$\alpha$-approximate Reductions: a Novel Source of Heuristics for Better Approximation Algorithms

TL;DR

This paper introduces the concept of using carefully designed $oldsymbol{ extalpha}$-approximate reduction rules as practical heuristics to improve approximation algorithms, demonstrated through a new 2-approximate reduction rule for the Dominating Set problem that enhances results on benchmark instances.

Contribution

Proposes that $ extalpha$-approximate reduction rules can be effectively used as heuristics to improve approximation ratios in practice, a novel application beyond their original theoretical purpose.

Findings

A new 2-approximate reduction rule for Dominating Set.

Improved approximation ratios on benchmark instances.

First exploration of $ extalpha$-approximate reduction rules as a practical design technique.

Abstract

Lokshtanov et al.~[STOC 2017] introduced \emph{lossy kernelization} as a mathematical framework for quantifying the effectiveness of preprocessing algorithms in preserving approximation ratios. \emph{$\alpha$-approximate reduction rules} are a central notion of this framework. We propose that carefully crafted $\alpha$-approximate reduction rules can yield improved approximation ratios in practice, while being easy to implement as well. This is distinctly different from the (theoretical) purpose for which Lokshtanov et al. designed $\alpha$-approximate Reduction Rules. As evidence in support of this proposal we present a new 2-approximate reduction rule for the \textsc{Dominating Set} problem. This rule, when combined with an approximation algorithm for \textsc{Dominating Set}, yields significantly better approximation ratios on a variety of benchmark instances as compared to the latter algorithm alone. The central thesis of this work is that $\alpha$-approximate reduction rules can be used as a tool for designing approximation algorithms which perform better in practice. To the best of our knowledge, ours is the first exploration of the use of $\alpha$-approximate reduction rules as a design technique for practical approximation algorithms. We believe that this technique could be useful in coming up with improved approximation algorithms for other optimization problems as well.