Wednesday, Aug 6: 10:30 AM - 12:20 PM
4181
Contributed Papers
Music City Center
Room: CC-Davidson Ballroom A2
Main Sponsor
International Chinese Statistical Association
Presentations
Traditionally, design criteria related to effect aliasing are commonly used to evaluate and select optimal designs of the same size in factorial experiments. However, limited attention has been given to the assignment of experimental factors to design columns once an optimal design is selected. This study introduces a modified version of the Summary of Effect Aliasing Structure (SEAS) to assess the severity of aliasing for each design column relative to others within the design matrix. In addition, a systematic approach is proposed to guide the assignment of important experimental factors to design columns with minimal aliasing. Simulation results demonstrate improved performance in experimental analysis when factors are assigned using the proposed method.
Keywords
Summary of Effect Aliasing Structure (SEAS)
supersaturated design (SSD)
factorial designs
factor assignment
Given a set of parameters, several non-isomorphic order-of-addition orthogonal arrays can be generated to design an order-of-addition experiment. Under resource constraints, selecting the best from these candidate designs for the experiment can be practical to extract as much information as possible from the observed data. In this talk, I will introduce a series of numerical indices called centralized generalized wordlength pattern to characterize and compare order-of-addition orthogonal arrays. First, the J-characteristics will be justified for pairwise order matrices. Next, the centralized generalized wordlength pattern will be defined based on the sums of squared differences between the normalized J-characteristics of the pairwise order matrices determined by the fractional and full designs. Some theoretical and computational results will also be introduced for future work.
Keywords
Experimental design
Hadamard matrix
Hamming distance
Inversion
J-characteristic
Projection property
Shapley value is a well-known concept in cooperative game theory which provides a fair way to distribute the revenues or costs to each player. Recently, it has been widely applied in data science for data quality evaluation and model interpretation. There are also other applications beyond economics such as marketing and biology. However, the computation of the Shapley value is an NP-hard problem. For a cooperative game with $n$ players, calculating Shapley values for all players requires calculating the values for $2^n$ different coalitions, which makes it infeasible for a large $n$. \revB{In this paper, we find that the value function of a cooperative game can be viewed as the expected response of a two-level factorial experiment. Based on this perspective, we derive a factorial effects representation of the Shapley value. Then, a fast approximation approach for Shapley values based on fractional factorial designs is proposed.} Under certain conditions, our approach can obtain true Shapley values by calculating values of fewer than $4n^2-4$ different coalitions. Generally, highly accurate approximations of Shapley values can also be obtained by calculating values of additional
Keywords
interoperable artificial intelligence
design of experiments
computation
game theory
In this presentation, we introduce generalized method of moments (GMM) approaches for analyzing recurrent event data with informative censoring. Our framework employs a shared frailty model to account for the correlation between the recurrent event process and censoring time, allowing the frailty variable to be covariate-dependent. Unlike traditional shared-frailty proportional intensity models, our approach is based on rate models, enabling non-proportional rate functions across different covariate groups over time. The proposed GMM methods are robust, as they do not rely on Poisson process assumptions for recurrent events or specific distributional assumptions for frailty and censoring times. We establish the large-sample properties of our methods and evaluate their finite-sample performance through extensive simulation studies. Finally, we apply the proposed methods to a real dataset.
Keywords
Generalized method of moments
Recurrent event data
Informative censoring
Covariate-dependent frailty
First Author
Yu-Jen Cheng, National Tsing Hua University, Taiwan
Presenting Author
Yu-Jen Cheng, National Tsing Hua University, Taiwan
The fraction of variance explained (FVE) in a linear model quantifies the extent to which predictors account for outcome variability. In high dimensional settings, where traditional FVE estimators do not apply, modern FVE estimators such as GWASH struggle with strong correlation among predictors, often found, for example, in brain imaging data. We propose a decomposition framework that partitions the FVE into two components: a low dimensional component capturing the strong correlation, estimable by low dimensional methods, and a high dimensional component with remaining weak correlation, estimable by high dimensional methods. Simulations demonstrate that decomposition of dominant PCs improves bias reduction in FVE estimation compared to standard approaches, such as GCTA. Our method shows consistent performance asymptotically. Application to the Adolescent Brain Cognitive Development (ABCD) dataset validates its real-world applicability, capturing nuanced heritability signals in high-resolution brain imaging data. This work offers a robust framework for unbiased FVE estimation in high-dimensional models.
Keywords
The fraction of variance explained (FVE)
brain imaging
high dimensional
principal component decomposition
Co-Author(s)
Chun Chieh Fan, Laureate Institute for Brain Research
David Azriel, Technion
Armin Schwartzman, University of California, San Diego
First Author
Man Luo, UC San Diego, Department of Family Medicine & Public Health
Presenting Author
Man Luo, UC San Diego, Department of Family Medicine & Public Health
Robust estimation is primarily concerned with providing reliable parameter estimates in the
presence of outliers. Numerous robust loss functions have been proposed in regression and
classification, along with various computing algorithms. In modern penalised generalised
linear models (GLMs), however, there is limited research on robust estimation that can
provide weights to determine the outlier status of the observations. This article proposes
a unified framework based on a large family of loss functions, a composite of concave
and convex functions (CC-family). Properties of the CC-family are investigated, and CC-
estimation is innovatively conducted via the iteratively reweighted convex optimisation
(IRCO), which is a generalisation of the iteratively reweighted least squares in robust
linear regression. For robust GLM, the IRCO becomes the iteratively reweighted GLM.
The unified framework contains penalised estimation and robust support vector machine
(SVM) and is demonstrated with a variety of data applications.
Keywords
robust
MM algorithm
variable selection
SVM
iteratively reweighted
GLM