Momentum-Accelerated Sparse Signal Recovery With Signal Constraints

What is it about?

We developed computationally efficient algorithms for minimization of general nonlinear negative log-likelihood (NLL) functions with sparse signal regularizations and convex-set constraints on the signal. Our algorithms can solve inverse problems under various statistical models, including the popular Gaussian and Poisson generalized linear models (GLMs).

Featured Image

Why is it important?

The proposed optimization framework is matrix-free, (mostly) tuning-free, achieves state-of-the-art empirical performance, and has strong convergence guarantees (of the iterates and the objective function).