This paper presents an algorithmic paradigm that will address a malware model in the form of a nonlinear set of differential equations.
The neural networks are Morlet wavelet as estimate the state variables of the malware behavior.
To maximize the network performance, we use a heuristic optimization algorithm that optimizes the network parameters to achieve rapid convergence and accurate performance.
In both, the mean squared errors (MSE) of between \(\:{10}^{-09}\) and \(\:{10}^{-07}\), which point to the accuracy of the scheme.
The stability of optimization process is checked by the comparative behavior with the several runs.
This paper presents an algorithmic paradigm that will address a malware model in the form of a nonlinear set of differential equations. The neural networks are Morlet wavelet as estimate the state variables of the malware behavior. To maximize the network performance, we use a heuristic optimization algorithm that optimizes the network parameters to achieve rapid convergence and accurate performance. The learning stage is designed and constructed on the basis of a fitness criterion which is based on the residuals of the governing equations and the initial conditions. We discuss two various cases in order to test the strength of proposed approach. In both, the mean squared errors (MSE) of between \(\:{10}^{-09}\) and \(\:{10}^{-07}\), which point to the accuracy of the scheme. Furthermore, the values of Theil inequality coefficient (TIC) are between \(\:{10}^{-06}\) and \(\:{10}^{-04}\) with strong predictive powers. The stability of optimization process is checked by the comparative behavior with the several runs. We find the results in perfect agreement with reference solutions, which suggests that our framework is an effective and useful substitute to traditional numerical solvers to malware dynamic systems.
