TY - JOUR
T1 - Adapting Bayes Network Structures to Non-stationary Domains
AU - Nielsen, Søren Holbech
AU - Nielsen, Thomas Dyhre
PY - 2008
Y1 - 2008
N2 - When an incremental structural learning method graduallymodifies a Bayesian network (BN) structure to fit a sequential streamof observations, we call the process structuraladaptation. Structural adaptation is useful when the learner is set towork in an unknown environment, where a BN is graduallybeing constructed as observations of the environment are made. Existingalgorithms for incremental learning assume that the samples in thedatabase have been drawn from a single underlying distribution. Inthis paper we relax this assumption, so that the underlyingdistribution can change during the sampling of the database. Theproposed method can thus be used in unknown environments, whereit is not even known whether the dynamics of the environment arestable. We state formal correctness results for our method,and demonstrate its feasibility experimentally
AB - When an incremental structural learning method graduallymodifies a Bayesian network (BN) structure to fit a sequential streamof observations, we call the process structuraladaptation. Structural adaptation is useful when the learner is set towork in an unknown environment, where a BN is graduallybeing constructed as observations of the environment are made. Existingalgorithms for incremental learning assume that the samples in thedatabase have been drawn from a single underlying distribution. Inthis paper we relax this assumption, so that the underlyingdistribution can change during the sampling of the database. Theproposed method can thus be used in unknown environments, whereit is not even known whether the dynamics of the environment arestable. We state formal correctness results for our method,and demonstrate its feasibility experimentally
KW - Bayesian networks
KW - Learning
KW - Adaption
KW - Non-stationary domains
U2 - doi:10.1016/j.ijar.2008.02.007
DO - doi:10.1016/j.ijar.2008.02.007
M3 - Journal article
SN - 0888-613X
VL - 49
SP - 379
EP - 397
JO - International Journal of Approximate Reasoning
JF - International Journal of Approximate Reasoning
IS - 2
ER -