| Abstract | Learning Bayesian networks is a central prob-lem for pattern recognition, density estima-
tion and classification. In this paper, we
propose a new method for speeding up the
computational process of learning Bayesian
network structure. This approach uses con-
straints imposed by the statistics already col-
lected from the data to guide the learning al-
gorithm. This allows us to reduce the num-
ber of statistics collected during learning and
thus speed up the learning time. We show
that our method is capable of learning struc-
ture from data more efficiently than tradi-
tional approaches. Our technique is of partic-
ular importance when the size of the datasets
is large or when learning from incomplete
data. The basic technique that we introduce
is general and can be used to improve learn-
ing performance in many settings where suf-
ficient statistics must be computed. In ad-
dition, our technique may be useful for al-
ternate search strategies such as branch and
bound algorithms.
|
