Multi-Agent Submodular Optimization

TL;DR

This paper investigates multi-agent submodular optimization, establishing approximation bounds that relate multi-agent problems to their single-agent counterparts, and explores classes with tight approximation gaps.

Contribution

It introduces reductions linking multi-agent and single-agent submodular optimization, expanding tractability and analyzing approximation gaps for various problem classes.

Findings

Maximization admits an O(α)-approximation if single-agent multilinear formulation does

Certain classes like spanning trees and matroids have a multi-agent gap of 1

Minimization has an O(α · min{k, log^2 n})-approximation under convex formulations

Abstract

Recent years have seen many algorithmic advances in the area of submodular optimization: (SO) $\min/\max~f(S): S \in \mathcal{F}$, where $\mathcal{F}$ is a given family of feasible sets over a ground set $V$ and $f:2^V \rightarrow \mathbb{R}$ is submodular. This progress has been coupled with a wealth of new applications for these models. Our focus is on a more general class of \emph{multi-agent submodular optimization} (MASO) which was introduced by Goel et al. in the minimization setting: $\min \sum_i f_i(S_i): S_1 \uplus S_2 \uplus \cdots \uplus S_k \in \mathcal{F}$. Here we use $\uplus$ to denote disjoint union and hence this model is attractive where resources are being allocated across $k$ agents, each with its own submodular cost function $f_i()$. In this paper we explore the extent to which the approximability of the multi-agent problems are linked to their single-agent {\em primitives}, referred to informally as the {\em multi-agent gap}. We present different reductions that transform a multi-agent problem into a single-agent one. For maximization we show that (MASO) admits an $O(\alpha)$-approximation whenever (SO) admits an $\alpha$-approximation over the multilinear formulation, and thus substantially expanding the family of tractable models. We also discuss several family classes (such as spanning trees, matroids, and $p$-systems) that have a provable multi-agent gap of 1. In the minimization setting we show that (MASO) has an $O(\alpha \cdot \min \{k, \log^2 (n)\})$-approximation whenever (SO) admits an $\alpha$-approximation over the convex formulation. In addition, we discuss the class of "bounded blocker" families where there is a provably tight O$(\log n)$ gap between (MASO) and (SO).