Stability analysis of gradient-based neural networks for optimization problems (Q5942314): Difference between revisions

The authors consider for the unconstrained optimization problem \[ \min E(x), \quad x\in \mathbb{R}^n. \] A neural network model consists of the general gradient-based motion equation \[ {dx(t)\over dt}=Hg \bigl(x(t) \bigr), \quad x(t_0)= x_0\in \mathbb{R}^n \] where \(g(x)\) is the gradient of \(E(x)\), \(x_0\in \mathbb{R}^n\) is an arbitrary initial point and \(H\) is a symmetric positive definite matrix which can be viewed as a scaling matrix. The authors introduce a new approach to analyze the stability of neural network models without using any Lyapunov function.