A Functional Link Network With Ordered Basis Functions

A procedure is presented for selecting and ordering the polynomial basis functions in the functional link net (FLN). This procedure, based upon a modified Gram Schmidt orthonormalization, eliminates linearly dependent and less useful basis functions at an early stage, reducing the possibility of combinatorial explosion. The number of passes through the training data is minimized through the use of correlations. A one-pass method is used for validation and network sizing. Function approximation and learning examples are presented. Results for the ordered FLN are compared with those for the FLN, group method of data handling, and multi-layer perceptron.