We have begun work on creating a classifier whose
inspiration is drawn from Boolean function minimization.
The classifier called Circle possesses a number
of attractive properties, e.g., non-monotonicity. On
the initial synthetic datasets we examined, Circle performed
quite well. We are also encouraged by preliminary
success on the bioinformatics data that drove all
of this in the first place. A number of serious limitations
in terms of complexity still need to be addressed.
We have opted for an easy first-solution, probabilistically
employing Circle. There are, no doubt, other
techniques that can be adopted, by smartly guiding
Circle’s classification through the use of information
theory. There is a lot work to do on the rule sets themselves,
perhaps mining them or adding support and
confidence as is done with association rules. We are
also interested in applying Circle to streams, where
work has begun in developing classifiers and clustering.
Circle can likely be changed slightly to cluster.
One can imagine agglomeration based on logical adjacency.
We would like to thank Sun Kim for helpful
comments.