Complexity issues in discrete neurocomputing

Abstract

An overview of the basic results in complexity theory of discrete neural computations is presented. Especially, the computational power and efficiency of single neurons, neural circuits, symmetric neural networks (Hopfield model), and of Boltzmann machines is investigated and characterized. Corresponding intractability results are mentioned as well. The evidence is presented why discrete neural networks (inclusively Boltzmann machines) are not to be expected to solve intractable problems more efficiently than other conventional models of computing.