Poisson Exchange Beyond Submodularity: Effective Approximation Algorithms for Offline and Online Subset Selection over Matroids
Abstract
Over the past decade, a growing body of research has shown that -weak submodularity broadly arises in numerous subset selection tasks, including feature selection, neural network pruning, and video summarization. Despite its prevalence, maximizing a -weakly submodular function subject to a general matroid constraint remains challenging. To date, the only known approximation guarantee is the conservative factor established by \citet{chen2018weakly}. To improve upon this result, this paper proposes a novel algorithm called \MGPE, which repeatedly performs maximum-gain local exchanges through careful control of a non-homogeneous Poisson clock, and proves that this \MGPE\ can attain an approximation ratio arbitrarily close to . In sharp contrast to the previous guarantee, our obtained factor not only strictly improves upon for every , but also can asymptotically approach the optimal -approximation for submodular maximization as . Furthermore, we surprisingly find that when the matroid constraint reduces to a cardinality or the objective satisfies the stronger notion of -weak DR-submodularity, \MGPE\ can automatically recover the tight approximation ratios of and , respectively. Here, denotes the DR ratio.