Selection by pairwise comparisons with limited resources

TL;DR

This paper explores methods for sorting and selecting objects through uncertain pairwise comparisons, introducing two novel approaches that improve efficiency and establish benchmarks for optimal tournament design.

Contribution

It presents two new methods, ran-fil and min-ent, for object selection via pairwise comparisons, with ran-fil requiring no prior knowledge and min-ent serving as an optimal benchmark.

Findings

ran-fil performs well without prior knowledge

min-ent sets a benchmark for optimality

methods improve efficiency in uncertain comparisons

Abstract

We analyze different methods of sorting and selecting a set of objects by their intrinsic value, via pairwise comparisons whose outcome is uncertain. After discussing the limits of repeated Round Robins, two new methods are presented: The {\it ran-fil} requires no previous knowledge on the set under consideration, yet displaying good performances even in the least favorable case. The {\it min-ent} method sets a benchmark for optimal dynamic tournaments design.