Search

Shapley value is originally a concept in econometrics to fairly distribute both gains and costs to players in a coalition game. In the recent decades, its application has been extended to other areas such as marketing, engineering and machine learning. For example, it produces reasonable solutions for problems in sensitivity analysis, local model explanation toward the interpretable machine learning, node importance in social network, attribution models, etc. However, it could be very expensive to compute the Shapley value. Specifically, in a d-player coalition game, calculating a Shapley value requires the evaluation of d ! or 2 d marginal contribution values, depending on whether we are taking the permutation or combination formulation of the Shapley value. Hence, it becomes infeasible to calculate the Shapley value when d is reasonably large. A common remedy is to take a random sample of the permutations to surrogate for the complete list of permutations. We find an advanced sampling scheme can be designed to yield much more accurate estimation of the Shapley value than the simple random sampling (SRS). Our sampling scheme is based on combinatorial structures in the field of design of experiments (DOE), particularly the order-of-addition experimental designs for the study of how the orderings of components would affect the output. We show that the obtained estimates are unbiased, and can sometimes deterministically recover the original Shapley value. Both theoretical and simulations results show that our DOE-based sampling scheme outperforms SRS in terms of estimation accuracy. Surprisingly, it is also slightly faster than SRS. Lastly, real data analysis is conducted for the C. elegans nervous system and the 9/11 terrorist network. for this article are available online. [ABSTRACT FROM AUTHOR]

Row-column designs have been widely used in experiments involving double confounding. Among them, one that provides unconfounded estimation of all main effects and as many two-factor interactions as possible is preferred, and is called optimal. Most current work focuses on the construction of two-level row-column designs, while the corresponding optimality theory has been largely ignored. Moreover, most constructed designs contain at least one replicate of a full factorial design, which is not flexible as the number of factors increases. In this study, a theoretical framework is built up to evaluate the optimality of row-column designs with prime level. A method for constructing optimal row-column designs with prime level is proposed. Subsequently, optimal full factorial three-level row-column designs are constructed for any parameter combination. Optimal fractional factorial two-level and three-level row-column designs are also constructed for cost saving. [ABSTRACT FROM AUTHOR]

Quasi-Monte Carlo (QMC) methods have been successfully used for the estimation of numerical integrations arising in many applications. In most QMC methods, low-discrepancy sequences have been used, such as digital nets and lattice rules. In this paper, we derive the convergence rates of order of some improved discrepancies, such as centered L 2 -discrepancy, wrap-around L 2 -discrepancy, and mixture discrepancy, and propose a randomized QMC method based on a uniform design constructed by the mixture discrepancy and Baker's transformation. Moreover, the numerical results show that the proposed method has better approximation than the Monte Carlo method and many other QMC methods, especially when the number of dimensions is less than 10. [ABSTRACT FROM AUTHOR]

Computer experiments play a crucial role when physical experiments are expensive or difficult to be carried out. As a kind of designs for computer experiments, maximin distance designs have been widely studied. Many existing methods for obtaining maximin distance designs are based on stochastic algorithms, and these methods will be infeasible when the run size or number of factors is large. In this paper, we propose some deterministic construction methods for maximin L 2 -distance designs in two to five dimensions based on densest packings. The resulting designs have large L 2 -distances and are mirror-symmetric. Some of them have the same L 2 -distances as the existing optimal maximin distance designs, and some of the others are completely new. Especially, the resulting 2-dimensional designs possess a good projection property. [ABSTRACT FROM AUTHOR]

The minimum aberration criterion is popular for selecting good designs with qualitative factors under an ANOVA model, and the minimum β\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\beta $$\end{document}-aberration criterion is more suitable for selecting designs with quantitative factors under a polynomial model. However, numerous computer experiments involve both qualitative and quantitative factors, while there is a lack of a reasonable criterion to assess the effectiveness of such designs. This paper proposes some important properties of the β\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\beta $$\end{document}-wordlength pattern for mixed-level designs, and introduces the minimum θ\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\theta $$\end{document}-aberration criterion for comparing and selecting designs with qualitative and quantitative factors based on a full model involving all interactions of the factors. The computation of the θ\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\theta $$\end{document}-wordlength pattern is optimized by the generalized wordlength enumerator. Then we provide some construction methods for designs with less θ\documentclass[12pt]{minimal}\usepackage{amsmath}\usepackage{wasysym}\usepackage{amsfonts}\usepackage{amssymb}\usepackage{amsbsy}\usepackage{mathrsfs}\usepackage{upgreek}\setlength{\oddsidemargin}{-69pt}\begin{document}$$\theta $$\end{document}-aberration, and apply this criterion to screen the marginally coupled designs and the doubly coupled designs. [ABSTRACT FROM AUTHOR]

This paper introduces a novel approach to reducing statistical errors in generative models, with a specific focus on generative adversarial networks (GANs). Inspired by the error analysis of GANs, we find that statistical errors mainly arise from random sampling, leading to significant uncertainties in GANs. To address this issue, we propose a selective sampling mechanism called space‐filling sampling. Our method aims to increase the sampling probability in areas with insufficient data, thereby improving the learning performance of the generator. Theoretical analysis confirms the effectiveness of our approach in reducing statistical errors and accelerating convergence in GANs. This research represents a pioneering effort in targeting the reduction of statistical errors in GANs, and it demonstrates the potential for enhancing the training of other generative models. [ABSTRACT FROM AUTHOR]

Strong orthogonal arrays have better space-filling properties than ordinary orthogonal arrays for computer experiments. We consider column-orthogonal strong orthogonal arrays of strength two plus and three minus, and present methods of constructing such designs. Several situations are examined, including those of four or higher levels and mixed levels. The methods are based on both regular and nonregular designs. The resulting designs inherit the good property of strong orthogonal arrays of strength two plus or three and have the additional property of column orthogonality. This type of design is a better choice for computer experiments. [ABSTRACT FROM AUTHOR]

AbstractApproximate Bayesian computation (ABC) is a class of Bayesian inference algorithms that targets problems with intractable or unavailable likelihood functions. It uses synthetic data drawn from the simulation model to approximate the posterior distribution. However, ABC is computationally intensive for complex models in which simulating synthetic data is very expensive. In this article, we propose an early rejection Markov chain Monte Carlo (ejMCMC) sampler based on Gaussian processes to accelerate inference speed. We early reject samples in the first stage of the kernel using a discrepancy model, in which the discrepancy between the simulated and observed data is modeled by Gaussian process (GP). Hence, synthetic data is generated only if the parameter space is worth exploring. We demonstrate through theory, simulation experiments, and real data analysis that the new algorithm significantly improves inference efficiency compared to existing early-rejection MCMC algorithms. In addition, we employ our proposed method within an ABC sequential Monte Carlo (SMC) sampler. In our numerical experiments, we use examples of ordinary differential equations, stochastic differential equations, and delay differential equations to demonstrate the effectiveness of the proposed algorithm. We develop an R package that is available at https://github.com/caofff/ejMCMC. [ABSTRACT FROM AUTHOR]

Space-filling designs are useful for exploring the relationship between the response and factors, especially when the true model is unknown. The wrap-around L 2 -discrepancy is an important measure of the uniformity, and has often been used as a type of space-filling criterion. However, most obtained designs are generated through stochastic optimization algorithms, and cannot achieve the lower bound of the discrepancies and are only nearly uniform. Then deterministic construction methods for uniform designs are desired. This paper constructs uniform designs under the wrap-around L 2 -discrepancy by generator matrices of linear codes. Several requirements on the generator matrices, such as a necessary and sufficient condition for generating uniform designs, are derived. Based on these, two simple deterministic constructions for uniform designs are given. Some examples illustrate the effectiveness of them. Moreover, the resulting designs can be regarded as a generalization of good lattice point sets, and also enjoy good orthogonality. • The deterministic construction methods for uniform designs under the wrap-around L 2 -discrepancy are given. • A necessary and sufficient condition for generating uniform designs are derived. • The resulting designs can be regarded as a generalization of good lattice point sets, and also enjoy good orthogonality. [ABSTRACT FROM AUTHOR]

Subsampling plays a crucial role in tackling problems associated with the storage and statistical learning of massive datasets. However, most existing subsampling methods are model-based, which means their performances can drop significantly when the underlying model is misspecified. Such an issue calls for model-free subsampling methods that are robust under diverse model specifications. Recently, several model-free subsampling methods have been developed. However, the computing time of these methods grows explosively with the sample size, making them impractical for handling massive data. In this article, an efficient model-free subsampling method is proposed, which segments the original data into some regular data blocks and obtains subsamples from each data block by the data-driven subsampling method. Compared with existing model-free subsampling methods, the proposed method has a significant speed advantage and performs more robustly for datasets with complex underlying distributions. As demonstrated in simulation experiments, the proposed method is an order of magnitude faster than other commonly used model-free subsampling methods when the sample size of the original dataset reaches the order of 107. Moreover, simulation experiments and case studies show that the proposed method is more robust than other model-free subsampling methods under diverse model specifications and subsample sizes. [ABSTRACT FROM AUTHOR]