Core Concepts
The authors propose an Oracle-Net, a graph neural network, to predict the support of the sparse solution in a variational framework for nonlinear compressed sensing in Electrical Impedance Tomography reconstruction problems. The derived nonsmooth optimization problem is efficiently solved through a constrained proximal gradient method, providing error bounds on the approximate solution.
Abstract
The content discusses a framework for solving nonlinear inverse problems, with a focus on Electrical Impedance Tomography (EIT) reconstruction. The key points are:
Nonlinear compressed sensing: The authors consider a more general setting that extends the concepts of compressive sensing and sparse recovery to inverse and ill-posed nonlinear problems. The goal is to recover an unknown sparse vector σ† from incomplete and contaminated nonlinear measurements Λδ.
Variational regularization model: A regularized minimization problem is introduced to recover σ†, involving a sparsity-promoting penalty R and a constraint set K. The authors propose to use an "Oracle" operator to predict the support of the sparse solution and integrate it into the regularization model.
Oracle-Net for support estimation: A graph neural network, named Oracle-Net, is proposed to predict the support of the sparse solution from the nonlinear measurements. This is integrated into the regularized recovery model to enforce sparsity.
Proximal gradient method: The derived nonsmooth optimization problem is efficiently solved through a constrained proximal gradient method. Error bounds on the approximate solution are provided, showing that the accurate recovery of σ† is possible under certain conditions on the Jacobian of the measurement system Φ.
Application to EIT: The authors focus on the EIT inverse problem as a case study, but the proposed Oracle-based framework could be adapted to other nonlinear ill-posed problems.