The problem of the necessary complexity of neural networks is of interest in applications. In this paper, learning capability and storage capacity of feedforward neural networks are considered. We markedly improve the recent results by introducing neural-network modularity logically. This paper r…