Author
Abstract
Abstract –Adaptive systems include a vast range of living natural and artificial
systems. Reinforcement learning systems are one form of adaptive systems. The current
work will focus on a particular kind of reinforcement learning system: the classifier
system. A classifier system has the ability to categorize its environment and create rules
dynamically, thus making it able to adapt to differing circumstances. This work
investigates the effect of crowding factor on the classifier system to solve six-bit and
eleven-bit multiplexer problems. The six bit multiplexer problem is defined as six signal
lines that come into the multiplexer. The signals on the first two lines (the address or Alines)
are decoded as an assigned binary number. This address value is then used to
indicate which of the four remaining signals (on the data or D-lines) is to be passed
through the multiplexer output. The eleven bit multiplexer problem is defined as eleven
signal lines that come into the multiplexer. The signals on the first three lines (the
address or A-lines) are decoded as an assigned binary number. This address value is
then used to indicate which of the eight remaining signals (on the data or D-lines) is to
be passed through the multiplexer output. This work Investigates the classifier system
rule learning with no crowding and normal crowding settings by comparing and
contrasting the effectiveness of the rule sets learned and their composition in two cases.
Experiment results show that the run using classifiers without crowding replacement is
unable to perform as well as the run with crowding replacement. The time needed to
match the signal is shorter when using classifiers with crowding replacement and we are
more likely to achieve good results quickly.
Keywords
)