A greedy screening test strategy to accelerate solving LASSO problems with small regularization parameters

摘要: In the era of big data remarked by high dimensionality and large sample size, least absolute shrinkage selection operator (LASSO) problems demand efficient algorithms. Both static dynamic strategies based on screening test principle have been proposed recently, in order to safely filter out irrelevant atoms from dictionary. However, such only work well for LASSO with regularization parameters, lose their efficiency those small parameters. This paper presents a novel greedy strategy accelerate solving as its effectiveness through adoption relatively larger parameter which filters every iteration. Further more, convergence proof is given, computational complexity solvers integrated this investigated. Numerical experiments both synthetic real sets support strategy, results show it outperforms