Estimation de signaux par noyaux d'ondelettes
Vincent Guigue, Alain Rakotomamonjy and Stéphane Canu
Revue Traitement du Signal
This paper addresses the problem of regression in the case of non-uniform sampled signals.
Our method is based on supervised learning theory, we propose to use
L2 estimation with wavelet
kernel combined with
L1 multiscale regularization. The use of Least Angle Regression as solver
enable us to propose new solutions to set the regularization parameter.