Random Vector Functional Link (RVFL) Networks

Abstract

This chapter summarises the concept of the random vector functional link network (RVFL), that is, a multilayer perceptron (MLP) in which only the output weights are chosen as adaptable parameters, while the remaining parameters are constrained to random values independently selected in advance. It is shown that an RVFL is an efficient universal approximator that avoids the curse of dimensionality. The proof is based on an integral representation of the function to be approximated and subsequent evaluation of the integral by the Monte-Carlo approach. This is compared with the universal approximation capability of a standard MLP. The chapter terminates with a simple experimental illustration of the concept on a toy problem.