Iterative Algorithms

Sub-package containing optimization routines.

DeepMR provides optimization routines for compressed sensing and network unfolding.

Optimization Steps

Operators representing single iterations of classical optimization algorithms.

DeepMR expose these operators as torch.nn objects to be chained e.g., in unrolled Neural Network architectures and the corresponding functional versions for standalone usage.

deepmr.optim.CGStep	Conjugate Gradient method step.
deepmr.optim.ADMMStep	Alternate Direction of Multipliers Method step.
deepmr.optim.PGDStep	Proximal Gradient Method step.
deepmr.optim.cg_solve	Solve inverse problem using Conjugate Gradient method.
deepmr.optim.admm_solve	Solve inverse problem using Alternate Direction of Multipliers Method.
deepmr.optim.pgd_solve	Solve inverse problem using Proximal Gradient Method.

In addition, we provide utils to estimate matrix-free operator properties, such as maximum eigenvalue.

Linop linear algebra

deepmr.optim.power_method

Use power iteration to calculate the spectral norm of a Linop.