Constructing Low Star Discrepancy Point Sets with Genetic Algorithms

Abstract

Geometric discrepancies are standard measures to quantify the irregularity of distributions. They are an important notion in numerical integration. One of the most important discrepancy notions is the so-called \emph{star discrepancy}. Roughly speaking, a point set of low star discrepancy value allows for a small approximation error in quasi-Monte Carlo integration. It is thus the most studied discrepancy notion. In this work we present a new algorithm to compute point sets of low star discrepancy. The two components of the algorithm (for the optimization and the evaluation, respectively) are based on evolutionary principles. Our algorithm clearly outperforms existing approaches. To the best of our knowledge, it is also the first algorithm which can be adapted easily to optimize inverse star discrepancies.

Extended abstract appeared at GECCO 2013. v2: corrected 3 numbers in table 4

References in corpus (1)

• Calculation of Discrepancy Measures and Applications

Cited by in corpus (6)

• Calculation of Discrepancy Measures and Applications
• Message-Passing Monte Carlo: Generating low-discrepancy point sets via Graph Neural Networks
• Computing Star Discrepancies with Numerical Black-Box Optimization Algorithms
• One-Shot Decision-Making with and without Surrogates
• Subgroup-based Rank-1 Lattice Quasi-Monte Carlo
• Relationship between low-discrepancy sequence and static solution to multi-bodies problem