Deconvolution in a network

Abstract

We analyse linear models between functional variables. We first recall the main results about unconstrained models estimated by penalized OLS but we will concentrate the presentation on convolution models defined as constrained specifications. In the network example convolution models means that the relation between variables only depends on a “distance” between individuals. The main objective is the construction of this concept in order to perform a Fourier analysis on general spaces (graphs, homogenous spaces for examples).