Discretization Error Analysis for Tikhonov Regularization
in Learning Theory
Ernesto De Vito, Andrea Caponetto and Lorenzo Rosasco
We study the connections between learning from examples and inverse
problems. We show that learning from
examples can be seen as the discretization of a stochastic
inverse problem defined by a Carleman operator.
In particular we develop a discretization strategy for this class of
inverse problems and we give a
convergence analysis. Our approach can be applied to other classes
of problems such as integral equations.