Line-search aided non-negative least-square learning for random neural network

Abstract

Recently, Timotheou has formulated the learning problem of the random neural network (RNN) into a convex non-negative least-square problem that can be solved to optimality. By incorporating this work of problem formulation and the line-search technique, this paper designs a line-search aided non-negative leastsquare (LNNLS) learning algorithm for the RNN, which is able to find a nearly optimal solution efficiently. (The source code is available at www.yonghuayin.icoc. cc.) Numerical experiments based on datasets with different dimensions have been conducted to demonstrate the efficacy of the LNNLS learning algorithm.