Adaptive Computation of Sparse Solutions
Speaker: Prof. Dr. Martin Burger
Affiliation: FAU, Germany
Zoom: Meeting ID: 618 3161 7110 , Passcode: 442549
Abstract: In this talk we will discuss approaches of adaptivity to the computation of sparse solutions in an underlying continuum setting. A canonical application of interest and the guiding is the deconvolution problem in superresolution microscopy, but there are several other relevant applications, e.g. the efficient design of imaging techniques for exploration robots or the training of compact neural networks. Our approach is based on convexified variational formulations of sparsity, which allows for a rigorous mathematical analysis. A key idea we discuss is the adaptive refinement on grids based on a-posteriori error estimation, a second one the superresolved estimation of peak locations from multiple nonzero entries in adjacent grid points.