In practice, the process of learning is considered as follows. A learning machine ust choose from a given set of functions F = {f(x,α), α∈} the one which best (in

ome predefined sense) approximates the unknown dependency. is an abstract set f parameters, chosen beforehand. This choice is actually an optimization problem in he parameter space of α. For the sake of convenience in general discussions, the set f parameters of the machine is denoted with a vector α.