PROPERTIES OF BLOCK FEEDBACK NEURAL NETWORKS

In this paper, we discuss some properties of Block Feedback Neural Net works (BFN). In the first part of the paper, we study network structur es. We define formally what a structure is, and then show that the set F-n of n-layers BFN structures can be expressed as the direct sum of the set A(n) of n-layers BFN architectures and the set D-n of n-layers BFN dimensions. Both A(n) and D-n are shown to have the structure of a distributive lattice and to indice such structure in F-n. Moreover, we show that the computing capabilities of BFN are monotonically nonde creasing with the elements of A(n) ordered according to the lattice st ructure. In the second part we show that the increasing in the computi ng power allows the BFN to be universal computers, having the same com puting power as a Turing machine.