Structured statistical and computational tools in high-dimensional recovery, with applications in medical imaging

The researchers plan to develop information criteria for sparse, structured selection and estimation of high-dimensional, linear models (such as in inverse problems). The new criteria will take the sparsity into account, especially by paying attention to bias that results from the effect of false positives on the optimization. By linking concepts of structured sparse and low-rank recovery to convex optimization, practical and efficient techniques for the regularization and solution of large scale inverse problems will be developed.

They will apply these methods to the domain of mathematical imaging, in particular to medical image reconstruction and magneto-encephalography. They will also develop new multiscale, data-adaptive transformations for sparse representations of signals, using statistical smoothing techniques, including smoothing splines and kernel based methods.

Principal investigators: Maarten Jansen - Statistique Mathématique -, Ignace Loris - Mécanique et mathématique appliquée -, Caroline Verhoeven - Service de biostatistique et informatique médicale -.

Spokesperson

Ignace Loris

Mechanics and Applied Mathematics
Faculty of Sciences

Dates
Created on August 31, 2018