Inference Bayesian network for multitopographic neural network communication: a case study in documentary data

This paper presents an original approach consisting in assimilating the behavior of the MultiSOM model to the one of a Bayesian inference network in documentary data. This approach is used both for validating the MultiSOM intermap communication principles and for enhancing the accuracy of the probabilistic correlation computation mode. In a complementary way, the approach also led to prove that a neural multimap model provided with unsupervised learning might well behave as a Bayesian inference network in which the estimation of posterior probabilities becomes a simple process only using prior similarity measures. A performance comparison between former probabilistic intercommunication mode and new probabilistic intercommunication mode based on Bayesian inference is finally proposed in the paper.