.nl
o Note that you can keep the java applet running off-line!
o An eventual solution will be posted on comp.ai and
comp.ai.neural-nets
Kind regards,
Mervyn van Kuyen
============================
mervyn at xs4all.nlAI at Home:
http://www.xs4all.nl/~mervyn
============================
P.S. In case you want to experiment with
the settings yourself:
o trials - how many different settings will be tried
o agent - how many new neurons are created after each
evaluation
o kill - 45 means 45% chance that new connections which
do not benefit performance will be removed
o balance - 50 means 50% of the new connections will be
excitatory, the rest inhibitory
o exi - 20 means the maximum weight of the excitatory
neurons will be 20 (threshold is 10)
o inh - 20 means the maximum weight of the inhibitory
neurons will be 20 (threshold is 10)
