Correctness of the FPNA neural paradigm
Abstract
Neural networks are usually considered as naturally parallel computing models. But the number of operators and the complex connection graphs of standard neural models can not be handled by digital hardware devices. A new theoretical and practical framework allows to reconcile simple hardware topologies with complex neural architectures: Field Programmable Neural Arrays (FPNA) lead to powerful neural architectures that are easy to map onto digital hardware, thanks to a simplified topology and an original data exchange scheme. This report describes the basic principles of the FPNA paradigm. Formal definitions are introduced and illustrated. Two computation methods for feedforward FPNAs are introduced. The proof of their correctness is sketched.