arXiv subjects

Christiane Lemieux

Publications and source records attributed to Christiane Lemieux.

9 recordsLinked to original sources

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 $ϕ$. We provide a complete framework to study equidistribution properties of nets in base $ϕ$, 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 $ϕ$ and two-dimensional Hammersley point sets in base $ϕ$ and we prove some properties for $(0,1)-$sequences and $(0,m,2)-$nets in base $ϕ$ respectively. We also include numerical studies of the discrepancy of point sets and sequences in base $ϕ$ 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 $ϕ$ can be generalized to other irrational bases.

math.NT↗

arXiv2022

A negative dependence framework to assess different forms of scrambling

We use the framework of dependence to assess the benefits of scrambling randomly versus deterministically for Faure and Halton sequences. We attempt to answer the following questions: when a deterministic sequence has known defects for small sample sizes, should we address these defects by applying random scrambling or should we find a "good" deterministic scrambling yielding a sequence that can then be randomized using a less computer-intensive randomization method such as a digital shift? And in the latter case, how do we choose a deterministic scrambling and how do we assess whether it is good or not?

stat.CO↗

arXiv2021

Single-Index Importance Sampling with Stratification

In many stochastic problems, the output of interest depends on an input random vector mainly through a single random variable (or index) via an appropriate univariate transformation of the input. We exploit this feature by proposing an importance sampling method that makes rare events more likely by changing the distribution of the chosen index. Further variance reduction is guaranteed by combining this single-index importance sampling approach with stratified sampling. The dimension-reduction effect of single-index importance sampling also enhances the effectiveness of quasi-Monte Carlo methods. The proposed method applies to a wide range of financial or risk management problems. We demonstrate its efficiency for estimating large loss probabilities of a credit portfolio under a normal and t-copula model and show that our method outperforms the current standard for these problems.

stat.CO↗

arXiv2020

On the dependence structure and quality of scrambled $(t,m,s)-$nets

In this paper we develop a framework to study the dependence structure of scrambled $(t,m,s)$-nets. It relies on values denoted by $C_b(\mathbf{k};P_n)$, which are related to how many distinct pairs of points from $P_n$ lie in the same elementary $\mathbf{k}-$interval in base $b$. These values quantify the equidistribution properties of $P_n$ in a more informative way than the parameter $t$. They also play a key role in determining if a scrambled set $\tilde{P}_n$ is negative lower orthant dependent (NLOD). Indeed this property holds if and only if $C_b(\mathbf{k};P_n) \le 1$ for all $\mathbf{k} \in \mathbb{N}^s$, which in turn implies that a scrambled digital $(t,m,s)-$net in base $b$ is NLOD if and only if $t=0$. Through numerical examples we demonstrate that these $C_b(\mathbf{k};P_n)$ values are a powerful tool to compare the quality of different $(t,m,s)$-nets, and to enhance our understanding of how scrambling can improve the quality of deterministic point sets.

math.PR↗

arXiv2020

Normal variance mixtures: Distribution, density and parameter estimation

Normal variance mixtures are a class of multivariate distributions that generalize the multivariate normal by randomizing (or mixing) the covariance matrix via multiplication by a non-negative random variable W. The multivariate t distribution is an example of such mixture, where W has an inverse-gamma distribution. Algorithms to compute the joint distribution function and perform parameter estimation for the multivariate normal and t (with integer degrees of freedom) can be found in the literature and are implemented in, e.g., the R package mvtnorm. In this paper, efficient algorithms to perform these tasks in the general case of a normal variance mixture are proposed. In addition to the above two tasks, the evaluation of the joint (logarithmic) density function of a general normal variance mixture is tackled as well, as it is needed for parameter estimation and does not always exist in closed form in this more general setup. For the evaluation of the joint distribution function, the proposed algorithms apply randomized quasi-Monte Carlo (RQMC) methods in a way that improves upon existing methods proposed for the multivariate normal and t distributions. An adaptive RQMC algorithm that similarly exploits the superior convergence properties of RQMC methods is presented for the task of evaluating the joint log-density function. This allows the parameter estimation task to be accomplished via an EM-like algorithm where all weights and log-densities are numerically estimated. It is demonstrated through numerical examples that the suggested algorithms are quite fast; even for high dimensions around 1000 the distribution function can be estimated with moderate accuracy using only a few seconds of run time. Even log-densities around -100 can be estimated accurately and quickly. An implementation of all algorithms presented in this work is available in the R package nvmix (version >= 0.0.4).

stat.CO↗

arXiv2016

Quasi-random numbers for copula models

The present work addresses the question how sampling algorithms for commonly applied copula models can be adapted to account for quasi-random numbers. Besides sampling methods such as the conditional distribution method (based on a one-to-one transformation), it is also shown that typically faster sampling methods (based on stochastic representations) can be used to improve upon classical Monte Carlo methods when pseudo-random number generators are replaced by quasi-random number generators. This opens the door to quasi-random numbers for models well beyond independent margins or the multivariate normal distribution. Detailed examples (in the context of finance and insurance), illustrations and simulations are given and software has been developed and provided in the R packages copula and qrng.

stat.CO↗

arXiv2013

Lattice Particle Filters

A standard approach to approximate inference in state-space models isto apply a particle filter, e.g., the Condensation Algorithm.However, the performance of particle filters often varies significantlydue to their stochastic nature.We present a class of algorithms, called lattice particle filters, thatcircumvent this difficulty by placing the particles deterministicallyaccording to a Quasi-Monte Carlo integration rule.We describe a practical realization of this idea, discuss itstheoretical properties, and its efficiency.Experimental results with a synthetic 2D tracking problem show that thelattice particle filter is equivalent to a conventional particle filterthat has between 10 and 60% more particles, depending ontheir "sparsity" in the state-space.We also present results on inferring 3D human motion frommoving light displays.

cs.AI↗