CONVERGENCE OF ITERATIVE ALGORITHMS FOR LEARNING BAYESIAN NETWORKS

Resumo

The formalism of the bayesian networks has been employed in the development of many intelligent systems. This work considers applicatins which demand the utilization of online learning methods, more specifically methods for online parameter learning. Online learning methods update the bayesian network parameters as new data samples/observations are collected. In this context, it is import to consider the convergence of the learning method in relation to the empirical distribution of the data. Given that this work  proposes a experimental protocol to quantify the convergence/divergence of models generated by online learning procedures. An application example it is also presented.