Adaptive basis functions. Alternatively, we might make a function y(x) from basis functions that depend on additional parameters included in the vector w. In a two-layer feed forward neural network with nonlinear hidden units and a linear output, the function can be written
