LTI ODE-valued neural networks

摘要: A dynamical version of the classical McCulloch & Pitts' neural model is introduced in this paper. In new approach, artificial neurons are characterized by: i) inputs form differentiable continuous-time signals, ii) linear time-invariant ordinary differential equations (LTI ODE) for connection weights, and iii) activation functions evaluated frequency domain. It will be shown that characterization constitutive nodes an network, namely LTI ODE-valued network ODEVNN), allows solving multiple problems at same time using a single structure. Moreover, it demonstrated ODEVNNs can interpreted as complex-valued networks (CVNNs). Hence, research on topic applied straightforward form. Standard boolean implemented to illustrate operation ODEVNNs. Concluding paper, several future lines highlighted, including need developing learning algorithms newly