Radial basis functions are a form of kernal famlies of functions, often non-linear, which are used as early layers in a neural network to create non-linear diversity, used especially for image processing.. The functions are be applied at each point in the input data and include members with diffeeent levels of 'spread' to enable represtentions at different scales. Support vector machines use an initial layer like this followed by a relatively simple upper layer.
Used on Chap. 8: page 155