Andreas M. Tillmann

The Computational Complexity of the Restricted Isometry Property, the Nullspace Property, and Related Concepts in Compressed Sensing

This paper deals with the computational complexity of conditions which guarantee that the NP-hard problem of finding the sparsest solution to an underdetermined linear system can be solved by efficient algorithms. In the literature, several such conditions have been introduced. The most well-known ones are the mutual coherence, the restricted isometry property (RIP), and the nullspace property (NSP). While evaluating the mutual coherence of a given matrix is easy, it has been suspected for some time that evaluating RIP and NSP is computationally intractable in general. We confirm these conjectures by showing that for a given matrix A and positive integer k, computing the best constants for which the RIP or NSP hold is, in general, NP-hard. These results are based on the fact that determining the spark of a matrix is NP-hard, which is also established in this paper. Furthermore, we also give several complexity statements about problems related to the above concepts.

On the Computational Intractability of Exact and Approximate Dictionary Learning

The efficient sparse coding and reconstruction of signal vectors via linear observations has received a tremendous amount of attention over the last decade. In this context, the automated learning of a suitable basis or overcomplete dictionary from training data sets of certain signal classes for use in sparse representations has turned out to be of particular importance regarding practical signal processing applications. Most popular dictionary learning algorithms involve NP-hard sparse recovery problems in each iteration, which may give some indication about the complexity of dictionary learning but does not constitute an actual proof of computational intractability. In this technical note, we show that learning a dictionary with which a given set of training signals can be represented as sparsely as possible is indeed \ssb NP-hard. Moreover, we also establish hardness of approximating the solution to within large factors of the optimal sparsity level. Furthermore, we give \ssb NP-hardness and non-approximability results for a recent dictionary learning variation called the sensor permutation problem. Along the way, we also obtain a new non-approximability result for the classical sparse recovery problem from compressed sensing.

DOLPHIn—Dictionary Learning for Phase Retrieval

We propose a new algorithm to learn a dictionary for reconstructing and sparsely encoding signals from measurements without phase. Specifically, we consider the task of estimating a two-dimensional image from squared-magnitude measurements of a complex-valued linear transformation of the original image. Several recent phase retrieval algorithms exploit underlying sparsity of the unknown signal in order to improve recovery performance. In this work, we consider such a sparse signal prior in the context of phase retrieval, when the sparsifying dictionary is not known in advance. Our algorithm jointly reconstructs the unknown signal-possibly corrupted by noise-and learns a dictionary such that each patch of the estimated image can be sparsely represented. Numerical experiments demonstrate that our approach can obtain significantly better reconstructions for phase retrieval problems with noise than methods that cannot exploit such “hidden” sparsity. Moreover, on the theoretical side, we provide a convergence result for our method.

Solving Basis Pursuit: Heuristic Optimality Check and Solver Comparison

The problem of finding a minimum ℓ1-norm solution to an underdetermined linear system is an important problem in compressed sensing, where it is also known as basis pursuit. We propose a heuristic optimality check as a general tool for ℓ1-minimization, which often allows for early termination by “guessing” a primal-dual optimal pair based on an approximate support. Moreover, we provide an extensive numerical comparison of various state-of-the-art ℓ1-solvers that have been proposed during the last decade, on a large test set with a variety of explicitly given matrices and several right-hand sides per matrix reflecting different levels of solution difficulty. The results, as well as improvements by the proposed heuristic optimality check, are analyzed in detail to provide an answer to the question which algorithm is the best.

Projection onto the cosparse set is NP-hard

The computational complexity of a problem arising in the context of sparse optimization is considered, namely, the projection onto the set of k-cosparse vectors w.r.t. some given matrix Ω. It is shown that this projection problem is (strongly) NP-hard, even in the special cases in which the matrix Ω contains only ternary or bipolar coefficients. Interestingly, this is in contrast to the projection onto the set of k-sparse vectors, which is trivially solved by keeping only the k largest coefficients.

Visualization of Astronomical Nebulae via Distributed Multi-GPU Compressed Sensing Tomography

The 3D visualization of astronomical nebulae is a challenging problem since only a single 2D projection is observable from our fixed vantage point on Earth. We attempt to generate plausible and realistic looking volumetric visualizations via a tomographic approach that exploits the spherical or axial symmetry prevalent in some relevant types of nebulae. Different types of symmetry can be implemented by using different randomized distributions of virtual cameras. Our approach is based on an iterative compressed sensing reconstruction algorithm that we extend with support for position-dependent volumetric regularization and linear equality constraints. We present a distributed multi-GPU implementation that is capable of reconstructing high-resolution datasets from arbitrary projections. Its robustness and scalability are demonstrated for astronomical imagery from the Hubble Space Telescope. The resulting volumetric data is visualized using direct volume rendering. Compared to previous approaches, our method preserves a much higher amount of detail and visual variety in the 3D visualization, especially for objects with only approximate symmetry.

An infeasible-point subgradient method using adaptive approximate projections

We propose a new subgradient method for the minimization of nonsmooth convex functions over a convex set. To speed up computations we use adaptive approximate projections only requiring to move within a certain distance of the exact projections (which decreases in the course of the algorithm). In particular, the iterates in our method can be infeasible throughout the whole procedure. Nevertheless, we provide conditions which ensure convergence to an optimal feasible point under suitable assumptions. One convergence result deals with step size sequences that are fixed a priori. Two other results handle dynamic Polyak-type step sizes depending on a lower or upper estimate of the optimal objective function value, respectively. Additionally, we briefly sketch two applications: Optimization with convex chance constraints, and finding the minimum ℓ1-norm solution to an underdetermined linear system, an important problem in Compressed Sensing.