From neural networks to graphical models: a brief introduction

摘要: One of the most important and interesting aspects neural networks is that they exhibit collective information-processing capabilities by connecting processing elements (“neurons”), each which performs a very simple limited information processing. The central issue network research thus to explore how why such system with exhibits complex functionalities. Being looked at from another direction, studying problem, we are asking about global characterization performed network, on basis local (i.e., neuron), as well interconnected. Essentially same problem also arises in study graphical models (Bayesian networks). main objective this section give brief review models, some emphasis their relation networks. In one class an N -dimensional random vector x characterized set conditional probabilities form: p(xi|x\i), where x\i denotes all except xi. If, for i, xi independent xj , j > then defines probability distribution