The approximation capabilities of Durrmeyer-type neural network operators (Q6989222)

In this paper, the authors introduce a new family of neural network operators of Durrmeyer type. These operators originate from sigmoidal activation functions, with coefficients defined through a convolution integral involving a kernel function and the target function.\N\NThe theoretical analysis establishes pointwise and uniform convergence for continuous functions, together with quantitative estimates for the approximation order based on the modulus of continuity. The study is further extended to Lebesgue spaces \(L^p([a,b])\), where convergence theorems are proved by exploiting norm inequalities and the density of continuous functions in these spaces.\N\NFinally, several examples of activation functions, including also ReLU and RePU functions, are discussed.