ELM learning theories show that effective learning algorithms can be derived based on randomly generated hidden neurons (biological neurons, artificial neurons, wavelets, Fourier series,etc) as long as they are nonlinear piecewise continuous, independent of training data and application environments.