Classifiers based on Bayesian neural networks
Artem A. Maksutov, Anastasiya V. Simonenko, Igor S. Shmakov · 2017
Artificial neural networks have occupied significant niche in IT world, but the general concept still has some unresolved issues. This paper is devoted to Bayesian network, which is the probabilistic model organized in acyclic graph. Providing brief introduction in the world of artificial neural networks and Bayesian approach in particular we then move to the proof of the idea that Bayesian networks do not suffer from retraining, which is the real problem for industrial usage of neural networks. This assumption is proved by an application to the real problem of classification. Our goal is to build Bayesian network which will classify the investigated objects and show that this network takes into account the noise and doesn't suffer from retraining.