On different faces of the representer theorem, non-euclidean geometries and randomness
Vincent Guillemet

Meeting • 2022-07-12

Abstract
We will study indirect applications of the excellent "Splines Are Universal Solutions of Linear Inverse Problems with Generalized TV Regularization" (and other articles). Those will feature spatio-temporal temperature monitoring and epidemic (COVID19 - Malaria) monitoring. Our technical concern will be first on finding a representer theorem (RT) for spaces with non-euclidean geometries and performing regressions on functional spaces with multiple components (trend-seasonal decomposition). Then, our technical concern will move towards adapting the RT for regressions on spaces of stochastic processes, specifically for Hawkes processes. The presentation shall be decorated with fancy plots and videos.