You also need to know Probability.
There is a research paper which is about the function approximation of RBF network and RBF network with node fault.
Given an trainning set, find out the weight vector, fed the training set back, then got the training set error
Fed the test set in the trained networks, test set error can be found
Using the fault vector multiplying the weight , and get the faulty weight vector, for each fault network, when ptrue > 0.005, generate 10,000 faulty network
When ptrue < 0.005, generate 50,000 faulty network. Fed the test set to find out the test set error of each faulty network and find out the average of the test set error of each trained network. use the data to plot the graph in page 19 of test set error. Given an trainning set, find out the weight vector, fed the training set back, then got the training set error