|
What is
JellyFish?
|
|
JellyFish is a neural net based backgammon
program that plays at a very high level. On the highest playing
level it matches the best humans in the world, and on the very
fast level 5 a top human will hardly win more than 55% of the
time. Also, its use of the doubling cube is outstanding.
JellyFish is able to play matches of any length, or 'money games'
where each point is equally valuable.
JellyFish uses artificial neural networks,
trained to play backgammon from self play. An artificial neural
network is a model of how the human brain works on the level of
individual brain cells. Fredrik Dahl developed the neural nets of
JellyFish, and used the neural net compiler nn from Neureka
ANS.
This program can be used for fun, for
testing your game, for analyzing recorded matches, or most
importantly: To improve your game.
JellyFish can give a running commentary on
the moves and cube decisions you make. You may play against
JellyFish, or use the "2 Players" mode to have
JellyFish keep track of the score and comment on both opponents
play. It's almost like having your own private professional to
comment on your game.
Why did we call it "JellyFish"?
The name JellyFish comes from the neural
net that the program is build over. A neural net is a simulation
of a biologic brain, and like a human, JellyFish has learned the
strategy of backgammon by experience, by playing itself a very
large number of games.
The human brain, however, contains more
than 100.000.000.000 neurons, and this is not possible to include
into a computer program. The size of our programs
"brain" is more the size of some of the lower organism
- like a jellyfish.
But - our JellyFish is hopefully a better
backgammon player than the jellyfish you find in the sea...
System requirements:
- JellyFish requires a PC running
Windows95 or Windows NT
- 8 MB RAM
- i486 processor (Pentium recommended
for maximum benefit)
- 2 MB Disk space
- CD-ROM Player
Copyright © 1998 - 2002
JellyFish AS. All rights reserved. Last updated Comments to info@jelly.effect.no