Dynamical Nonlinear Neural Networks with Perturbations Modeling and Global Robust Stability Analysis

Abstract

This paper is devoted to studying both the global and local stability of dynamical neural networks. In particular, it has focused on nonlinear neural networks with perturbation. Properties relating to asymptotic and exponential stability and instability have been detailed. This paper also looks at the robustness of neural networks to perturbations and examines if the related properties have been preserved. Circumstances for global and local exponential stability of nonlinear neural network dynamics have been studied. We mentioned that the local exponential stability of any equilibrium point of dynamical neural networks is equivalent to the stability of the linearized system around that equilibrium point. From this, some well-known and new sufficient conditions for local exponential stability of neural networks have been obtained. The Lyapunov's procedure has been used to analyze the stability property of nonlinear dynamical systems and many outcomes have been combined and generalized. A kind of Lyapunov's stability of the stable points of Hopfield neural network (HNN) have been proven, which means that if the initial state of the network is close enough to a stable point, then the network state will remain in a small neighborhood of the stable point. These stability results indicate the convergence of the memory process of HNN. The theoretical results are illustrated through a few problem cases for a nonlinear dynamical system with perturbation behavior.

An Overview of Neuromorphic Computing for Artificial Intelligence Enabled Hardware-Based Hopfield Neural Network

“…Hopfield is based on the idea of energy function to create a new calculation method, which is through the nonlinear dynamics method for developing this neural network. It has clarified the relationship between the neural network and dynamics model [61]. Then, established the stability criterion of the neural network on this algorithm.…”

Hardware-Based Hopfield Neuromorphic Computing for Fall Detection

“…The node states of the network take the binarized +1 and −1. Hopfield neural network is derived from a nonlinear dynamical system, and DHNN can be described by a set of nonlinear difference equations [ 24 ], while differential equations usually describe to the CHNN [ 25 ]. Compared to other machine-learning algorithms, the Hopfield neural network is more straightforward and less dependent on the data.…”