The outlier-corrected-data-adaptive Lasso: A new robust estimator for the independent contamination model

Many of today's signal processing tasks consider sparse models where the number of explanatory variables exceeds the sample size. When dealing with real-world data, the presence of impulsive noise and outliers must also be accounted for. Accurate and robust parameter estimation and consistent variable selection are needed simultaneously. Recently, some popular robust methods have been adapted to such complex settings. Especially, in high dimensional settings, however, it is possible to have a single contaminated predictor being responsible for many outliers. The amount of outliers introduced by this predictor easily exceeds the breakdown point of any existing robust estimator. Therefore, we propose a new robust and sparse estimator, the Outlier-Corrected-Data-(Adaptive) Lasso (OCD-(A) Lasso). It simultaneously handles highly contaminated predictors in the dataset and performs well under the classical contamination model. In a numerical study, it outperforms competing Lasso estimators, at a largely reduced computational complexity compared to its robust counterparts.