Multilayer feedforward networks are universal approximators

  title={Multilayer feedforward networks are universal approximators},
  author={Kurt Hornik and Maxwell B. Stinchcombe and Halbert L. White},
  journal={Neural Networks},
Neural networks for localized approximation
We prove that feedforward artificial neural networks with a single hidden layer and an ideal sigmoidal response function cannot provide localized approximation in a Euclidean space of dimension
A constructive method for multivariate function approximation by multilayer perceptrons
It is shown how to construct a perceptron with two hidden layers for multivariate function approximation, which can perform function approximation in the same manner as networks based on Gaussian potential functions, by linear combination of local functions.
Constructive Approximation of Discontinuous Functions by Neural Networks
A constructive proof that a real, piecewise continuous function can be almost uniformly approximated by single hidden-layer feedforward neural networks (SLFNNs) is given.
Function approximation using a partition of the input space
  • P. Koiran
  • Computer Science, Mathematics
    [Proceedings 1992] IJCNN International Joint Conference on Neural Networks
  • 1992
It is shown that a simple geometric proof of this theorem can be extended to networks of units using a smooth output function, and a recent result on the approximation of polynomials by fixed size networks is improved.
Generalization and Approximation Capabilities of Multilayer Networks
The theory allows one to specify a finite discrete set of training data and a network structure that generalizes and approximates any given continuous mapping between sets of contours on a plane within any given permissible error.
We prove that neural networks with a single hidden layer are capable of providing an optimal order of approximation for functions assumed to possess a given number of derivatives, if the activation


Universal approximation using feedforward networks with non-sigmoid hidden layer activation functions
Multilayer feedforward networks possess universal approximation capabilities by virtue of the presence of intermediate layers with sufficiently many parallel processors; the properties of the intermediate-layer activation function are not so crucial.
There exists a neural network that does not make avoidable mistakes
  • A. Gallant, H. White
  • Computer Science
    IEEE 1988 International Conference on Neural Networks
  • 1988
The authors show that a multiple-input, single-output, single-hidden-layer feedforward network with (known) hardwired connections from input to hidden layer, monotone squashing at the hidden layer
Approximation by superpositions of a sigmoidal function
  • G. Cybenko
  • Computer Science
    Math. Control. Signals Syst.
  • 1989
In this paper we demonstrate that finite linear combinations of compositions of a fixed, univariate function and a set of affine functionals can uniformly approximate any continuous function ofn real
Nonlinear dynamics of artificial neural systems
It is shown that there are many powerful techniques for reducing the number of spurious terms, and that the high order approach has many advantages over a cascaded slab approach in certain problem areas.
Parallel distributed processing: explorations in the microstructure of cognition, vol. 1: foundations
The fundamental principles, basic mechanisms, and formal analyses involved in the development of parallel distributed processing (PDP) systems are presented in individual chapters contributed by
On the Representation of Continuous Functions of Several Variables as Superpositions of Continuous Functions of one Variable and Addition
The aim of this paper is to present a brief proof of the following theorem: Theorem. For any integer n ≥ 2 there are continuous real functions ψ p q (x) on the closed unit interval E 1 = [0;1] such
Principles of mathematical analysis
Chapter 1: The Real and Complex Number Systems Introduction Ordered Sets Fields The Real Field The Extended Real Number System The Complex Field Euclidean Spaces Appendix Exercises Chapter 2: Basic
Theory of the Back Propagation Neural Network