Core Concepts
This work presents scalable distributed algorithms for size-constrained submodular maximization that achieve constant-factor approximation ratios, constant number of MapReduce rounds, and sublinear adaptive complexity, while also providing the first linear-time distributed algorithm for this problem.
Abstract
The paper focuses on developing efficient distributed algorithms for the problem of size-constrained submodular maximization (SMCC). The key contributions are:
Analysis of low-adaptive algorithms (THRESHSEQMOD and LAG) that satisfy the randomized consistency property (RCP), enabling their use in the distributed MapReduce (MR) setting without loss of approximation.
Development of two constant-round MR algorithms, R-DASH and T-DASH, that achieve approximation ratios of 1/2(1-1/e-ε) and 3/8-ε respectively, with sublinear adaptive complexity.
Introduction of the first linear-time distributed algorithm, L-DIST, that achieves a constant approximation ratio with a constant number of MR rounds.
Proposal of the MED framework that increases the maximum cardinality constraint supported by MR algorithms, at the cost of additional MR rounds.
Extensive empirical evaluation demonstrating significant speedups of the proposed algorithms over state-of-the-art MR algorithms, especially for larger cardinality constraints.
Stats
The paper does not contain any explicit numerical data or statistics. The focus is on the theoretical analysis and design of the distributed algorithms.