arXiv subjects

Nathan Kirk

Publications and source records attributed to Nathan Kirk.

16 recordsLinked to original sources

arXiv2026

A Bayesian Approach to Low-Discrepancy Subset Selection

Low-discrepancy designs play a central role in quasi-Monte Carlo methods and are increasingly influential in other domains such as machine learning, robotics and computer graphics, to name a few. In recent years, one such low-discrepancy construction method called subset selection has received a lot of attention. Given a large population, one optimally selects a small low-discrepancy subset with respect to a discrepancy-based objective. Versions of this problem are known to be NP-hard. In this text, we establish, for the first time, that the subset selection problem with respect to kernel discrepancies is also NP-hard. Motivated by this intractability, we propose a Bayesian Optimization procedure for the subset selection problem utilizing the recent notion of deep embedding kernels. We demonstrate the performance of the BO algorithm to minimize discrepancy measures and note that the framework is broadly applicable any design criteria.

stat.ME↗

arXiv2025

Optimizing Kernel Discrepancies via Subset Selection

Kernel discrepancies are a powerful tool for analyzing worst-case errors in quasi-Monte Carlo (QMC) methods. Building on recent advances in optimizing such discrepancy measures, we extend the subset selection problem to the setting of kernel discrepancies, selecting an m-element subset from a large population of size $n \gg m$. We introduce a novel subset selection algorithm applicable to general kernel discrepancies to efficiently generate low-discrepancy samples from both the uniform distribution on the unit hypercube, the traditional setting of classical QMC, and from more general distributions $F$ with known density functions by employing the kernel Stein discrepancy. We also explore the relationship between the classical $L_2$ star discrepancy and its $L_\infty$ counterpart.

stat.ML↗

arXiv2025

Neural Low-Discrepancy Sequences

Low-discrepancy points are designed to efficiently fill the space in a uniform manner. This uniformity is highly advantageous in many problems in science and engineering, including in numerical integration, computer vision, machine perception, computer graphics, machine learning, and simulation. Whereas most previous low-discrepancy constructions rely on abstract algebra and number theory, Message-Passing Monte Carlo (MPMC) was recently introduced to exploit machine learning methods for generating point sets with lower discrepancy than previously possible. However, MPMC is limited to generating point sets and cannot be extended to low-discrepancy sequences (LDS), i.e., sequences of points in which every prefix has low discrepancy, a property essential for many applications. To address this limitation, we introduce Neural Low-Discrepancy Sequences (NeuroLDS), the first machine learning-based framework for generating finite LDS. Drawing inspiration from classical LDS, we train a neural network to map indices to points such that the resulting sequences exhibit minimal discrepancy across all prefixes. To this end, we deploy a two-stage learning process: supervised approximation of classical constructions followed by unsupervised fine-tuning to minimize prefix discrepancies. We demonstrate that NeuroLDS outperforms all previous LDS constructions by a significant margin with respect to discrepancy measures. Moreover, we demonstrate the effectiveness of NeuroLDS across diverse applications, including numerical integration, robot motion planning, and scientific machine learning. These results highlight the promise and broad significance of Neural Low-Discrepancy Sequences. Our code can be found at https://github.com/camail-official/neuro-lds.

cs.LG↗

arXiv2025

High-Dimensional Quasi-Monte Carlo via Combinatorial Discrepancy

Monte Carlo (MC) and Quasi-Monte Carlo (QMC) methods are classical approaches for the numerical integration of functions $f$ over $[0,1]^d$. While QMC methods can achieve faster convergence rates than MC in moderate dimensions, their tractability in high dimensions typically relies on additional structure -- such as low effective dimension or carefully chosen coordinate weights -- since worst-case error bounds grow prohibitively large as $d$ increases. In this work, we study the construction of high-dimensional QMC point sets via combinatorial discrepancy, extending the recent QMC method of Bansal and Jiang. We establish error bounds for these constructions in weighted function spaces, and for functions with low effective dimension in both the superposition and truncation sense. We also present numerical experiments to empirically assess the performance of these constructions.

math.NA↗

arXiv2025

On the optimization of discrepancy measures

Points in the unit cube with low discrepancy can be constructed using algebra or, more recently, by direct computational optimization of a criterion. The usual $L_\infty$ star discrepancy is a poor criterion for this because it is computationally expensive and lacks differentiability. Its usual replacement, the $L_2$ star discrepancy, is smooth but exhibits other pathologies shown by J. Matou\v{s}ek. In an attempt to address these problems, we introduce the \textit{average squared discrepancy} which averages over $2^d$ versions of the $L_2$ star discrepancy anchored in the different vertices of $[0,1]^d$. Not only can this criterion be computed in $O(dn^2)$ time, like the $L_2$ star discrepancy, but also we show that it is equivalent to a weighted symmetric $L_2$ criterion of Hickernell's by a constant factor. We compare this criterion with a wide range of traditional discrepancy measures, and show that only the average squared discrepancy avoids the problems raised by Matou\v{s}ek. Furthermore, we present a comprehensive numerical study showing in particular that optimizing for the average squared discrepancy leads to strong performance for the $L_2$ star discrepancy, whereas the converse does not hold.

math.NA↗

arXiv2025

Enhancing Neural Autoregressive Distribution Estimators for Image Reconstruction

Autoregressive models are often employed to learn distributions of image data by decomposing the $D$-dimensional density function into a product of one-dimensional conditional distributions. Each conditional depends on preceding variables (pixels, in the case of image data), making the order in which variables are processed fundamental to the model performance. In this paper, we study the problem of observing a small subset of image pixels (referred to as a pixel patch) to predict the unobserved parts of the image. As our prediction mechanism, we propose a generalized version of the convolutional neural autoregressive distribution estimation (ConvNADE) model adapted for real-valued and color images. Moreover, we investigate the quality of image reconstruction when observing both random pixel patches and low-discrepancy pixel patches inspired by quasi-Monte Carlo theory. Experiments on benchmark datasets demonstrate that, where design permits, pixels sampled or stored to preserve uniform coverage improves reconstruction fidelity and test performance.

eess.IV↗

arXiv2025

Multilevel Sampling in Algebraic Statistics

This paper proposes a multilevel sampling algorithm for fiber sampling problems in algebraic statistics, inspired by Henry Wynn's suggestion to adapt multilevel Monte Carlo (MLMC) ideas to discrete models. Focusing on log-linear models, we sample from high-dimensional lattice fibers defined by algebraic constraints. Building on Markov basis methods and results from Diaconis and Sturmfels, our algorithm uses variable step sizes to accelerate exploration and reduce the need for long burn-in. We introduce a novel Fiber Coverage Score (FCS) based on Voronoi partitioning to assess sample quality, and highlight the utility of the Maximum Mean Discrepancy (MMD) quality metric. Simulations on benchmark fibers show that multilevel sampling outperforms naive MCMC approaches. Our results demonstrate that multilevel methods, when properly applied, provide practical benefits for discrete sampling in algebraic statistics.

stat.CO↗

arXiv2025

Low Stein Discrepancy via Message-Passing Monte Carlo

Message-Passing Monte Carlo (MPMC) was recently introduced as a novel low-discrepancy sampling approach leveraging tools from geometric deep learning. While originally designed for generating uniform point sets, we extend this framework to sample from general multivariate probability distributions with known probability density function. Our proposed method, Stein-Message-Passing Monte Carlo (Stein-MPMC), minimizes a kernelized Stein discrepancy, ensuring improved sample quality. Finally, we show that Stein-MPMC outperforms competing methods, such as Stein Variational Gradient Descent and (greedy) Stein Points, by achieving a lower Stein discrepancy.

cs.LG↗

arXiv2025

Quasi-Monte Carlo Methods: What, Why, and How?

Many questions in quantitative finance, uncertainty quantification, and other disciplines are answered by computing the population mean, $\mu := \mathbb{E}(Y)$, where instances of $Y:=f(\boldsymbol{X})$ may be generated by numerical simulation and $\boldsymbol{X}$ has a simple probability distribution. The population mean can be approximated by the sample mean, $\hat{\mu}_n := n^{-1} \sum_{i=0}^{n-1} f(\boldsymbol{x}_i)$ for a well chosen sequence of nodes, $\{\boldsymbol{x}_0, \boldsymbol{x}_1, \ldots\}$ and a sufficiently large sample size, $n$. Computing $\mu$ is equivalent to computing a $d$-dimensional integral, $\int f(\boldsymbol{x}) \varrho(\boldsymbol{x}) \, \mathrm{d} \boldsymbol{x}$, where $\varrho$ is the probability density for $\boldsymbol{X}$. Quasi-Monte Carlo methods replace independent and identically distributed sequences of random vector nodes, $\{\boldsymbol{x}_i \}_{i = 0}^{\infty}$, by low discrepancy sequences. This accelerates the convergence of $\hat{\mu}_n$ to $\mu$ as $n \to \infty$. This tutorial describes low discrepancy sequences and their quality measures. We demonstrate the performance gains possible with quasi-Monte Carlo methods. Moreover, we describe how to formulate problems to realize the greatest performance gains using quasi-Monte Carlo. We also briefly describe the use of quasi-Monte Carlo methods for problems beyond computing the mean, $\mu$.

math.NA↗

arXiv2024

Message-Passing Monte Carlo: Generating low-discrepancy point sets via Graph Neural Networks

Discrepancy is a well-known measure for the irregularity of the distribution of a point set. Point sets with small discrepancy are called low-discrepancy and are known to efficiently fill the space in a uniform manner. Low-discrepancy points play a central role in many problems in science and engineering, including numerical integration, computer vision, machine perception, computer graphics, machine learning, and simulation. In this work, we present the first machine learning approach to generate a new class of low-discrepancy point sets named Message-Passing Monte Carlo (MPMC) points. Motivated by the geometric nature of generating low-discrepancy point sets, we leverage tools from Geometric Deep Learning and base our model on Graph Neural Networks. We further provide an extension of our framework to higher dimensions, which flexibly allows the generation of custom-made points that emphasize the uniformity in specific dimensions that are primarily important for the particular problem at hand. Finally, we demonstrate that our proposed model achieves state-of-the-art performance superior to previous methods by a significant margin. In fact, MPMC points are empirically shown to be either optimal or near-optimal with respect to the discrepancy for low dimension and small number of points, i.e., for which the optimal discrepancy can be determined. Code for generating MPMC points can be found at https://github.com/tk-rusch/MPMC.

cs.LG↗

arXiv2024

An improved Halton sequence for implementation in quasi-Monte Carlo methods

Despite possessing the low-discrepancy property, the classical d dimensional Halton sequence is known to exhibit poorly distributed projections when d becomes even moderately large. This, in turn, often implies bad performance when implemented in quasi-Monte Carlo (QMC) methods in comparison to, for example, the Sobol' sequence. As an attempt to eradicate this issue, we propose an adapted Halton sequence built by integer and irrational based van der Corput sequences and show empirically improved performance with respect to the accuracy of estimates in numerical integration and simulation. In addition, for the first time, a scrambling algorithm is proposed for irrational based digital sequences.

math.NA↗

arXiv2023

Golden Ratio Nets and Sequences

In this paper we introduce and study nets and sequences constructed in an irrational base, focusing on the case of a base given by the golden ratio $\phi$. We provide a complete framework to study equidistribution properties of nets in base $\phi$, which among other things requires the introduction of a new concept of prime elementary intervals which differ from the standard definition used for integer bases. We define the one-dimensional van der Corput sequence in base $\phi$ and two-dimensional Hammersley point sets in base $\phi$ and we prove some properties for $(0,1)-$sequences and $(0,m,2)-$nets in base $\phi$ respectively. We also include numerical studies of the discrepancy of point sets and sequences in base $\phi$ showing an improvement in distribution properties over traditional integer based Hammersley constructions. As motivation for future research, we show how the equidistribution notions that are introduced for base $\phi$ can be generalized to other irrational bases.

math.NT↗

arXiv2022

On the expected L2-discrepancy of jittered sampling

For $m, d \in \mathbb{N}$, a jittered sample of $N=m^d$ points can be constructed by partitioning $[0,1]^d$ into $m^d$ axis-aligned equivolume boxes and placing one point independently and uniformly at random inside each box. We utilise a formula for the expected $\mathcal{L}_2-$discrepancy of stratified samples stemming from general equivolume partitions of $[0,1]^d$ which recently appeared, to derive a closed form expression for the expected $\mathcal{L}_2-$discrepancy of a jittered point set for any $m, d \in \mathbb{N}$. As a second main result we derive a similar formula for the expected Hickernell $\mathcal{L}_2-$discrepancy of a jittered point set which also takes all projections of the point set to lower dimensional faces of the unit cube into account.

math.PR↗

arXiv2022

Partitions for stratified sampling

Classical jittered sampling partitions $[0,1]^d$ into $m^d$ cubes for a positive integer $m$ and randomly places a point inside each of them, providing a point set of size $N=m^d$ with small discrepancy. The aim of this note is to provide a construction of partitions that works for arbitrary $N$ and improves straight-forward constructions. We show how to construct equivolume partitions of the $d$-dimensional unit cube with hyperplanes that are orthogonal to the main diagonal of the cube. We investigate the discrepancy of such point sets and optimise the expected discrepancy numerically by relaxing the equivolume constraint using different black-box optimisation techniques.

math.CO↗

arXiv2020

On Proinov's lower bound for the diaphony

In 1986, Proinov published an explicit lower bound for the diaphony of both finite and infinite sequences of points in the d-dimensional unit cube. To the best of our knowledge, the proofs of these results were so far only available in a monograph of Proinov written in Bulgarian. The first contribution of our paper is to give a self contained version of Proinov's proof in English. Along the way, we improve the asymptotic constants using recent and corrected results of Hinrichs and Markhasin. Finally, as a main result, we use the method of Proinov to derive an explicit lower bound for the dyadic diaphony of finite and infinite sequences in a similar fashion.

math.NT↗

arXiv2014

A computational study of the effects of remodelled electrophysiology and mechanics on initiation of ventricular fibrillation in human heart failure

The study of pathological cardiac conditions such as arrhythmias, a major cause of mortality in heart failure, is becoming increasingly informed by computational simulation, numerically modelling the governing equations. This can provide insight where experimental work is constrained by technical limitations and/or ethical issues. As the models become more realistic, the construction of efficient and accurate computational models becomes increasingly challenging. In particular, recent developments have started to couple the electrophysiology models with mechanical models in order to investigate the effect of tissue deformation on arrhythmogenesis, thus introducing an element of nonlinearity into the mathematical representation. This paper outlines a biophysically-detailed computational model of coupled electromechanical cardiac activity which uses the finite element method to approximate both electrical and mechanical systems on unstructured, deforming, meshes. An ILU preconditioner is applied to improve performance of the solver. This software is used to examine the role of electrophysiology, fibrosis and mechanical deformation on the stability of spiral wave dynamics in human ventricular tissue by applying it to models of both healthy and failing tissue. The latter was simulated by modifying (i) cellular electrophysiological properties, to generate an increased action potential duration and altered intracellular calcium handling, and (ii) tissue-level properties, to simulate the gap junction remodelling, fibrosis and increased tissue stiffness seen in heart failure. The resulting numerical experiments suggest that, for the chosen mathematical models of electrophysiology and mechanical response, introducing tissue level fibrosis can have a destabilising effect on the dynamics, while the net effect of the electrophysiological remodelling stabilises the system.

cs.CE↗