Quick Streaming Algorithms for Maximization of Monotone Submodular Functions in Linear Time
This work addresses efficient optimization for large-scale data processing, offering incremental advances in streaming algorithm design for submodular functions.
The authors tackled the problem of monotone submodular maximization under cardinality constraints by introducing the first deterministic linear-time streaming algorithms, achieving constant approximation ratios with query complexities as low as ⌈n/c⌉ + c and demonstrating empirical improvements over state-of-the-art methods.
We consider the problem of monotone, submodular maximization over a ground set of size $n$ subject to cardinality constraint $k$. For this problem, we introduce the first deterministic algorithms with linear time complexity; these algorithms are streaming algorithms. Our single-pass algorithm obtains a constant ratio in $\lceil n / c \rceil + c$ oracle queries, for any $c \ge 1$. In addition, we propose a deterministic, multi-pass streaming algorithm with a constant number of passes that achieves nearly the optimal ratio with linear query and time complexities. We prove a lower bound that implies no constant-factor approximation exists using $o(n)$ queries, even if queries to infeasible sets are allowed. An empirical analysis demonstrates that our algorithms require fewer queries (often substantially less than $n$) yet still achieve better objective value than the current state-of-the-art algorithms, including single-pass, multi-pass, and non-streaming algorithms.