Complexity Of Neural Network Learning In The Real Number Model

This paper presents several results concerning complezity of neural network learning in the Blum-ShubSmale real number model. Two different types of loading problems are defined in this model. The relationships between loading problems and some problems in the real number model are studied. First, we prove the polynomial equivalence between the second loading problem and linear progmmming problem in the real number model. Second, we show that the TSP problem can be polynomially reduced t o the first loading problem. At last, we prove loading a neural network consisting of linear sum units and linear threshold units is NP-complete.