Explaining Supervisor Set for Machine Learning Methods
Andrzej Nowakowski, Piotr Fulmański, Marta Lipnicka · 2024
The problem of explaining learning considered in this paper is based on Vapnik’s learning machine model and is that of choosing from the given set of functions the one that best approximates the supervisor’s response. We define that family in the rigorous way using a set of partial differential equations. Such a formulation of the family of functions together with a risk functional allows to apply tools from optimal control theory. We build a new dynamic programming method to derive verification theorem for approximate solution to that new control problem. That approach to machine learning enables a construction of a numerical algorithm to compute a function which approximates supervisor set.