Comparison of Approximation Capabilities of Neural Networks and Linear Models
Abstrakt
Using method from nonlinear approximation and integral representations tailored to computational units, we describe some cases when neural networks outperform any linear approximator.