Random initialization

Definition

Random initialization is an approach used in iterative learning algorithms where the initial parameter vector is selected with its coordinates chosen randomly. Random initialization works better than deterministic processes if we specifically want all coordinates of the parameter vector to be different so that they are capable of learning different features. This is particularly important in learning algorithms associated with neural networks.