Latitudinal and longitudinal neural network structures for function approximations

摘要: This paper proposes a novel neural network structure for approximating arbitrary functions. The convergence property of this kind is given. Another aim devoted to constructing networks approximate continuous functions by local piecewise quadratic It shown that such constructive can be applied any function with sufficiently small error. Compared existing works, provides the strategy use much fewer neurons achieve desired precision, and given more favorable smoothness using nonlinear function. relationship between two structures mentioned above discussed.