Game-based varying parameters in zeroing neural networks

This paper studies new game-based varying parameters (GBVP) in zeroing neural networks (ZNN). Models with such parameters are termed GBVP-ZNN. To the best of our knowledge, this is the first introduction of game-based varying parameters into ZNNs. Meanwhile, GBVP-ZNN can reduce to many ZNN models found in previous literature. First, the solvability of the expected parameters is theoretically analyzed. This is equivalent to the existence of Nash equilibria in a specific differential game, for which several sufficient conditions are derived. Second, the convergence and finite-time convergence of GBVP-ZNN are investigated. The error function converges to zero under certain conditions. Third, the discretization process of GBVP-ZNN is presented. Finally, an application to time-varying matrix inversion is investigated.