Formally specifying neural networks via MOFEU and equivalence of well-formed neural networks to normal dynamic systems

摘要

This work presents an overview of the formal model of universal specification for neural networks, known as MOFEU (Modelo Formal de Especificacao Universal para Redes Neurais). The model is represented by a mathematical structure that allows for specification at the levels of neuron, architecture, learning, and operational mode. Such a model is not only useful for universal specification, but has the highly important characteristic that the paradigms of neural networks specified in this model are equivalent to a normal dynamic system.