In ELM, the inputweights
(linking the input layer to the hidden layer) and hidden biases are
randomly chosen, and the output weights (linking the hidden layer to
the output layer) are analytically determined by using the Moore–
Penrose (MP) generalized inverse.