Stability of discrete time recurrent neural networks and nonlinear optimization problems

Area of Science:

• * Control Theory
• * Machine Learning
• * Nonlinear Systems Analysis

Background:

• * Discrete-time recurrent neural networks (DTRNNs) are essential nonlinear systems.
• * Standard absolute stability criteria often yield insufficient results for DTRNNs.
• * Proving global Lyapunov stability is crucial for understanding DTRNN behavior.

Purpose of the Study:

• * To introduce and validate the Reduction of Dissipativity Domain (RDD) method for DTRNN stability analysis.
• * To demonstrate the superiority of RDD over traditional absolute stability criteria.
• * To establish conditions for effective application of the RDD method.

Main Methods:

• * The study employs the Reduction of Dissipativity Domain (RDD) method.
• * RDD involves a multi-step procedure requiring maximization of nonconvex functions over polytopes.
• * Analysis focuses on conditions guaranteeing at most one local maximum for these functions.

Main Results:

• * The Reduction of Dissipativity Domain method is shown to be more powerful than standard criteria.
• * Conditions are derived ensuring at most one local maximum for the objective functions.
• * This result holds for a broad spectrum of neuron transfer functions.

Conclusions:

• * The Reduction of Dissipativity Domain offers a more effective approach for proving global Lyapunov stability in DTRNNs.
• * The derived conditions facilitate the practical application of this advanced stability analysis technique.
• * The method's applicability across diverse neuron transfer functions enhances its utility.