USPACOR: Universal sparsity-controlling outlier rejection

The recent upsurge of research toward compressive sampling and parsimonious signal representations hinges on signals being sparse, either naturally, or, after projecting them on a proper basis. The present paper introduces a neat link between sparsity and a fundamental aspect of statistical inference, namely that of robustness against outliers, even when the signals involved are not sparse. It is argued that controlling sparsity of model residuals leads to statistical learning algorithms that are computationally affordable and universally robust to outlier models. Analysis, comparisons, and corroborating simulations focus on robustifying linear regression, but succinct overview of other areas is provided to highlight universality of the novel framework.