92

does not require auxiliary matching information for

the alignment. In particular, our method can align

longitudinal data across heterogeneous subjects in a

common latent space to capture the dynamics of

shared patterns while utilizing temporal dependency

within subjects. Our numerical studies on both simulation settings and neuronal activity data indicate that

the proposed data integration approach improves prediction accuracy compared to existing machine learning methods.

Joint work with Yubai Yuan, Babak Shahbaba, Robert Fortin, Keiland Cooper and Qing Nie.

Estimating Heritability of Time-to-Event Traits

Using Censored Multiple Variance Component

Model

Jin Zhou

University of California, Los Angeles

Abstract: Genome-wide association studies (GWASs)

have spotlighted genetic variants linked to the onset

age for diseases such as Type 2 diabetes, Alzheimer's,

and heart disease. Central to GWASs is heritability,

which represents the proportion of phenotypic variation attributable to genetic variation. Historical approaches rooted in animal breeding are inadequate for

modern human genomic data. Alternatively, heritability relies on binary disease-status traits that fail to

capture the temporal aspect of disease development.

In this context, our research presents the censored multiple variance component model (CMVC)

based on an accelerated failure time model with

syntenic variables. Designed for individual-level genotype data, it's scalable for biobank data and handles

time-to-event data with random right-censoring. Simulations affirm its unbiased nature for right-censored

outcomes. We offer heritability assessments for various diseases, explore per-allele effect sizes across

genomic segments, and apply our methods to UK

Biobank.

Conclusively, our study introduces an advanced

method tailored for human genomic data, shedding

profound insights into the genetic determinants of

disease onset and progression.

Invited Session IS073: Statistical Interdisciplinary

Studies II

A New Statistic That Integrates Statistical Significance and Clinical Significance for Assessing the

Progression of Pulmonary Nodules

Jing Zhou

Renmin University of China

Abstract: Lung cancer remains one of the most prevalent malignant tumors worldwide. Nearly all lung

cancers evolve from pulmonary nodules, which are

the lesions that manifest as distinct \"spots\" within the

lung areas. For most patients with diagnosed pulmonary nodules, achieving a benign/malignant diagnosis

only based on baseline CT scans is particularly challenging. Therefore, adopting regular follow-up treatment becomes one of the crucial strategies in clinical

practice. In this study, we develop a novel statistic that

would be able to quantify the changes in nodules observed between baseline and follow-up CT scans for

one specific patient. Compared with previous studies,

we have three significant contributions. Firstly, the

proposed statisitc relies solely on 2D CT images for

calculation, circumvents the cumbersome process

required by traditional volumetric methods that necessitate precise delineation of the nodule’s 3D shape.

Additionally, we provided the empirical distribution

of this metric, enabling hypothesis testing for change

detection. Secondly, we proposed a novel technique

for generating invariant nodule samples based on

Gaussian random perturbation, overcoming the limitations of existing studies that could only simulate invariant nodules based on phantom, and breaking

through the bottleneck in mass-producing invariant

nodule samples. Lastly, to make this method more

aligned with clinical practice, we innovatively introduced the concept of clinical significance. This ensures that the assessment of nodule progression not

only considers statistical significance but also incorporates the clinical experience of physicians, making

the test statistic proposed in this paper more valuable

for clinical application.

Joint work with Hang Yu, Hansheng Wang, Ying Ji.

Correlation Trilogy of UWF-Based Myopia Predic-

93

tion: Copula, OU, and Large Model

Catherine Liu

The Hong Kong Polytechnic University

Abstract: The talk is to introduce a series of interdisciplinary work that are motivated from ophthalmic

practices. We tackle three challenges in myopia prediction via AI: (1) high conditional correlation between target responses, spherical equivalent (SE) and

axial length (AL); (2) interocular asymmetries and

high correlations of bilateral eye (OU) data; and (3)

overfitting of Vision Transformers (ViT) on small

datasets. Specifically, we aim to tell, i) Statistics empowers AI We introduce a novel Copula loss to

model the conditional correlation between bivariate

labels for both Regression- regression (RR) and Regression-Classification (RC) tasks, enhancing the

prediction capability of various backbone models. ii)

AI resolves interocular asymmetries We propose a

bi-channel multi-task architecture with a shared

non-frozen backbone and separate adapters. The association with a Multi-RR Copula loss sharply improves

the prediction accuracy. iii) ViT on small OU datasets We propose a novel framework that fuses

ViT, LoRa, and adapters together to overcome overfitting and allow for interocular asymmetries. Equipped

with a Multi-RC copula loss, the framework successfully enhances the prediction capability of large model

and thus achieves the highest accuracy for myopia

prediction.

Joint work with Factulty of Data Science and Eye &

ENT Hospital, Fudan University.

Powerful, Scalable and Resource-Efficient MetaAnalysis of Rare Variant Associations in Large

Whole Genome Sequencing Studies

Zilin Li

Northeast Normal University

Abstract: Meta-analysis of whole-genome/exome

sequencing (WGS/WES) studies provides an attractive solution to the problem of collecting large sample

sizes for discovering rare variants associated with

complex phenotypes. Existing rare variant meta-analysis approaches are not scalable to biobank-scale WGS data. Here we present MetaSTAAR,

a powerful and resource-efficient rare variant meta-analysis framework for large-scale WGS/WES

studies. MetaSTAAR accounts for relatedness and

population structure, can analyze both quantitative

and dichotomous traits, and boosts the power of rare

variant tests by incorporating multiple variant functional annotations. Through meta-analysis of four

lipid traits in 30,138 ancestrally diverse samples from

14 studies of the Trans-Omics for Precision Medicine

(TOPMed) Program, we show that MetaSTAAR performs rare variant meta-analysis at scale and produces

results comparable to using pooled data. Additionally,

we identified several conditionally significant rare

variant associations with lipid traits. We further

demonstrate that MetaSTAAR is scalable to biobank-scale cohorts through meta-analysis of TOPMed WGS data and UK Biobank WES data of

~200,000 samples.

Joint work with Xihao Li, Xihong Lin.

Construction and Application of Crime Prediction

Effectiveness Index

犯罪预测效果指数构建与应用

Yaofeng Zhang

Hubei University of Economics

摘要：针对现有犯罪预测评价指标的不足, 本文构

建了更为合理的犯罪预测效果指数(Prediction Effectiveness Index: PEI). 并通过对不同单元网格面

积和预测热点数下的主要犯罪预测评价指标进行

对比分析, 验证了犯罪预测效果指数的合理性. 在

此基础上, 提出了一种能提升犯罪预测效果的动态

阈值犯罪预测方法, 以 WH 市 2015 年 1 月

-2019 年 6 月 WH 市入室盗窃犯罪数据为实证案

例, 使用长短期记忆模型进行了测试验证. 实验结

果表明, 相较其他犯罪预测评价指标, 犯罪预测效

果指数考虑的因素更全面, 更合理. 本文提出的动

态阈值犯罪预测方法比传统固定阈值方法的实验

效果平均提升了 4.74%.

Joint work with Jinling Yao, Li Li, Zhilin Geng.

Towards Precision Oncology Discovery: Four Less

Known Genes and Their Unknown Interactions as

Highest-Performed Biomarkers for Colorectal

Cancer

94

Zhengjun Zhang

University of Chinese Academy of Sciences

Abstract: The goal of this study was to use a new

interpretable machine-learning framework based on

max-logistic competing risk factor models to identify

a parsimonious set of differentially expressed genes

(DEGs) that play a pivotal role in the development of

colorectal cancer (CRC). Transcriptome data from

nine public datasets were analyzed, and a new Chinese cohort was collected to validate the findings. The

study discovered a set of four critical DEGs - CXCL8,

PSMC2, APP, and SLC20A1 - that exhibit the highest

accuracy in detecting CRC in diverse populations and

ethnicities. Notably, PSMC2 and CXCL8 appear to

play a central role in CRC, and CXCL8 alone could

potentially serve as an early-stage marker for CRC.

This work represents a pioneering effort in applying

the max-logistic competing risk factor model to identify critical genes for human malignancies, and the

interpretability and reproducibility of the results

across diverse populations suggests that the four

DEGs identified can provide a comprehensive description of the transcriptomic features of CRC. The

practical implications of this research include the

potential for personalized risk assessment and precision diagnosis and tailored treatment plans for patients.

Joint work with Yongjun Liu, Yuqing Xu, Xiaoxing

Li, Mengke Chen, Xueqin Wang, Ning Zhang, Heping

Zhang and published on https://www.nature.com

/articles/s41698-024-00512-1.

Invited Session IS087: Data Science and Business

Intelligence Statistical Analysis

数据科学与商业智能统计分析

New Thinking and Development in Data Analysis

数据分析的新思维与新发展

Jianping Zhu

Xiamen University

Abstract: 本报告探讨了数据分析在当前技术进步

与多学科交叉融合背景下的创新性发展。报告共分

为四个部分，特别强调了包含统计学的交叉学科的

重要性及其在推动现代社会进步中的关键作用。首

先，报告回顾了大数据发展的历程，阐述了政府对

大数据发展的积极推进及战略支持。其次，讨论了

学科间的融合与创新，尤其是数据科学与其他学科

进行交叉融合，形成新的研究领域，展示了各高校

对交叉学科发展的重视，这一趋势为统计学的未来

提供新的机遇。接着，报告从十个关键方面深入探

讨了数据分析中的创新思维，突出了统计学在解决

现实问题中的重要作用和巨大潜力。最后，展示了

本团队在数据分析领域的研发工作，分享了在实际

应用中积累的经验与成果，强调了理论与实践结合

的重要性，并对未来数据分析技术的发展提出了前

瞻性展望。

From Specific Models to LLMs: Revolutionizing

Problem Solving with Generative AI

Jialin Hua

Jiangxi University of Finance and Economics

Abstract: The advent of Generative AI, particularly

Large Language Models (LLMs), has transformed the

landscape of problem-solving across diverse fields.

Traditionally, researchers focused on designing and

refining specific models tailored to particular problems. However, the paradigm has shifted towards

leveraging the expansive capabilities of LLMs to address a wide range of challenges. This talk explores

this transition, highlighting the advantages, limitations,

and practical implications of using LLMs compared to

traditional methods. We will delve into real-world

applications, illustrating how LLMs can streamline

processes and enhance decision-making. Additionally,

we will discuss future prospects and the evolving role

of Generative AI in problem-solving frameworks,

emphasizing its potential to drive innovation and efficiency across various industries.

Predicting Total Retail Sales of Consumer Goods

Based on Web Search Data and Deep Neural Networks

基于网络搜索数据和深度神经网络的社会消费品

零售总额预测研究

Kaiming Cheng

Zhejiang Gongshang University

摘要: 为弥补传统预测变量及预测技术的不足，本

文基于深度学习长期和短期时间序列网络(LSTNet)，

结合网络搜索数据与政府统计指标，构建

95

LSTNet&BI 模型开展浙江省及地级市社会消费品

零售总额的预测研究。研究发现：(1)引入网络搜索

数据能够有效提高 LSTNet 模型的预测性能与预测

精度；(2) LSTNet&BI 模型具有较好的泛化能力，

对浙江省社会消费品零售总额的短期和长期预测

效果较稳定，其预测性能与预测精度均优于其他基

准模型；(3) LSTNet&BI 模型具备较强的稳健性，

对杭州市、绍兴市和衢州市社会消费品零售总额的

预测效果也较好。

Joint work with Shucheng Liu.

A Study on Empowering High Quality Development and Efficiency Improvement Path of Industrial Industry with New Productivity: Empirical

Analysis Based on Random Forest Algorithm

新质生产力赋能工业行业高质量发展及其效率提

升路径研究—基于随机森林算法的实证分析

Min Zhang

Chongqing Technology and Business University

摘要: 基于新质生产力视角, 精准识别地方工业发

展态势和制定符合个体特征差异的提质增效优化

策略是推动全国高质量发展的内在要求和重要着

力点. 选取全国 31 个省份在 2003-2022 年期间的

39 个工业行业, 将新质生产力的新型劳动者、新型

劳动对象、新型劳动资料三要素与人力、教育、自

然、结构、生态、科技、数据、资本、制度等九大

要素禀赋相结合, 分别构建工业 62 个要素投入和

以规模以上工业企业主营业务收入作为产出的指

标体系. 采用 MissForest 算法填补缺失值. 通过

十折交叉验证, 比较传统线性回归、决策树、

bagging、随机森林的预测精度, 发现随机森林预测

效果最好. 因此, 采用随机森林算法精准预测和提

取变量之间的关系. 首先, 找出各地区各行业预测

产出值与实际产出值的差距, 同时结合各地区十三

五和四十五规划的核心行业, 精准识别各地区的非

充分发展行业、充分发展行业、不重要行业、要素

约束行业. 利用可解释的机器学习算法, 研究各个

行业各自重要的投入要素. 此外, 立足于全国频数

最高急迫需要改善的新兴产业的非充分发展行业,

运用产出对投入的部分依赖算法及其可视化, 探索

其各自最为重要的要素禀赋的动态效率优化路径,

以供各省份依托自身条件参考, 选择发展合适的工

业行业及其配置新质生产力投入要素.

Joint work with Linyu Zhang.

Invited Session IS013: Design and Modeling for

Computer Experiments

计算机试验的设计与建模

Robust Design for Order-of-Addition Experiments

Jianfeng Yang

Nankai University

Abstract: Order-of-addition experiment is a kind of

experiment that considers the addition order of ?

components. Finding an effective design is crucial

since the full design consists of ?! combinations

which are unaffordable even for computer simulation.

Several types of designs have been proposed for certain models in recent years, however they may be

singular under other models. In this paper, we propose

a kind of design called maximin distance component

orthogonal array whose distance is maximized while

the balance of any two columns from a component

orthogonal array still holds. A genetic algorithm is

used to search for such designs. The design, with excellent space-filling property, is robust since its ?

-optimality under the component-position model and

good performances under other models. Numerical

simulation and case study show the good performance

of the proposed design compared to the existing two

types of designs under different models.

Joint work with Yiran Huang.

Physical Parameter Calibration

Shifeng Xiong

Chinese Academy of Sciences

Abstract: Computer simulation models are widely

used to study complex physical systems. A related

fundamental topic is the inverse problem, also called

calibration, which aims at learning about the values of

parameters in the model based on observations. In

most real applications, the parameters have specific

physical meanings, and we call them physical parameters. To recognize the true underlying physical system, we need to effectively estimate such parameters.

However, existing calibration methods cannot do this

well due to the model identifiability problem. This

paper proposes a semi-parametric model, called the

discrepancy decomposition model, to describe the

96

discrepancy between the physical system and the

computer model. The proposed model possesses a

clear interpretation, and more importantly, it is identifiable under mild conditions. Under this model, we

present estimators of the physical parameters and the

discrepancy, and then establish their asymptotic properties. Numerical examples show that the proposed

method can better estimate the physical parameters

than existing methods.

Joint work with Yang Li.

Adaptive Grid Designs for Classifying Monotonic

Binary Computer Simulations

Xu He

Chinese Academy of Sciences

Abstract: This research is motivated by the

ice-breaking dynamic simulation, which is developed

to classify the successful conditions under which an

underwater vehicle will break through the ice. This

simulation is extremely time-consuming and the output of this simulation is deterministic, binary, and

monotonic. To accurately detecting the edge that separates the negative-outcome and positive-outcome

regions using as few simulation runs as possible, it is

crucial to use an efficient experimental design to determine the input values of simulation runs. Adaptive

designs that select input values according to obtained

outcomes are completely better than static designs

because a great proportion of design points of static

designs are redundant and can be omitted without

losing any information. In this paper, we propose a

new class of adaptive designs called adaptive grid

designs. An adaptive grid is a sequence of grids with

increasing resolution such that lower resolution grids

are proper subsets of higher resolution grids. From

carrying out simulation runs corresponding to lower

resolution points before higher resolution points and

skipping redundant runs, adaptive grid designs require

an order of magnitude less functional evaluations to

ensure certain level of classification accuracy than the

best possible static design and the same order of magnitude of evaluations with the best possible adaptive

design. We also provide numerical results on test

functions, the road crash simulation, and the

ice-breaking simulation to corroborate the superiority

of adaptive grid designs.

Joint work with Tian Bai, Dianpeng Wang and

Kuangqi Chen.

Construction of Orthogonal-Maxpro Latin Hypercube Designs

Yaping Wang

East China Normal University

Abstract: Orthogonal Latin hypercube designs

(LHDs) and maximum projection (MaxPro) LHDs are

widely used in computer experiments. They are efficient for estimating the trend part and the Gaussian

process part of the universal Kriging (i.e. the Gaussian

process) model, respectively, especially when only

some of the factors are active. Yet, the orthogonality

and the MaxPro criteria often do not agree with each

other. In this work, we propose a new class of optimal

designs, called orthogonal-MaxPro LHDs, optimizing

a well-defined multi-objective criterion combining the

correlation and the MaxPro metrics. An efficient parallel algorithm via level permutations and expansions

is developed, whose efficiency is guaranteed by theories. Numerical results are presented to show that the

construction is fast and the obtained designs are attractive, especially for large computer experiments.

Joint work with Sixu Liu, Qian Xiao.

Invited Session IS001: Advanced Estimation

Methods and Machine Learning

General Pairwise Comparison Models

Ruijian Han

The Hong Kong Polytechnic University

Abstract: Statistical inference using pairwise comparison data is an effective approach to analyzing

large-scale sparse networks. In this paper, we propose

a general framework to model the mutual interactions

in a network, which enjoys ample flexibility in terms

of model parametrization. Under this setup, we show

that the maximum likelihood estimator for the latent

score vector of the subjects is uniformly consistent

under a near-minimal condition on network sparsity.

This condition is sharp in terms of the leading order

asymptotics describing the sparsity. Our analysis uti-

97

lizes a novel chaining technique and illustrates an

important connection between graph topology and

model consistency. Our results guarantee that the

maximum likelihood estimator is justified for estimation in large-scale pairwise comparison networks

where data are asymptotically deficient. Simulation

studies are provided in support of our theoretical

findings.

Joint work with Yiming Xu, Kani Chen.

RMA: Ranking Based on Model Averaging

Baihua He

University of Science and Technology of China

Abstract: Ranking problems are commonly encountered in practical applications, including order priority

ranking, wine quality ranking, and piston slap noise

performance ranking. The responses of these ranking

applications are often considered as continuous responses and there is uncertainty on which scoring

function is used to model the responses. In this paper,

we address the scoring function uncertainty of continuous response ranking problems by proposing a

Ranking Model Averaging (RMA) method. With a set

of candidate models varied by scoring functions,

RMA assigns weights for each model determined by a

K-fold cross-validation criterion based on pairwise

loss. We provide two main theoretical properties for

RMA. First, we prove that the averaging ranking predictions of RMA are asymptotically optimal in

achieving the lowest possible ranking risk. Second,

we provide a bound on the difference between the

empirical RMA weights and theoretical optimal ones

and show that RMA weights are consistent. Simulation results validate RMA superiority over competing

methods in reducing ranking risk. Moreover, when

applied to empirical examples-order priority, wine

quality, and piston slap noise, RMA shows its effectiveness in building accurate ranking systems.

Joint work with Ziheng Feng.

Demographic Parity-Aware Individualized Treatment Rules

Wen Su

City University of Hong Kong

Abstract: There has been growing interest in developing advanced methodologies aimed at estimating

optimal individualized treatment rules (ITRs) in various fields, such as business decision-making, precision medicine, and social welfare distribution. The

application of ITRs within a societal context raises

substantial concerns regarding potential discrimination. Customized policies, learned from biased data,

can inadvertently lead to disparities based on sensitive

attributes such as age, gender, or race. To address this

concern directly, we introduce the concept of demographic parity (DP) in ITRs. However, estimating an

optimal ITR that satisfies the demographic parity

definition requires solving a non-convex constrained

optimization problem. To overcome these computational challenges, we employ tailored fairness proxies

inspired by DP and transform it into a convex quadratic programming problem. Additionally, we establish the consistency and convergence rate of the proposed estimator. The performance of the proposed

method is demonstrated through extensive simulation

studies and real data analysis.

Joint work with Wenhai Cui, Xiaodong Yan and

Xingqiu Zhao.

An Optimal Two-Step Estimation Approach for

Two-Phase Studies

Kin Yau Wong

The Hong Kong Polytechnic University

Abstract: Two-phase sampling is commonly adopted

for reducing cost and improving estimation efficiency.

In this project, we consider the two-phase study design where the outcome and some cheap covariates

are observed for a large cohort at Phase I, and expensive covariates are obtained for a selected subset of

the cohort at Phase II. As a result, the analysis of the

association between the outcome and covariates faces

a missing data problem. The complete case analysis

that uses only the Phase II sample is generally inefficient. We develop a two-step estimation approach,

which first obtains an estimator using the complete

data, and then updates it using an asymptotically

mean-zero estimator obtained from a working model

between the outcome and cheap covariates using the

98

full data. The two-step estimator is asymptotically at

least as efficient as the complete-data estimator and is

robust to misspecification of the working model. We

propose a kernel-based method to construct a two-step

estimator that achieves optimal efficiency, and also

develop a simple joint update approach based on multiple working models to approximate the optimal estimator. The proposed method is based on the influence function and is generally applicable as long as

the complete-data estimator is asymptotically linear.

We demonstrate the advantages of the proposed

method over the existing approaches via simulation

studies and provide applications to real biomedical

studies.

Joint work with Qingning Zhou.

Invited Session IS003: Advancements in Statistical

Inference of Point Processes and Their Applications

Filtering and Estimating the Renewal Hawkes

Process

Jiancang Zhuang

The Institute of Statistical Mathematics, Japan

Abstract: The Hawkes self-exciting models is one of

the most popular point-process model in many areas

in natural and social science because of its capacity of

investing the clustering effect and positive interactions

among individual events/particles. In recent years, this

model has been generalized to the renewal Hawkes

process, in which the background (immigrant) process

is a renewal process and the background events and

triggering events are not distinguishable from observation. This study first discusses the likelihood based

filtering techniques and corresponding statistical inference related to the renewal Hawkes process, and

then explores the Markov-Chain Monte-Carlo based

inferences methods. The outputs are connected to and

compared with those by Stindl and Chen (2018,

CSDA; 2022, JRSSC; 2023, AOAS).

Nonparametric Second-Order Estimation for Spatiotemporal Point Patterns

Jialing Liu

Sun Yat-sen University

Abstract: Many existing methodologies for analyzing

spatiotemporal point patterns are developed based on

the assumption of stationarity in both space and time

for the second-order intensity or pair correlation. In

practice, however, such an assumption often lacks

validity or proves to be unrealistic. In this paper, we

propose a novel and flexible nonparametric approach

for estimating the second-order characteristics of spatiotemporal point processes, accommodating

non-stationary temporal correlations. Our proposed

method employs kernel smoothing and effectively

accounts for spatial and temporal correlations differently. Under a spatially increasing-domain asymptotic

framework, we establish consistency of the proposed

estimators, which can be constructed using different

first-order intensity estimators to enhance practicality.

Simulation results reveal that our method, in comparison with existing approaches, significantly improves

statistical efficiency. An application to a COVID-19

dataset further illustrates the flexibility and interpretability of our procedure.

Joint work with Decai Liang and Yongtao Guan.

Fitting Multivariate Hawkes Process with Interval

Count Data

Feng Chen

The University of New South Wales

Abstract: The multivariate Hawkes process is a

prominent model for analyzing multi-type event sequences, wherein events exhibit intra- and inter-type

excitation phenomena. While fitting this process with

complete event sequence data comprising exact occurrence times and type labels of all events within the

observation time window can be accomplished

through direct likelihood maximization or the EM

algorithm, challenges arise when dealing with interval

count data of different event types in regular intervals.

In this study, we introduce a novel method to estimate/approximate the likelihood of the multivariate

Hawkes process relative to interval count data, along

with two parameter estimation techniques based on

this likelihood estimator. The likelihood approximation leverages importance sampling, while the parameter estimation methods involve numerical maximiza-

99

tion of the approximate (log-)likelihood and Markov

Chain Monte Carlo, respectively. We evaluate the

performance of these proposed parameter estimation

methods via simulation experiments and compare

them with a recently published alternative method.

Results demonstrate that the proposed estimators exhibit smaller mean square errors than the benchmark

method. Furthermore, we apply these estimators to

analyze two real-life datasets one on global terrorist

activities and the other focusing on youth suicides in

Hong Kong. Our findings highlight the ease of implementation and efficacy of the proposed methods in

practical scenarios.

Bayesian Nonparametric Learning for Point Processes with Spatial Homogeneity: A Spatial Analysis of NBA Shot Locations

Guanyu Hu

University of Texas Health Science Center at Houston

Abstract: Basketball shot location data provide valuable summary information regarding players to

coaches, sports analysts, fans, statisticians, as well as

players themselves. Represented by spatial points,

such data are naturally analyzed with spatial point

process models. We present a novel nonparametric

Bayesian method for learning the underlying intensity

surface built upon a combination of Dirichlet process

and Markov random field. Our method has the advantage of effectively encouraging local spatial homogeneity when estimating a globally heterogeneous

intensity surface. Posterior inferences are performed

with an efficient Markov chain Monte Carlo (MCMC)

algorithm. Simulation studies show that the inferences

are accurate and the method is superior compared to a

wide range of competing methods. Application to the

shot location data of 20 representative NBA players in

the 2017-2018 regular season offers interesting insights about the shooting patterns of these players. A

comparison against the competing methods shows that

the proposed method can effectively incorporate spatial contiguity into the estimation of intensity surfaces.

Invited Session IS007: Asymptotic Theory and

High-Dimensional Statistics

High-Dimensional Bootstrap and Asymptotic Expansion

Yuta Koike

The University of Tokyo

Abstract: The recent seminal work of Chernozhukov,

Chetverikov and Kato has shown that bootstrap approximation for the maximum of a sum of independent random vectors is justified even when the dimension is much larger than the sample size. In this context, numerical experiments suggest that third-moment

match bootstrap approximations would outperform

normal approximation even without studentization,

but the existing theoretical results cannot explain this

phenomenon. In this talk, we show that Edgeworth

expansion, if justified, can give an explanation for this

phenomenon. In particular, we derive an asymptotic

expansion formula of the bootstrap coverage probability and show that the third-moment match wild bootstrap is second-order accurate in high-dimensions

even without studentization when the covariance matrix has identical diagonal entries and bounded eigenvalues. In addition, we show the validity of the asymptotic expansion when appropriate random vectors

have Stein kernels.

Central Limit Theorem in Exponential Graphs

指数图中心极限定理

Songhao Liu

Southern University of Science and Technology

Abstract: The question of whether the central limit

theorem (CLT) holds for the total number of edges in

exponential random graph models (ERGMs) in the

subcritical region of parameters has remained an open

problem. In this paper, we establish the CLT in a subset of the subcritical region known as Dobrushin’s

uniqueness region. As a result of our proof, we also

derive a convergence rate for the CLT and an explicit

formula for the asymptotic variance. To establish our

main result, we develop Stein’s method for the normal

approximation for general functionals of nonlinear

exponential families of random variables, which is of

independent interest. In addition to ERGM, our general theorem can also be applied to other models.

Joint work with Xiao Fang and Qiman Shao.

100

Large Dimensional Robust Spiked Covariance

Matrix and Order Estimation

Mengxi Yi

Beijing Normal University

Abstract: This presentation delves into the behaviors

of robust scatter estimators within the intricate landscapes of large-dimensional datasets, compounded by

information and impulsive noise. It is shown that the

robust estimator of scatter has bounded spectrum and

may contain isolated eigenvalues, as opposed to the

sample covariance matrix. From these insights, we

develop innovative estimation procedures for eigenvalues and eigenvectors. Additionally, we introduce a

robust and consistent estimator to precisely identify

the number of spikes. Empirical validation through

numerical studies will illustrate the finite-sample performance, comparing favorably against existing

methods. Furthermore, we'll discuss a real-world application to underscore the practical significance of

our approach.

Double Shrinkage Priors for a Normal Mean Matrix

Takeru Matsuda

The University of Tokyo & RIKEN Center for Brain

Science

Abstract: We consider estimation of a normal mean

matrix under the Frobenius loss. Motivated by the

Efron-Morris estimator, a generalization of Stein's

prior has been recently developed, which is superharmonic and shrinks the singular values towards zero.

The generalized Bayes estimator with respect to this

prior is minimax and dominates the maximum likelihood estimator. However, here we show that it is inadmissible by using Brown's condition. Then, we

develop two types of priors that provide improved

generalized Bayes estimators and examine their performance numerically. The proposed priors attain risk

reduction by adding scalar shrinkage or column-wise

shrinkage to singular value shrinkage. Parallel results

for Bayesian predictive densities are also given.

Invited Session IS092: Statistical Network Analysis

and Its Application

Modularity Based Methods for Network Data

Yuguo Chen

University of Illinois Urbana-Champaign

Abstract: We introduce several network modularity

measures for both single-layer and multi-layer networks under different null models of the network,

motivated by empirical observations in networks from

a diverse field of applications. We describe a statistical framework for modularity-based network community detection. A hypothesis testing procedure is also

proposed for selecting an appropriate null model for

data. These null models are used to define modularity

measures as well as model likelihood based quality

functions. The proposed measures are then optimized

to detect the optimal community assignment of nodes.

The effectiveness of the measures is demonstrated in

simulated networks and real networks.

Distribution-Free Matrix Prediction under Arbitrary Missing Pattern

Yuan Zhang

The Ohio State University

Abstract: This paper studies the open problem of

conformalized entry prediction in a row/column- exchangeable matrix. The matrix setting presents novel

and unique challenges, but there exists little work on

this interesting topic. We meticulously define the

problem, differentiate it from closely related problems,

and rigorously delineate the boundary between

achievable and impossible goals. We then propose two

practical algorithms. The first method provides a fast

emulation of the full conformal prediction, while the

second method leverages the technique of algorithmic

stability for acceleration. Both methods are computationally efficient and can effectively safeguard coverage validity in presence of arbitrary missing pattern.

Further, we quantify the impact of missingness on

prediction accuracy and establish fundamental limit

results. Empirical evidence from synthetic and real-world data sets corroborates the superior performance of our proposed methods.

Joint work with Meijia Shao.

101

ReHLine: Regularized Composite ReLU-ReHU

Loss Minimization with Linear Computation and

Linear Convergence

Ben Dai

The Chinese University of Hong Kong

Abstract: Empirical risk minimization (ERM) is a

crucial framework that offers a general approach to

handling a broad range of machine learning tasks. In

this paper, we propose a novel algorithm, called

ReHLine, for minimizing a set of regularized ERMs

with convex piecewise linear-quadratic loss functions

and optional linear constraints. The proposed algorithm can effectively handle diverse combinations of

loss functions, regularization, and constraints, making

it particularly well-suited for complex domain-specific problems. Examples of such problems

include FairSVM, elastic net regularized quantile

regression, Huber minimization, etc. In addition,

ReHLine enjoys a provable linear convergence rate

and exhibits a per-iteration computational complexity

that scales linearly with the sample size. The algorithm is implemented with both Python and R interfaces, and its performance is benchmarked on various

tasks and datasets. Our experimental results demonstrate that ReHLine significantly surpasses generic

optimization solvers in terms of computational efficiency on large-scale datasets. Moreover, it also outperforms specialized solvers such as liblinear in

SVMs, hqreg in Huber minimization and lightning(SAGA, SAG, SDCA, SVRG) in smooth SVMs,

exhibiting exceptional flexibility and efficiency.

Joint work with Yixuan Qiu.

Efficient Estimation for Longitudinal Networks via

Adaptive Merging

Haoran Zhang

Southern University of Science and Technology

Abstract: Longitudinal network consists of a sequence of temporal edges among multiple nodes,

where the temporal edges are observed in real time. It

has become ubiquitous with the rise of online social

platform and e-commerce, but largely under- investigated in literature. In this paper, we propose an efficient estimation framework for longitudinal network,

leveraging strengths of adaptive network merging,

tensor decomposition and point process. It merges

neighboring sparse networks so as to enlarge the

number of observed edges and reduce estimation variance, whereas the estimation bias introduced by network merging is controlled by exploiting local temporal structures for adaptive network neighborhood. A

projected gradient descent algorithm is proposed to

facilitate estimation, where the upper bound of the

estimation error in each iteration is established. A

thorough analysis is conducted to quantify the asymptotic behavior of the proposed method, which

shows that it can significantly reduce the estimation

error and also provides guideline for network merging

under various scenarios. We further demonstrate the

advantage of the proposed method through extensive

numerical experiments on synthetic datasets and a

militarized interstate dispute dataset.

Joint work with Junhui Wang.

Invited Session IS009: Complex Data, Geometry

and Related Fields

Regression Analysis of Neuroimaging Data Leveraging Deep Neural Networks

Jian Kang

University of Michigan

Abstract: The complex interplay between neuroimaging data and variables of interest poses significant

challenges for conventional regression models. These

challenges are due to the ultra-high dimensionality,

varying levels of noise, and limited sample sizes inherent to this type of data. In this talk, I will introduce

a series of regression models specifically designed for

neuroimaging data analysis, utilizing Deep Neural

Networks (DNN) to enable more accurate statistical

inference. Unlike traditional approaches, our innovative methods offer enhanced flexibility in capturing

intricate patterns in brain activity, accommodating the

heterogeneity in noise levels and spatial dependencies

across different brain regions. I will delve into parameter estimation, inference procedures, and the theoretical underpinnings of our advanced models, ultimately

demonstrating their superior performance over exist-

102

ing methodologies through rigorous simulations and

real-world neuroimaging case studies.

Interpretable Super-Resolution Dimension Reduction of Spatial Transcriptomics Data by DeepFuseNMF

Ruibin Xi

Peking University

Abstract: Spatial transcriptomic (ST) technologies

enable measurements of gene expression across thousands of spots from a tissue slice while preserving

location information of the spots. ST spot locations

are spatially sparse, but the coupled image data are of

super-resolution. Using a deep neural network guided

by non-negative matrix factorization, we develop a

computational method named DeepFuseNMF that can

integrate ST expression and image data, and achieve

interpretable super-resolution dimension reduction as

well as super-resolution spatial domain detection and

gene expression recovery. In applications to mouse

brain and tumor ST datasets, we demonstrate that

embedding dimensions given by DeepFuseNMF can

faithfully recover spatial and molecular patterns of the

profiled tissue, and find that DeepFuseNMF can identify fine spatial structures that are difficult for available methods.

Algorithms for Ridge Estimation with Convergence Guarantees

Wanli Qiao

George Mason University

Abstract: The extraction of filamentary structure

from a point cloud is discussed. The filaments are

modeled as ridge lines or higher dimensional ridges of

an underlying density, which are low-dimensional

structures where the density is higher than in the surrounding area when moving away from the set in an

orthogonal direction. We propose two novel algorithms, and provide theoretical guarantees for their

convergences, by which we mean that the algorithms

can asymptotically recover the full ridge set. We consider the new algorithms as alternatives to the existing

Subspace Constrained Mean Shift (SCMS) algorithm

for which no such theoretical guarantees are known.

Joint work with Wolfgang Polonik.

Manifold Fitting: The Next Frontier in Single-Cell

RNA Clustering and Its Potential Applications

Bingjie Li

National University of Singapore

Abstract: Single-cell RNA sequencing (scRNA-seq)

has emerged as a powerful tool for exploring cellular

heterogeneity and unraveling the complexities of disease pathogenesis. However, the analysis of

scRNA-seq data presents significant challenges due to

technical variability, high-dimensional data complexity, and inherent biological noise. Existing scRNA-seq

analysis methods often struggle to provide accurate

and robust clustering results. To address these limitations, we introduce a ground breaking framework that

leverages manifold fitting techniques. In contrast to

conventional approaches that focus on dimensionality

reduction, our method directly fits a low-dimensional

manifold within the high-dimensional ambient space

and subsequently unfolds the data along this manifold.

By reducing the distances between cells of similar

types while preserving the richness of gene expression

information, our approach achieves superior

scRNA-seq clustering and visualization compared to

current state-of-the-art methods. This innovative

framework sets a new benchmark in scRNA-seq analysis, paving the way for transformative insights into

cellular diversity and opening up exciting possibilities

for future research and clinical applications.

Joint work with Zhigang Yao, Yukun Lu, Shing-Tung

Yau.

Invited Session IS067: Stastistical Analysis with

Complex Data

Local Clustering for Functional Data

Qingzhao Zhang

Xiamen University

Abstract: In functional data analysis, unsupervised

clustering has been extensively conducted and has

important implications. In most of the existing functional clustering analyses, it is assumed that there is a

single clustering structure across the whole domain of

measurement (say, time interval). In some data anal-

103

yses, for example, the analysis of normalized

COVID-19 daily confirmed cases for the U.S. states,

it is observed that functions can have different clustering patterns in different time subintervals. To tackle

the lack of flexibility of the existing functional clustering techniques, we develop a local clustering approach, which can fully data-dependently identify

subintervals, where, in different subintervals, functions have different clustering structures. It is built on

the popular basis expansion technique and has a novel

penalization form. It simultaneously achieves subinterval identification, clustering, and estimation. Its

estimation and clustering consistency properties are

rigorously established. In simulation, it significantly

outperforms multiple competitors. In the analysis of

the COVID-19 case trajectory data, it identifies sensible subintervals and clustering structures.

Joint work with Yuanxing Chen, Qingzhao Zhang,

Shuangge Ma, Yuanxing Chen.

Gaussian Approximation for Thresholding Statistics

Yumou Qiu

Peking University

Abstract: Thresholding statistics that sum the

thresholded standardized statistics over many components are more powerful than the sum-of-square type

and maximum type test statistics in detecting sparse

and weak signals for global hypotheses. However, the

asymptotic distribution of the thresholding statistics

has only been derived under the assumption of independent variables or certain conditions on the mixing

dependence among variables. Gaussian approximation

result is established for the thresholding statistics

under general covariance structures and

high-dimensionality. Due to the non-smoothness of

the thresholding function, existing techniques to show

Gaussian approximation results for the sum-of-square

and maximum statistics can not be applied. A novel

method has been developed to establish the Gaussian

approximation results for thresholding statistics.

Based on this result, a bootstrap procedure is constructed to approximate the distribution of the thresholding statistics under a high-dimensional setting.

Simulation studies are conducted to show the utility of

the proposed approach.

Distribution-Free Prediction Intervals under Covariate Shift, with an Application to Causal Inference

Yukun Liu

East China Normal University

Abstract: Owing to its appealing distribution-free

feature, conformal inference has become a popular

tool for constructing prediction intervals with a desired coverage rate. In scenarios involving covariate

shift, where the shift function needs to be estimated

from data, many existing methods resort to data-splitting techniques. However, these approaches

often lead to wider intervals and less reliable coverage

rates, especially when dealing with finite sample sizes.

To address these challenges, we propose methods

based on a pivotal quantity derived under a parametric

working model and employ a resampling-based

framework to approximate its distribution. The

resampling-based approach can produce prediction

intervals with a desired coverage rate without splitting

the data and can be easily applied to causal inference

settings where a shift in the covariate distribution can

occur between treatment and control arms. Additionally, the proposed approaches enjoy a double robustness property and are adaptable to different prediction

tasks. Our extensive numerical experiments demonstrate that, compared to existing methods, the proposed novel approaches can produce substantially

shorter conformal prediction intervals with lower

variability in the interval lengths while maintaining

promising coverage rates and advantages in versatile

usage.

Knockoff-Based Statistics for the Identification of

Putative Causal Genes in Genetic Studies

Shiyang Ma

Shanghai Jiao Tong University

Abstract: Gene-based tests are important tools for

elucidating the genetic basis of complex traits. Despite

substantial recent efforts in this direction, the existing

tests are still limited, owing to low power and detec-

104

tion of false-positive signals due to the confounding

effects of linkage disequilibrium. In this paper, we

describe a gene-based test that attempts to address

these limitations by incorporating data on long-range

chromatin interactions, several recent technical advances for region-based testing, and the knockoff

framework for synthetic genotype generation.

Through extensive simulations and applications to

multiple diseases and traits, we show that the proposed test increases the power over state-of-the-art

gene-based tests and provides a narrower focus on the

possible causal genes involved at a locus. We apply

BIGKnock to the UK Biobank data with 405,296

participants for multiple binary and quantitative traits,

and show that relative to conventional gene-based

tests, BIGKnock produces smaller sets of significant

genes that contain the causal gene(s) with high probability.

Invited Session IS102: Statistical Measurement,

Evaluation and Decision（统计测度、评价与决策

分会场）

Interval-Valued Functional Clustering Based on

the Wasserstein Distance Considering the Interval

Distribution Information with Application to Air

Quality Data

Lirong Sun

Zhejiang Gongshang University

Abstract: Interval-valued functional data is a special

kind of symbolic data in which each value in a functional dataset is represented by an interval-valued data

entry. Interval-valued functional clustering combines

interval-valued clustering and functional clustering,

and the existing research on interval-valued data

characteristic extraction methods has been carried out

under the uniform distribution. However, the uniform

distribution may not adequately represent the interval

distribution, and consequently the clustering results

obtained may be distorted, Therefore, our paper utilizes the empirical distribution function to approximately estimate the interval distribution to improve

the information utilization of interval distribution

within interval- valued functional data. We propose a

novel distance metric termed the interval-valued functional Wasserstein distance under the empirical distribution function and a clustering method based on this

metric with the improved K-means clustering algorithm. Finally, the proposed clustering method is applied to simulated data and multi-index air quality

data, and the effectiveness and superiority of the proposed metric is demonstrated.

Research on the Measurement of Income Distribution of Data Value Chain Balance Fairness and

Efficiency

Wenjin Zuo

Shanghai University of Finance and Economics

Zhejiang College

Abstract: Scientific measurement of data value is the

basis of sustainable development of data resource

market. The income distribution measurement of data

value chain, which balance the fairness and efficiency,

challenges the traditional measurement methods. In

order to solve the above problems, this paper combines the fair efficiency coefficient, the minimum

variance and optimization methods to improve the

classical bankruptcy distribution model, and then puts

forward two measurement methods of data value

chain income distribution that balance the fairness and

efficiency. Theoretical and case analysis show that the

two new methods balance the influence of fairness

and efficiency, improve the performance of classical

bankruptcy allocation methods, and provide an effective tool for scientific measurement of data value. The

new methods proposed in this study enrich the system

of data value measurement.

Multidimensional Cloud Model Based Assessment

and its Application to the Risk of Supply Chain

Financial Companies

Jinming Zhou

Anhui Polytechnic University

Abstract: The multidimensional cloud model is proposed as the expansion of the one-dimensional cloud

model. The features of ambiguity and stochasticity in

complex information situations are considered, thus,

this optimized model can be utilized upon multiple

value classifications and ordering via which the ob-

105

jects’ attributes of physical and social can be reflected.

Therefore, this promoted model is widely used. This

paper provides a knowledge graph by reviewing the

theoretical research of the multidimensional cloud

model and its related bibliographies, and CiteSpace is

applied here to give a visualization conclusion. In

recent years, a multitude of theories and methods have

emerged to address the challenges posed by fuzzy and

stochastic uncertainty in various domains, such as

image segmentation, data mining, prediction techniques, and comprehensive evaluation of multiple

metrics and dimensions using uncertain linguistic

variables.

Integrated Optimization of Kernel Correlation

Filtering Algorithm Based on the Benefit of Doubt

Method

Sibo Chen

Anhui Polytechnic University

Abstract: The kernel correlation filtering algorithm is

an efficient and accurate video stream analysis tool,

which is widely used in dynamic object tracking and

identification scenarios such as security monitoring,

driving assistance, and remote medical assistance.

However, since its training process requires the construction of positive and negative samples through a

circulant matrix, when environmental factors such as

light and humidity change, the collection and construction process of positive and negative samples in

the video stream will be altered correspondingly, resulting in tracking failure. To address this problem,

this study proposed a benefit of doubt method based

on multi-environment sampling to optimize the kernel

correlation filtering algorithm. By summarizing samplers in multiple simulated environments such as different color temperature, contrast, and clarity a relatively stable and accurate tracking judgment result is

given. The results show that the kernel correlation

filtering algorithm optimized by the benefit of doubt

method can effectively reduce the environmental impact of the kernel correlation filtering algorithm and

improve the stability and accuracy of video stream

discrimination.

Contributed Session CS021: Matrix Theory and

Sufficient Dimension Reduction

Multi Angle Subgroup Ensemble Learning Based

on Tensor Decomposition

基于张量分解的多角度亚组集成学习

Xun Zhao

Southwestern University Of Finance And Economics

摘要: 为学习多方向交叉产生的亚组结构, 本文提

出一种新的基于张量分解的多角度亚组集成方法.

较传统单一角度的亚组分析, 我们得到的亚组内部

有更强的同质性, 因此在保留解释性的同时, 有更

高的预测精度. 进一步, 结合两两判罚函数和

ADMM 工具, 我们发展有理论保证的快速算法估

计相应参数. 我们给出了估计量的大样本性质, 从

理论上保证方法的合理性. 数值模拟和健康养老数

据分析进一步展示了我们提出的方法在估计的准

确度和亚组识别上显著优于现有方法, 并且发现隔

代照料会减轻四川、广东等 21 个区域的老年人患

中风、认知类疾病的风险, 但是会增加与饮食及劳

累相关类疾病（如三高、胃部疾病、关节炎）的患

病风险.

Joint work with Ling Zhou, Weijia Zhang and

Huazhen Lin.

Dynamic Matrix Recovery 动态矩阵恢复

Ziyuan Chen

Peking University

Abstract: Matrix recovery from sparse observations

is an extensively studied topic emerging in various

applications, such as recommendation system and

signal processing, which includes the matrix completion and compressed sensing models as special cases.

In this article, we propose a general framework for

dynamic matrix recovery of low-rank matrices that

evolve smoothly over time. We start from the setting

that the observations are independent across time, then

extend to the setting that both the design matrix and

noise possess certain temporal correlation via modified concentration inequalities. By pooling neighboring observations, we obtain sharp estimation error

bounds of both settings, showing the influence of the

underlying smoothness, the dependence and effective

samples. We propose a dynamic fast iterative shrinkage-thresholding algorithm that is computationally

106

efficient, and characterize the interplay between algorithmic and statistical convergence. Simulated and

real data examples are provided to support such findings.

Joint work with Ying Yang and Fang Yao.

An Improved Principal Hessian Directions Method

for Dimension Reduction

Zheng Li

Northeast Normal University

Abstract: Sufficient dimension reduction (SDR) is a

valuable approach for analyzing high-dimensional

data. In this paper, we present a novel SDR technique

termed Improved Principal Hessian Directions (IPHD)

for estimating the central subspace of the response

variable Y given the vector of predictor variables X.

The computational algorithm for implementing the

proposed IPHD method is provided. We also establish

the consistency and asymptotic properties of the IPHD

method. Monte Carlo simulation is used to evaluate

the performance and demonstrate the efficiency and

robustness of the proposed method. When the variance of each component becomes more dissimilar, the

IPHD performs well in scenarios of the elliptical distribution.

Joint work with Wei Gao.

A Novel Method for Synthetic Controls with Interference

Peiyu He

Peking University

Abstract: Synthetic control methods have been

widely used for causal inference in comparative case

studies, which have typically assumed no interference

on the untreated units. However, in many studies,

there is suspect that interference may arise and jeopardize causal inference. In this paper, we propose a

novel synthetic control method that allows for interference. We establish consistent and asymptotically

normal estimation and inference of the direct and

interference effects averaged over post-intervention

periods. Identification of the effects is achieved under

the assumption that the number of interfered units is

roughly less than half of the total units. Our identification strategy does not require to model the interference structure. We evaluate the performance of our

method and compare it to existing methods by simulations. Application to the analysis of Middle Eastern

conflict data, where the intervention is the relocation

of the US embassy to Jerusalem, shows evidence that

this intervention not only increases the number of

conflicts in Israel-Palestine but also has interference

effects on several other Middle Eastern countries.

Joint work with Yilin Li and Wang Miao.

Identifying Causal Effects Using Instrumental

Variables From the Auxiliary Dataset

Kang Shuai

Peking University

Abstract: Instrumental variable approaches have

gained popularity for estimating causal effects in the

presence of unmeasured confounders. However, the

availability of instrumental variables in the primary

dataset is often challenged due to stringent and untestable assumptions. This paper presents a novel

method to identify and estimate causal effects by utilizing instrumental variables from the auxiliary dataset,

incorporating a structural equation model, even in

scenarios with nonlinear treatment effects. Our approach involves using two datasets: one called the

primary dataset with joint observations of treatment

and outcome, and another auxiliary dataset providing

information about the instrument and treatment. Our

strategy differs from most existing methods by not

depending on the simultaneous measurements

of instrument and outcome. The central idea for identifying causal effects is to establish a valid substitute

through the auxiliary dataset, addressing unmeasured

confounders. This is achieved by developing a control

function and projecting it onto the function space

spanned by the treatment variable. We then propose a

three-step estimator for estimating causal effects and

derive its asymptotic results. We illustrate the proposed estimator through simulation studies, and the

results demonstrate favorable performance. We also

conduct a real data analysis to evaluate the causal

effect between vitamin D status and body mass index.

107

Contributed Session CS022: Asymptotic Theory in

Probability and Statistics

Stochastic Dynamical Analysis for Renewable

Power System Excited by Multiplicative Fractional

Gaussian Noise Based on Graph Neural Network

Hufei Li

North Minzu University

Abstract: This paper focuses on the influence of multiplicative fractional Gaussian noise (MFGN) on renewable power systems. Firstly, based on the historical data of power system, the long-range correlation in

power system is found, and then the power system

model excited by MFGN is established. Secondly, we

derive the Fokker-Planck-Kolmogorov (FPK) equation for the proposed model. Thirdly, a novel deep

learning method based on graph neural network is

proposed to solve the corresponding time-dependent

FPK equation. Finally, several illustrative examples

are carried out in detail to verify the effectiveness and

feasibility of the proposed method. Besides, some

suggestions are put forward to improve the dynamical

characteristics of power system under random disturbance.

Joint work with Shaojuan Ma.

Limit Theorems for Continuous State Nonhomogeneous Markov Chain

Chengjun Ding

Yunnan University

Abstract: Continuous state nonhomogeneous Markov

chains are widely used to model the performance of

random variables continuously varied over time in

many fields such as population biology. Existing

works mainly focus on their strong law of large numbers. There is little work developed on their limit theorems. To this end, this paper investigates the limiting

properties of continuous state nonhomogene- ous

Markov chains, and establishes limit theorems for

multivariate functions of continuous state nonhomogeneous Markov chains, including the strong law of

lager numbers, the central limit theorem and almost

sure central limit theorem under some mild conditions,

which are some basic theoretical properties for statistical inference and predictions of continuoustime-varying random variables.

Green Matching: An Efficient Parameter Estimation Method for Complex Dynamical Systems

格林匹配：复杂动力系统的一种高效参数估计方法

Guoyu Zhang

Peking University

Abstract: Parameters of differential equations are

essential to characterize intrinsic behaviors of dynamic systems. Numerous methods for estimating parameters in dynamic systems are computationally and/or

statistically inadequate, especially for complex systems with general-order differential operators, such as

motion dynamics. This article presents Green's

matching, a computationally tractable and statistically

efficient two-step method, which only needs to approximate trajectories in dynamic systems but not

their derivatives due to the inverse of differential operators by Green's function. This yields a statistically

optimal guarantee for parameter estimation in general-order equations, a feature not shared by existing

methods, and provides an efficient framework for

broad statistical inferences in complex dynamic systems.

Joint work with Jianbin Tan, Xueqin Wang, Hui

Huang and Fang Yao.

Stratified Permutational Berry-Esseen Bounds

Pengfei Tian

Tsinghua University

Abstract: Stratification, commonly known as blocking, is one of the fundamental tools employed in sampling, experimental design, and inference. Many estimators used in these contexts can be expressed as

permutational statistics. Despite the prevalence of

stratification in practice, there is still a lack of

non-asymptotic results for stratified randomization,

which hinders comprehensive understanding and applications. We aim to bridge this gap. First, we extend

the existing Berry-Esseen bound to accommodate

various stratification regimes. But the Berry-Esseen

bound can not cover all possible stratification regimes.

Then leveraging Stein's method, we construct a versa-

108

tile Berry-Esseen bound capable of addressing the

diverse scenarios arising from stratification, ranging

from numerous small strata to a few large ones, as

well as their mixtures. This result can induce the classical non-stratified Berry-Esseen bound. In the proof,

we merge the method of two existing approaches to

deal with non-stratified Berry-Esseen bound, because

of the complexity of stratification. Furthermore, we

demonstrate the practical utility of our theory by applying it to stratified sampling, experimental designs,

post-stratification, and randomization inference with

an instrumental variable. These applications underscore the efficacy of our approach and contribute to

bridging the theoretical gap in the field.

Joint work with Fan Yang and Peng Ding.

Using Synthetic Data to Regularize Maximum

Likelihood Estimation

Weihao Li

National University of Singapore

Abstract: To overcome challenges in fitting complex

models with small samples, catalytic priors have recently been proposed to stabilize the inference by

supplementing observed data with synthetic data generated from simpler models. Based on a catalytic prior,

the Maximum A Posteriori (MAP) estimator is a regularized estimator that maximizes the weighted likelihood of the combined data. This estimator is

straightforward to compute, and its numerical performance is superior or comparable to other likelihood-based estimators. In this paper, we study several

theoretical aspects regarding the MAP estimator in

generalized linear models, with a particular focus on

logistic regression. We first prove that under mild

conditions, the MAP estimator exists and is stable

against the randomness in synthetic data. We then

establish the consistency of the MAP estimator when

the dimension of covariates diverges slower than the

sample size. Furthermore, we utilize the convex

Gaussian min-max theorem to characterize the asymptotic behavior of the MAP estimator as the dimension grows linearly with the sample size. These

theoretical results clarify the role of the tuning parameters in a catalytic prior, and provide insights in practical applications. We provide numerical studies to

confirm the effective approximation of our asymptotic

theory in finite samples and to illustrate adjusting

inference based on the theory.

Joint work with Dongming Huang.

Contributed Session CS023: Recent Advances in

Graphical Models and Image

Empowering Mental Health Insights: The Synergy

of Knowledge Graphs and Large Language Models

Shan Gao

Yunnan University

Abstract: Mental health has emerged as a major

global health concern, with increasing research efforts

in recent years. However, key findings are often scattered across various scientific studies and databases,

posing challenges for comprehensive analysis and

in-depth understanding of mental disorders. Here, we

introduce a novel Mental Disorders Knowledge Graph

(MDKG), a comprehensive, multi-relational platform

aggregating dispersed research findings into a unified

analytical framework. Our methodology involves the

innovative use of large language models, such as

ChatGPT and GPT-4.0, to streamline construction and

minimize manual effort, while incorporating contextual features like conditional statements, demographics, and additional contexts to ensure data

reliability. Extending beyond the unstructured text

data of the Pubmed abstract, we integrated triplets

from 12 leading biomedical databases, enhancing the

breadth and impartiality of the literature-based

MDKG. The resulting MDKG boasts over 10 million

triples, with more than 1.73 million derived from

mental health-related medical literature, unveiling

0.92 million unique insights not previously captured

in existing databases. We evaluate our MDKG's effectiveness by predicting major depressive disorder in the

UK Biobank cohort and comparing its mental health

information richness against other existing biomedical

knowledge graphs. This work signifies a step forward

in the data-driven exploration of mental disorders,

offering new avenues for understanding and intervention.

109

Joint work with Hongtu Zhu, Niansheng Tang,

Kaixian Yu, Yue Yang, Sheng Yu, Chenglong Shi,

Xueqin Wang.

Total Variation Regularized Multi-Matrices

Weighted Schatten p-norm Minimization for Image Denoising

Zuoxun Tan

Chongqing University

Abstract: Motivated by the superior performance of

nonconvex nonsmooth Lp (0 < p < 1) norm, this paper

introduces a novel method that combines the weighted

Schatten p-norm, Lp-norm, and total variation regularization based on the multiple matrices denoising

framework. The weighted Schatten p-norm encodes

the global low-rank to multiple matrices data, while

the Lp-norm provides noise robustness. Total variation

regularization is incorporated to promote structural

smoothness and edge preservation in the image. To

solve the nonconvex and nonsmooth model, an efficient alternating direction method of multipliers

(ADMM) is designed. In addition, we discuss how the

value of p in the three items affects the denoising

performance in simulated images. Extensive experiments on face datasets, videos, and real-world noisy

images demonstrate that the proposed method significantly improves denoising performance, particularly

for removing large sparse noise.

Joint work with Hu Yang.

Applying a Collaborative Learning Framework

Based on Pyramid Shaped Visual Transformers for

Uncertainty Information in Fundus Image Classification, Segmentation, and Object Detection Tasks

将基于金字塔型视觉变换器的不确定性信息共学

习框架用于眼底图像的分类，分割以及目标检测任

务

Xingshu Chen

Sun Yat-sen University

摘要: 在医学影像分析中，准确的分类、分割和目

标检测对疾病的有效诊断和疾病管理至关重要。传

统方法通常因强调学习共同的特征和共享模型参

数而在特征的可靠性和模型的表现上做出牺牲。本

文介绍了一种通过将 Pyramid Vision Transformer

（PVT）整合到不确定性信息共学习（UML）框架

中的改进方法，专门用于处理眼底图像分析任务。

UML 框架利用 PVT 无卷积操作的密集预测能力，

专门用于处理医学眼底图像数据。UML 框架包括

三个主要组成部分：1) 用于稳健的图像级和像素级

置信度估计的证据深度学习方法；2) 用于精确分割

出血和渗出的不确定性导航器；3) 确保将眼底条件

精确分类为正常、出血和渗出的不确定性指引器。

框架的功能已扩展到包括微血管瘤的目标检测

——这些小血管膨出是糖尿病视网膜病变等疾病

的早期指标，可能导致严重的视力损害。这一能力

促进了包括分类、分割和关键的早期病理特征检测

在内的全面诊断评估。在多个公共和私有数据集上

的广泛验证显示，我们的集成 UML-PVT 模型在

准确性、稳健性和可靠性方面显著优于现有方法。

将 PVT 整合到 UML 框架中为分析眼底图像提

供了一个强大的解决方案，可以有效处理分类、分

割和目标检测任务。这种方法不仅改进了诊断过程

，还促进了未来医学影像分析技术的发展，强调了

在医疗环境中使用先进、可靠和可解释的人工智能

的重要性。

Joint work with Ting Tian.

Aggregate Spectral Clustering for Multi-Layer

Popularity Adjusted Block Model

Jun Dai

University of Science and Technology of China

Abstract: Community detection aims to discover

underlying common community structures from network matrices or graphs. Our paper proposes a multi-layer network model that addresses the shortcomings of traditional multi-layer stochastic block models,

which often assume stronger intra-community connections than inter-ones. This model accounts for

outlier nodes within communities, thereby enhancing

its accuracy. We establish theoretical convergence

properties of the aggregate spectral clustering method

for this model and further propose a weighted aggregate approach to mitigate the interference of noisy

layers on the overall community signals. Our method

is data-driven and computationally efficient. Extensive simulation experiments demonstrate that our

refined method exhibits significant competitiveness

110

compared to traditional methods across a wide range

of parameter settings.

Information-Theoretic Thresholds for the Alignments of Partially Correlated Graphs

Dong Huang

Tsinghua University

Abstract: This paper studies the problem of recovering the hidden vertex correspondence between two

correlated random graphs. We propose the partially

correlated Erdős-Rényi graphs model, wherein a pair

of induced subgraphs with a certain number are correlated. We investigate the information-theoretic

thresholds for recovering the latent correlated subgraphs and the hidden vertex correspondence. We

prove that there exists an optimal rate for partial recovery for then umber of correlated nodes, above

which one can correctly match a fraction of vertices

and below which correctly matching any positive

fraction is impossible, and we also derive an optimal

rate for exact recovery. In the proof of possibility

results, we propose correlated functional digraphs,

which categorize the edges of the intersection graph

into two cases of components,and bound the error

probability by lower-order cumulant generating functions. The proof of impossibility results build upon the

generalized Fano’s inequality and the recovery

thresholds settled in correlated Erdős-Rényi graphs

model.

Joint work with Xianwen Song and Pengkun Yang.

Contributed Session CS024: Statistical Modeling

for Complex Data

Fast and Stable Estimation for Structural Break

Models with Endogenous Regressors

Changwei Zhao

Xiamen University

Abstract: This article introduces a new detection

method for multiple structural break models with

endogenous regressors based on the two-stage least

square estimation framework. The existing literature

either employs the information criterion methods, or

the sequential testing procedures, both of which have

practical limitations. To circumvent these limitations,

we propose an appealing cross-validation criterion

incorporating an order-preserved sample-splitting

strategy tailored for change-point models, and develop

a computationally scalable algorithm for implementation. Additionally, for testing purposes, we also develop a score-type test statistic as an alternative to the

Wald-type statistics. Intensive simulation studies are

conducted to make comparisons among various popular methods, and the results reflect the promising potential of our methods. Finally, we demonstrate the

proposed method through an application to the New

Keynesian Phillips Curve.

Joint work with Chuang Wan, Wei Zhong and Wenyang Zhang.

Simultaneous Outlier Detection and Prediction for

Kriging with True Identification

Youjie Zeng

University of Science and Technology of China

Abstract: Kriging with interpolation property is

widely used in various noise-free areas such as computer experiments. However, due to its Gaussian assumption, Kriging is susceptible to outliers which

affect statistical inference, and the resulting conclusions could be misleading. There is little work done

on outlier detection for kriging. To this end, we propose a novel kriging for simultaneous outlier detection

and prediction by introducing the normal-gamma

process to produce biases that results in an unbounded

penalty function at the origin to distinguish outliers

from normal data points. We develop a simple and

efficient method, avoiding expensive computation of

Markov chain Monte Carlo algorithm, to simultaneously detect outliers and make prediction. We establish the true identification property for outlier detection and the consistency of the estimated parameters

in kriging under the increasing domain framework, as

if the number and locations of the outliers were

known in advance. Under appropriate regularity conditions, we show information consistency for prediction in the presence of outliers. Numerical studies and

real data examples show that the proposed method

generally provides robust analyses in the presence of

outliers.

111

Joint work with Zhanfeng Wang, Youngjo Lee and

Niansheng Tang.

Time-Varying Mediation Analysis for Incomplete

Data with Application to DNA Methylation Study

for PTSD

Kecheng Wei

Fudan University

Abstract: DNA methylation (DNAm) has been shown

to mediate causal effects from traumatic experiences

to post-traumatic stress disorder (PTSD). However,

the scientific question about whether the mediation

effect changes over time remains unclear. In this paper,

we develop time-varying structural equation models to

identify cytosine-phosphate-guanine (CpG) sites

where DNAm mediates the effect of trauma exposure

on PTSD, and to capture dynamic changes in mediation effects. The proposed methodology is motivated

by the Detroit Neighborhood Health Study (DNHS)

with high-dimensional and longitudinal DNAm measurements. To handle the non-monotone missing

DNAm in the dataset, we propose a novel Longitudinal Multiple Imputation (LMI) method utilizing dependency among repeated measurements, and employ

the generalized method of moments to integrate the

multiple imputations. Simulations confirm that the

proposed method outperforms existing approaches in

various longitudinal settings. In DNHS data analysis,

our method identifies several CpG sites where DNAm

exhibits dynamic mediation effects. Some of the corresponding genes have been shown to be associated

with PTSD in the existing literature, and our findings

on their time-varying effects could deepen the understanding of the mediation role of DNAm on the causal

path from trauma exposure to PTSD risk.

Joint work with Fei Xue, Qi Xu, Yubai Yuan, Yuexia

Zhang, Guoyou Qin, Agaz H. Wani, Allison E. Aiello,

Derek E. Wildman, Monica Uddin and Annie Qu.

Time-Varying Home Advantage in the English

Premier League: A Causal Discovery Method for

Non-Stationary Causal Processes

Minhao Qi

Shanghai University of Finance and Economics

Abstract: In sports analytics, home advantage is a

robust phenomenon, with the home team usually winning more games than the away team. Yet causal discovery for home advantage encounters unique challenges due to the non-stationary, time-varying environment. In response, we propose a novel causal discovery method, DYnamic Non-stAtionary local

M-estimatOrs (DYNAMO), to learn the causal structures of home advantage. Our approach can identify

lagged and instantaneous causal relationships with

linear or nonlinear time-varying modules from a general class of non-stationary causal processes. By leveraging local information, we provide theoretical

guarantees for the identifiability of non-stationary

causal structures without imposing additional assumptions on specific classes of time-varying patterns.

Simulation studies validate the efficacy of DYNAMO

in recovering time-varying causal structures. We apply

our method to high-resolution tracking data from the

2021-2022 English Premier League and find that

home advantage, represented by the difference in

expected goals (EG) between playing at home and

away, is mainly affected by three important factors:

familiarity with the field, crowd support, and referee

bias. These causal relationships vary among different

teams and evolve over various match periods. The

code implementing DYNAMO is publicly available

at:https://anonymous.4open.science

/r/DYNAMO-750A/.

Joint work with Hengrui Cai, Guanyu Hu and Weining Shen.

A Semiparametric Gaussian Mixture Model for

Chest CT-based 3D Blood Vessel Reconstruction

Qianhan Zeng

Peking University

Abstract: Computed tomography (CT) has been a

powerful diagnostic tool since its emergence in the

1970s. Using CT data, three-dimensional (3D) structures of human internal organs and tissues, such as

blood vessels, can be reconstructed using professional

software. This 3D reconstruction is crucial for surgical

operations and can serve as a vivid medical teaching

example. However, traditional 3D reconstruction

112

heavily relies on manual operations, which are

time-consuming, subjective, and require substantial

experience. To address this problem, we develop a

novel semiparametric Gaussian mixture model tailored for the 3D reconstruction of blood vessels. This

model extends the classical Gaussian mixture model

by enabling nonparametric variations in the component-wise parameters of interest according to voxel

positions. We develop a kernel-based expectation-maximization algorithm for estimating the model

parameters, accompanied by a supporting asymptotic

theory. Furthermore, we propose a novel regression

method for optimal bandwidth selection. Compared to

the conventional cross-validation-based (CV) method,

the regression method outperforms the CV method in

terms of computational and statistical efficiency. In

application, this methodology facilitates the fully

automated reconstruction of 3D blood vessel structures with remarkable accuracy.

Joint work with Jing Zhou, Ying Ji and Hansheng

Wang.

July 14, 8:30-10:10

Invited Session IS023: Deep Generative Models

Efficient Training Algorithms for Neural Networks

神经网络的高效训练算法

Jianfei Chen

Tsinghua University

摘要: 训练大规模神经网络需要巨量的计算资源。

如何设计更高效的训练算法，减少训练神经网络所

需的资源是推动人工智能更快发展的重要问题。本

次报告介绍了近似梯度下降，一个有理论保证的高

效训练算法框架。基于该框架，我们针对训练过程

中的不同需求，开发了高效算法，并进行了相应的

算子实现。针对算力需求高的问题，本次报告将介

绍 INT4 训练算法、分块 INT8 训练算法及 2:4 稀疏

训练算法。针对显存需求高的问题，本次报告将介

绍 4 比特激活压缩训练算法及 4 比特优化器。实验

结果初步展示出了这些方法在高效训练大语言模

型方面的潜力。

Conditional Stochastic Interpolation for Generative Learning

Guohao Shen

The Hong Kong Polytechnic University

Abstract: We propose a conditional stochastic interpolation (CSI) approach for learning conditional distributions. CSI learns probability flow equations or

stochastic differential equations that transport a reference distribution to the target conditional distribution.

This is achieved by first learning the velocity function

and the conditional score function based on conditional stochastic interpolation, which is then used to

construct a deterministic process governed by an ordinary differential equation or a diffusion process for

conditional sampling. We establish the transport equation and derive the explicit form of the conditional

score function with mild conditions. We also incorporate an adaptive diffusion term in our proposed CSI

model to address the instability issues arising during

the training process. Furthermore, we establish

113

non-asymptotic error bounds for learning the target

conditional distribution via conditional stochastic

interpolation in terms of KL divergence, taking into

account the neural network approximation error. We

illustrate the application of CSI on image generation

using benchmark image data.

Joint work with Ding Huang, Jian Huang, Ting Li.

Score Identity Distillation: Exponentially Fast Distillation of Pretrained Diffusion Models for

One-Step Generation

Mingyuan Zhou

University of Texas, Austin

Abstract: Diffusion-based models, key to advancing

generative AI with their photorealistic outputs, face a

major hurdle: slow generation speed. Our Score identity Distillation (SiD) method challenges the belief

that quality in diffusion models requires iterative refinement, offering a groundbreaking solution. SiD

streamlines the generative process into a single, swift

step, achieving rapid improvements in Fréchet Inception Distance (FID) during the distillation process and,

in many cases, exceeding the quality of the original

models, which require extensive steps, from dozens to

hundreds. By reinterpreting the forward diffusion

process with semi-implicit distributions and three

novel score-based identities, we introduce a unique

loss mechanism. This allows for quick FID reductions

by training the generator with its synthesized images,

eliminating the need for real data or conventional

reverse diffusion, all within a significantly reduced

generation timeframe. Evaluated across four benchmark datasets, SiD demonstrates unparalleled efficiency and superior quality compared to current generative methods, setting new standards for diffusion

model distillation and expanding the potential of diffusion-based generation. This innovation makes

high-quality generative processes more accessible and

feasible for various applications, opening new research and application avenues in generative AI.

Gaussian Denoising for Generative Learning

Yuan Gao

The Hong Kong Polytechnic University

Abstract: Gaussian denoising has emerged as a powerful principle for constructing simulation-free continuous normalizing flows (CNFs) for generative

learning. Despite their empirical successes, theoretical

properties of these flows and the regularizing effect of

Gaussian denoising have remained largely unexplored.

In this talk, we aim to address this gap by presenting

several theoretical results. First, we investigate the

well-posedness of these CNFs built on Gaussian denoising. Through a unified framework termed Gaussian interpolation flow, we establish the Lipschitz regularity of the flow velocity field, the existence and

uniqueness of the flow, and the Lipschitz continuity of

the flow map and the time-reversed flow map for

several rich classes of target distributions. We further

study how well these CNFs learn probability distributions from a finite sample using the flow matching

objective function. We establish nonparametric convergence rates with the Wasserstein-2 distance, to

quantify the distribution estimation efficiency of the

simulation-free CNFs based on linear interpolation.

Our convergence analysis offers theoretical guarantees

for justifying the empirical success of these CNFs

from a statistical perspective.

Joint work with Prof. Jian Huang, Prof. Yuling Jiao,

and Prof. Shurong Zheng.

Invited Session IS096: New Statistical and Machine Learning Methods for Complex Data

A Parsimonious Joint Model of Survival Outcomes

and Time-Varying Biomarkers

Zhiyang Zhou

University of Wisconsin-Milwaukee

Abstract: Dynamic risk prediction dynamically updates an individual's risk assessment for a particular

outcome by integrating new information over time.

The core challenge of this approach involves estimating the intricate interplay between time-varying risk

factors and survival outcomes. The

shared-random-effects joint model, a key strategy for

dynamic risk prediction, simultaneously fits submodels for longitudinal/survival outcomes. However, as

the number of time-varying biomarkers increases, so

does the size of unknown parameters, making the

114

model computationally demanding. Additionally, this

inflation may potentially compromise the predictive

accuracy due to approximation errors in handling the

complex likelihood function. To mitigate these issues,

we introduce a parsimonious joint model with fewer

random effects to further enhance the computational

efficiency. Our method demonstrates a competitive

predictive accuracy, verified by numerical studies.

Joint work with Lihui Zha.

A Bayesian Approach to Response Optimization on

Data with Multistratum Structure

Po Yang

University of Manitoba

Abstract: Response optimization is a process of identifying the input variable settings that optimize the

response. Multistratum design arises naturally in industrial experiments due to the inconvenient and impractical completely randomization. Accounting for

the model uncertainty, we apply the Bayesian model

averaging method and predictive approach to investigate the optimization problem for data with multi-stratum structure. With the posterior probabilities of

models as weights, we consider the weighted average

of the predictive densities of the response over all

potential models. The goal of the optimization is to

identify the values of the factors that result in a maximum probability of a response in a given range. The

method is illustrated with two examples.

Joint work with Xiaohua Liu, Chang-Yun Lin.

Machine Learning-Driven Data Integration for

Precision Medicine

Pingzhao Hu

Western University

Abstract: With the advancement of high-throughput

technologies, an increasing amount of multiomic data

has been generated. These data sets offer a wealth of

information crucial for advancing precision medicine.

However, efficiently integrating and analyzing these

data for downstream applications poses significant

challenges. This talk will explore various tensor factorization-driven methods designed to address these

challenges in multiomic data integration. We will

delve into their applications in different domains such

as disease subtyping, radiogenomic analysis, drug

discovery and survival analysis, highlighting how

these innovative approaches can enhance our understanding and treatment of complex diseases through

more precise and comprehensive data analysis and

integration.

Improved Joint Modeling of Longitudinal and

Survival Data Using a Poisson Regression Approach

Depeng Jiang

University of Manitoba

Abstract: Studies often collect data on repeated

measurements (longitudinal) and time-to-events (survival). Recently, there has been significant discussion

on the joint model (JM) for analyzing both types of

outcomes simultaneously. JMs are computationally

demanding due to a large number of parameters and

the complexity of fitting the survival submodel, which

typically involves the piecewise constant proportional

hazard (PCPH). An alternative approach for survival

data analysis is the auxiliary Poisson regression model,

but its use in JMs has not been thoroughly explored.

In this study, we propose incorporating the auxiliary

Poisson model into the survival component of a JM

using a Bayesian framework. Through extensive simulation studies, we evaluated the performance of our

method under various conditions and compared it to

the JMbayes R package. Furthermore, we applied our

method to data from the Manitoba Follow-Up Study

to demonstrate its advantages and feasibility. Our

results suggest that utilizing the auxiliary Poisson

approach in the survival submodel is a promising

strategy for jointly analyzing longitudinal and survival

data, offering computational efficiency.

Joint work with Yixiu Liu, Depeng Jiang, Mahmoud

Torabi, Xuekui Zhang.

Invited Session IS024: High-Dimensional Statistical Learning

Kernel Variable Importance Measure

Liuhua Peng

University of Melbourne

115

Abstract: We propose a kernel variable importance

measure (KvIM) based on the maximum mean discrepancy (MMD). The KvIM can effectively measure

the importance of an individual dimension by quantifying its impact on the MMD. In addition, we propose

an objective-oriented penalized kernel variable importance measure (OOP-KvIM), which can objectively detect the number of important variables. Both

KvIM and OOP-KvIM do not rely on model assumptions and thus are model-free. In addition, they take

dependence across dimensions into account and can

capture important variables under different models.

Theoretical properties of the KvIM and OOP-KvIm

are established and supported by simulations and real-data examples.

Distribution-Free Prediction Bands for Clustered

Data with Missing Responses

Yanlin Tang

East China Normal University

Abstract: Existing methods for missing clustered data

often rely on strong model assumptions and are therefore prone to model misspecification. In this paper, we

construct covariate-dependent prediction bands for

new subjects at specific points or trajectories for

missing cluster data, without making any assumptions

about the model specification or within-cluster dependence structure. The proposed methods are based

on conformal inference combined with subsampling

and appropriate weighting to accommodate within-cluster correlation and the missing data mechanism,

thereby ensuring coverage guarantee in finite samples.

To provide an asymptotic conditional coverage guarantee for each given subject, we further propose predictions by establishing the highest posterior density

region of the target, which is more accurate under

complex error distributions, such as asymmetric and

multimodal distributions. Simulation studies show

that our methods have better finite-sample performance under various complex error distributions

compared to other alternatives. The practical usage of

the proposed method is demonstrated using motivating examples from serum cholesterol and CD4+ cell

data sets.

Joint work with Menghan Yi, Yingying Zhang,

Huixia Wang.

Integrative Nearest Neighbor Classifiers for

Block-Missing Multi-Modal Data

Guan Yu

University of Pittsburgh

Abstract: Classifiers leveraging multi-modal data

often have excellent classification performance.

However, in certain studies, due to various reasons,

some modalities are not collected from a sizable subset of participants and thus all data from those modalities are missing completely. Considering classification

problems with a block-missing multi-modal training

data set, we develop a new integrative nearest neighbor (INN) classifier. INN harnesses all available information in the training data set and the feature vector of the test data point effectively to predict the class

label of the test data point without deleting or imputing any missing data. Given a test data point, INN

determines the weights on the training samples adaptively by minimizing the worst-case upper bound on

the estimation error of the regression function over a

convex class of functions. As a weighted nearest

neighbor classifier, INN suffers from the curse of

dimensionality. Therefore, in high-dimensional scenarios, we propose a two-step INN, assuming that the

regression function depends on features via sparse

linear combinations of features. Our two-step INN

estimates those linear combinations first, and then use

them as new features to build the classifier. The effectiveness of our proposed methods have been demonstrated by both theoretical and numerical studies.

Residual Importance Weighted Transfer Learning

for High-Dimensional Linear Regression

Junlong Zhao

Beijing Normal University

Abstract: Transfer learning is an emerging paradigm

for leveraging multiple sources to improve the statistical inference on a single target. In this paper, we

propose a novel approach named residual importance

weighted transfer learning (RIW-TL) for

high-dimensional linear models built on penalized

116

likelihood. Compared to existing methods such as

Trans-Lasso that selects sources in an all-in-all-out

manner, RIW-TL includes samples via importance

weighting and thus may permit more effective sample

use. To determine the weights, remarkably RIW-TL

only requires the knowledge of one-dimensional densities dependent on residuals, thus overcoming the

curse of dimensionality of having to estimate

high-dimensional densities in naive importance

weighting. We show that the oracle RIW-TL provides

faster rate than its competitors and develop a

cross-fitting procedure to estimate this oracle. We

discuss variants of RIW-TL by adopting different

choices for residual weighting. The theoretical properties of RIW-TL and its variants are established and

compared with those of LASSO and Trans-Lasso.

Extensive simulation and a real data analysis confirm

its advantages.

Invited Session IS019: Financial Machine Learning

Corporate Risk Disclosures, Stock Price Comovement, and Expected Returns

Haifeng You

The Hong Kong University of Science and Technology

Abstract: This paper investigates the decision usefulness of risk factor disclosures in corporate annual

reports. We employ BERTopic, a state-of-the-art topic

modeling technique, to categorize paragraphs into

distinct risk factors, thereby quantifying each firm's

annual risk factor exposure. We find significant

comovements in stock returns and financial fundamentals between firms with similar risk factor profiles.

Our analysis demonstrates that risk-based peer groups

account for a substantially larger variance in stock

returns and financial ratios compared to industry peers

classified by the Global Industry Classification Standard (GICS) and the product-based peers introduced by

Hoberg and Phillips (2016). By utilizing the disclosure-based risk factor exposures as instrumental variables, we derive an implied expected return measure.

This measure shows a substantial and significant association with subsequent realized stock returns, underscoring the predictive power of risk factor disclosures in enhancing investment decision-making.

Mitigating Estimation Risk in High Dimensional

Portfolio Selection

Ming Yuan

Columbia University

Abstract: Usual estimated investment strategies are

subject to estimation risk and will not achieve the

optimal Sharpe ratio when the dimensionality is high

relative to sample size. We shall discuss the merits of

several common approaches to mitigation and explore

to what extent they can be practically useful.

Learning the Stochastic Discount Factor

Xinghua Zheng

The Hong Kong University of Science and Technology

Abstract: We develop a statistical framework to learn

the high-dimensional stochastic discount factor (SDF)

from a large set of characteristic-based portfolios.

Specifically, we build on the maximum-Sharpe ratio

estimated and sparse regression method proposed in

Ao, Li and Zheng (RFS,2019) to construct the SDF

portfolio, and develop a statistical inference theory to

test the SDF loadings. Applying our approach to 194

characteristic based portfolios, we find that the SDF

constructed by about 20 of them performs well in

achieving a high Sharpe ratio and explaining the

cross-section of expected returns of various portfolios.

Joint work with Zhanhui Chen, Yi Ding and Yingying Li.

Invited Session IS008: Causal Inference in Observational Studies

Debiased Estimating Equation Method for Versatile and Efficient Mendelian Randomization Using

Large Numbers of Correlated and Invalid SNPs

with Weak Effects

Ruoyu Wang

Harvard University

Abstract: Mendelian randomization (MR) is a powerful tool for identifying causal effects in the presence

of unobserved confounding, utilizing single nucleotide

polymorphisms (SNPs) as instrumental variables

117

(IVs). However, SNPs often have small effects on

complex traits, leading to bias and low statistical efficiency in MR analysis. Additionally, the strong linkage disequilibrium among SNPs is compounding this

issue, which poses additional statistical hurdles. To

address these challenges, this paper proposes DEEM

(Debiased Estimating Equation Method), a summary

statistics-based MR approach that can incorporate

numerous correlated SNPs with weak effects. DEEM

effectively eliminates the weak IV bias, adequately

accounts for the correlations among SNPs, and enhances efficiency by leveraging information from

correlated weak IVs. DEEM is a versatile method that

allows adjustment for pleiotropic effects and applies

to both two-sample and one-sample MR analyses. We

establish the consistency and asymptotic normality of

DEEM's estimates. Extensive simulations and two real

data examples demonstrate that DEEM can improve

the efficiency and robustness of MR analysis.

Joint work with Haoyu Zhang, Xihong Lin.

Policy Learning with Distributional Welfare

Yifan Cui

Zhejiang University

Abstract: In this paper, we explore optimal treatment

allocation policies that target distributional welfare.

Most literature on treatment choice has considered

utilitarian welfare based on the conditional average

treatment effect (ATE). While average welfare is intuitive, it may yield undesirable allocations especially

when individuals are heterogeneous (e.g., with outliers) - the very reason individualized treatments were

introduced in the first place. This observation motivates us to propose an optimal policy that allocates the

treatment based on the conditional quantile of individual treatment effects (QoTE). Depending on the

choice of the quantile probability, this criterion can

accommodate a policymaker who is either prudent or

negligent. The challenge of identifying the QoTE lies

in its requirement for knowledge of the joint distribution of the counterfactual outcomes, which is generally hard to recover even with experimental data.

Therefore, we introduce minimax policies that are

robust to model uncertainty. A range of identifying

assumptions can be used to yield more informative

policies. For both stochastic and deterministic policies,

we establish the asymptotic bound on the regret of

implementing the proposed policies. In simulations

and two empirical applications, we compare optimal

decisions based on the QoTE with decisions based on

other criteria. The framework can be generalized to

any setting where welfare is defined as a functional of

the joint distribution of the potential outcomes.

Joint work with Sukjin Han.

The Promises of Parallel Outcomes

Linbo Wang

University of Toronto

Abstract: A key challenge in causal inference from

observational studies is the identification and estimation of causal effects in the presence of unmeasured

confounding. In this paper, we introduce a novel approach for causal inference that leverages information

in multiple outcomes to deal with unmeasured confounding. The key assumption in our approach is conditional independence among multiple outcomes. In

contrast to existing proposals in the literature, the

roles of multiple outcomes in our key identification

assumption are symmetric, hence the name parallel

outcomes. We show nonparametric identifiability with

at least three parallel outcomes and provide parametric estimation tools under a set of linear structural

equation models. Our proposal is evaluated through a

set of synthetic and real data analyses.

Joint work with Ying Zhou, Dingke Tang, Dehan

Kong.

Few-Shot Multi-Task Learning with MetaSP

Lin Liu

Shanghai Jiao Tong University

Abstract: Data scarcity poses a serious threat to

modern machine learning and artificial intelligence, as

their practical success typically relies on the availability of big datasets. One effective strategy to mitigate

the issue of insufficient data is to first harness information from other data sources possessing certain

similarities in the study design stage, and then employ

the multi-task or meta learning framework in the

118

analysis stage. In this paper, we focus on multi-task

(or multi-source) linear models whose coefficients

across tasks share an invariant low-rank component, a

popular structural assumption considered in the recent

multi-task or meta learning literature. Under this assumption, we propose a new algorithm, called Meta

Subspace Pursuit that provably learns this invariant

subspace shared by different tasks. Under this stylized

setup for multi-task or meta learning, we establish

both the algorithmic and statistical guarantees of the

proposed method.

Invited Session IS002: Advancements in Integrative Statistical Inference

Guided Adversarial Robust Transfer Learning

with Source Mixing

Zijian Guo

Rutgers University

Abstract: Transfer learning is a critical technique that

enables the application of knowledge gained from

existing tasks or domains to improve performance on

a new one, reducing the need for extensive data and

training in each new context. Many existing transfer

learning methods rely on leveraging information from

source populations closely resembling the target population. However, this approach often overlooks valuable knowledge that may be present in different yet

potentially related auxiliary samples. When dealing

with a limited amount of target data and multiple

source data, we introduce a novel approach, Guided

Adversarial Robust Transfer (GART) learning, that

breaks free from strict similarity constraints. GART is

designed to optimize the most adversarial loss with

respect to a collection of source mixture populations

that guarantee excellent prediction performances for

the target data. We establish the closed form of the

population GART and show that the GART estimator

achieves a faster convergence rate than the model

fitted with the target data. Our simulation studies

suggest that GART outperforms existing transfer

learning methods, attaining higher robustness and

accuracy. We highlight GART's predictiveness and

robustness by applying it to form genetic prediction

models of high-density lipoprotein cholesterol using

multi-institutional biobank-linked electronic health

records data.

Joint work with Xin Xiong, Tianxi Cai.

Aggregating Dependent Signals with Heavy-Tailed

Combination Tests

Jingshu Wang

University of Chicago

Abstract: Combining dependent p-values presents a

longstanding challenge in statistical inference, particularly when aggregating results from diverse methods

to boost signal detection. P-value combination tests

using heavy-tailed distribution-based transformations,

such as the Cauchy combination test and the harmonic

mean p-value, have recently garnered significant interest in genetic applications for their potential to

efficiently handle arbitrary p-value dependencies. In

the talk, I will present our new results providing a

deeper understanding of the heavy-tailed combination

tests. Specifically, though researchers have shown that

these combination tests are asymptotically valid for

pairwise quasi-asymptotically independent test statistics, such as bivariate normal variables, we find out

that they are also asymptotically equivalent to the

Bonferroni test under the same conditions, making

them uninteresting. On the other hand, we show when

quasi-asymptotic independence is violated, such as

when the test statistics follow multivariate t distributions, these tests are still asymptotically valid, and can

be asymptotically much more powerful than the Bonferroni test. Our new results provide a broader justification of the heavy-tailed combination tests and indicate their practical utility when some p-values are

highly dependent.

An Integrative Multi-Context Mendelian Randomization Method for Identifying Risk Genes

Across Human Tissues

Fan Yang

Tsinghua University

Abstract: Mendelian randomization (MR) provides

valuable assessments of the causal effect of exposure

on outcome, yet the application of conventional MR

methods for mapping risk genes encounters new chal-

119

lenges. One of the issues is the limited availability of

expression quantitative trait loci (eQTLs) as instrumental variables (IVs), hampering the estimation of

sparse causal effects. Additionally, the often context/tissue-specific eQTL effects challenge the MR

assumption of consistent IV effects across eQTL and

GWAS data. To address these challenges, we propose

a multi-context multivariable integrative MR framework, mintMR, for mapping expression and molecular

traits as joint exposures. It models the effects of molecular exposures across multiple tissues in each gene

region, while simultaneously estimating across multiple gene regions. It uses eQTLs with consistent effects

across more than one tissue type as IVs, improving IV

consistency. A major innovation of mintMR involves

employing multi-view learning methods to collectively model latent indicators of disease relevance across

multiple tissues, molecular traits, and gene regions.

The multi-view learning captures the major patterns of

disease-relevance and uses these patterns to update the

estimated tissue relevance probabilities. The proposed

mintMR iterates between performing a multi-tissue

MR for each gene region and joint learning the disease-relevant tissue probabilities across gene regions,

improving the estimation of sparse effects across

genes. We apply mintMR to evaluate the causal effects of gene expression and DNA methylation for 35

complex traits using multi-tissue QTLs as IVs. The

proposed mintMR controls genome-wide inflation and

offers new insights into disease mechanisms.

Joint work with Yihao Lu, Lin Chen.

Contributed Session CS002: Recent Advances in

Deep Learning

Auxiliary Learning and Its Statistical Understanding

Hanchao Yan

Renmin University of China

Abstract: Modern statistical analysis often encounters

high-dimensional problems but with a limited sample

size. It poses great challenges to traditional statistical

estimation methods. To address this issue, various

statistical methods have been developed, most of

which focus on penalized estimation and feature

screening. In the field of machine learning, the method of auxiliary learning is demonstrated to be a useful

alternative in many applications. The key idea of auxiliary learning is introducing a set of auxiliary learning

tasks to enhance the performance of the primary

learning task. In this work, we adopt auxiliary learning to solve the estimation problem in

high-dimensional settings. We start with the linear

regression setup. To improve the statistical efficiency

of the parameter estimator for the primary task, we

consider several auxiliary tasks, which share the same

covariates with the primary task. Then a weighted

estimator for the primary task is developed, which is a

linear combination of the ordinary least squares estimators of both the primary task and auxiliary tasks.

The optimal weight is analytically derived and the

statistical properties of the corresponding weighted

estimator are studied. We then extend the weighted

estimator to generalized linear regression models.

Extensive numerical experiments are conducted to

verify our theoretical results. Last, a deep learning-related real-data example of smart vending machines is presented for illustration purposes.

Joint work with Feifei Wang, Chuanxin Xia, Hansheng Wang.

A Multi-Feature Fusion Deep Learning Prediction

Method for Production of Tight Gas Reservoirs

Chunlan Zhao

Southwest Petroleum University

Abstract: Due to the complex percolation mechanism

of tight gas reservoirs, the traditional method of production prediction often exhibits poor accuracy. This

paper proposes multi-feature fusion deep learning

prediction method (GKVT) for production of tight gas

reservoirs. This method integrates the optimization

techniques of dynamic production data dimensionality

reduction, data quality enhancement, and variational

mode decomposition (VMD) with Long short-term

memory (LSTM) to enhance the performance of deep

learning, which belongs to the artificial intelligence

method in the field of oil and gas exploration and

development technology. Firstly, the Granger causality test is utilized to select the features of dynamic

120

production data, identifying key features and eliminating factors that don't significantly impact cumulative production. Secondly, the selected features are

optimized by Kalman filter (KF) to improve the data

quality. Subsequently, multiple sub-modal sequences

are obtained from the optimized cumulative production data through variational mode decomposition.

Finally, multi-modal prediction and fusion based on

LSTM are carried out to obtain the cumulative prediction result. Compared with single recurrent neural

network models (RNN LSTM, GRU) as well as hybrid models (KF-LSTM,

CNN-LSTM, CNN-GRU), the R2 value of GKVT is

increased by 0.73%~12.84%, and MAPE is decreased

by 0.15%~1.19%. In summary, this paper through

dynamic feature selection, filtering and denoising,

mode decomposition and deep learning, optimizes

data representation, strengthens the model's understanding of dynamic data, and uncovers the uncertain

and nonlinear relationships between gas well production data, as well as the interpretability, so as to improve the prediction accuracy, especially for the production prediction of tight gas reservoirs, which has

theoretical reference value.

Joint work with Xiang Wu, Wei Xu.

An Analysis of Switchback Designs in Reinforcement Learning

Qianglin Wen

Yunnan University

Abstract: This paper offers a detailed investigation of

switchback designs in A/B testing, which alternate

between baseline and new policies over time. Our aim

is to thoroughly evaluate the effects of these designs

on the accuracy of their resulting average treatment

effect (ATE) estimators. We propose a novel \"weak

signal analysis\" framework, which substantially simplifies the calculations of the mean squared errors

(MSEs) of these ATEs in Markov decision process

environments. Our findings suggest that (i) when the

majority of reward errors are positively correlated, the

switchback design is more efficient than the alternating-day design which switches policies in a daily basis.

Additionally, increasing the frequency of policy

switches tends to reduce the MSE of the ATE estimator. (ii) When the errors are uncorrelated, however, all

these designs become asymptotically equivalent. (iii)

In cases where the majority of errors are negative

correlated, the alternating-day design becomes the

optimal choice. These insights are crucial, offering

guidelines for practitioners on designing experiments

in A/B testing. Our analysis accommodates a variety

of policy value estimators, including model-based

estimators, least squares temporal difference learning

estimators, and double reinforcement learning estimators, thereby offering a comprehensive understanding of optimal design strategies for policy evaluation in reinforcement learning.

Joint work with Chengchun Shi, Ying Yang, Niansheng Tang, Hongtu Zhu.

Unlocking the Power of AI: Deep Learning of Volatility is Indispensable

Wenxuan Ma

Renmin University of China

Abstract: Recently it is popular to explore return

predictability with machine learning in cross-sectional

studies. We demonstrate that the volatility predictability with deep learning is significant and the economic

gain of doing so is substantial. The predicted conditional volatility and the predicted expected return

(both monthly) can be used to form a double-sorted

long-short portfolio, achieving a Sharpe ratio of approximately 3, which is much higher than that of the

single-sorted long-short portfolio. Moreover, we find

that conditional volatility and expected return are

negatively correlated in cross-sections, which is significant and persistent and is consistent with some

previous studies. The code and data have been released, and all results are reproducible without data

snooping biases.

Joint work with Yan Xing.

Production Prediction of Tight Gas Reservoirs

Based on Deep Learning BWO-TFT Model

Shaodan Hou

Southwest Petroleum University

Abstract: Gas production prediction of tight sand-

121

stone gas reservoirs is very important for

their development and exploration. However, due to

the influence of various complex factors, the production data often exhibit uncertainty and nonlinear

characteristics, making it challenging to accurately

predict production. In order to deal

with these challenges, this paper proposes a novel

prediction method based on deep learning, called

BWO-TFT, which uses Temporal Fusion Transformer

(TFT) model to capture the uncertainty and nonlinearity in the data, and optimizes model’s parameters

by using the Beluga whale optimization (BWO) algorithm to improve the prediction accuracy. Firstly, K-Means clustering algorithm based

on Dynamic Time Warping (DTW) is used

to cluster the tight gas wells in the study area into

three clusters: high production

wells, medium production wells and low production

wells. Then, the influencing factors such as oil pressure, casing pressure and current formation pressure

are input into the BWO-TFT model as model feature

variables to predict the cumulative gas production of

these three types of tight gas wells. Compared with

seven baseline models, the BWO-TFT model has

higher prediction accuracy and narrower interval coverage width, which can effectively quantify the uncertainty of production and provide strong support for the

exploration and development of tight gas reservoirs.

In high, medium and low production wells, the mean

absolute percentage errors

(MAPE) are 0.1565%,0.4333% and 0.3899%, respectively, and the coverage width-based criterion (CWC) are 0.3474, 0.2120 and 0.1855, respectively.

Joint work with Chunlan Zhao.

Invited Session IS047: Recent Advances in Data

Integration in Survey Sampling

Issues with Inverse Probability Weighting and

Model-based Prediction for Non-probability Samples

Changbao Wu

University of Waterloo

Abstract: We provide an overview of recent developments in statistical inference for non-probability

survey samples. We discuss issues arising from methodological developments related to inverse probability

weighting and model-based prediction and concerns

with practical applications. Three procedures proposed in the recent literature on the estimation of participation probabilities, namely, the method of Valliant

and Dever (2011) based on the pooled sample, the

pseudo maximum likelihood method of Chen, Li and

Wu (2020), and the method of Wang, Valliant and Li

(2021) using a two-step computational strategy, are

examined under a joint randomization framework.

The inexplicit impact of standard assumptions on

model-based prediction approach is examined, and the

main issue of undercoverage is highlighted. We discuss potential strategies for dealing with undercoverage problems in practice.

Combining Data from Mobile Network Operators

and Others for Official Statistics

Li-Chun Zhang

University of Southampton

Abstract: Although mobile network operator (MNO)

data have great potentials for producing official statistics on population, tourism, mobility and environment,

MNO data would not suffice on their own whenever

the target statistical unit or the measurement unit is

not mobile device per se. However, the Official Statistics Agency (OSA) cannot have access to MNO data

at the signal or device level, due to confidentiality,

commercial interest and technology reasons. What is

being made available to the OSA is (anonymised)

macro data aggregated over devices, such as how

many devices moved from city A to B on a given day.

This creates many methodological challenges, and

there are hardly any official statistics made from

MNO data currently.

In this talk we present some recent development on

methods for combining MNO and non-MNO data to

make official statistics. Depending on how the associated uncertainty is defined, one can classify any statistical method for combining MNO and non-MNO data

under three broad approaches: randomisation, quasi-randomisation and super-population modelling.

122

First, randomisation requires a specialised survey to

convert the MNO data into the target statistical outputs, whose uncertainty is considered to be dominated

by the sampling error under the known survey sampling design. Next, although MNO data are not observed based on some known probabilities, one may

introduce a model of the underlying mechanism as if

they were, and assess the uncertainty accordingly.

Such a quasi-randomisation approach can be implemented together with suitable non-MNO population

data and, if fit-for-purpose, can remove the need of

specialised surveys altogether and considerably scale

up the use of MNO data. Finally, it is often possible to

build a so-called super-population model for specific

target variables from non-MNO sources, using features derived from relevant MNO data. However,

different models are needed for different statistics

generally, unlike building a quasi-randomisation model that is applicable to all the different variables associated with the same mobile devices.

A New Class of Robust Linear Estimators

Zhonglei Wang

Xiamen University

Abstract: Linear estimators are commonly used in

statistics, including survey sampling and causal inference, but existing works mainly ignore wrongly recorded extremes, leading to biased or inefficient estimators. In this paper, we propose a new class of robust linear estimators via median of means, and theoretical properties are investigated under regularity

conditions. The new robust linear estimators achieve

the same statistical efficiency as if there were wrongly

recorded extremes, and they can be easily implemented by standard statistical software. Numerical

results demonstrate the advantages of the proposed

estimators over state-of-art ones, and the coverage

rates of the proposed confidence interval are also satisfactory.

Joint work with Zhixiang Zhou, Shui Yao, Xiaojun

Mao.

Inductive Matrix Completion through Transfer

Learning

Hengfang Wang

Fujian Normal University

Abstract: The advent of big data has facilitated the

development of impactful models by enabling the

storage of vast amounts of data. Transfer learning, a

technique in machine learning, facilitates knowledge

transfer across different domains by employing

pre-existing models from a source domain to enhance

performance in a target domain. On the other hand,

inductive matrix completion utilizes supplementary

information from various sources to enhance task

efficacy. This study delves into the integration of inductive matrix completion within the framework of

transfer learning, wherein our proposed method incorporates group sparsity to account for differences

between the core matrices of the target and source

domains. We investigate the theoretical underpinnings

of our approach to demonstrate its advantages over

existing methods in achieving gains through transfer

learning. Synthetic experiments are conducted to assess the effectiveness of our proposed approach.

Invited Session IS049: Recent Advances in Efficient and Fair Machine Learning

Network-Adjusted Covariates for Community

Detection

Wanjie Wang

National University of Singapore

Abstract: Community detection is a crucial task in

network analysis that can be significantly improved

by incorporating subject-level information, i.e. covariates. Existing methods have shown the effectiveness of using covariates on the low-degree nodes, but

rarely discuss the case where communities have significantly different density levels, i.e. multiscale networks. In this paper, we introduce a novel method that

addresses this challenge by constructing network-adjusted covariates, which leverage the network

connections and covariates with a node-specific

weight to each node. This weight can be calculated

without tuning parameters. We present novel theoretical results on the strong consistency of our method

under degree-corrected stochastic blockmodels with

covariates, even in the presence of mis-specification

123

and multiple sparse communities. Additionally, we

establish a general lower bound for the community

detection problem when both network and covariates

are present, and it shows our method is optimal for

connection intensity up to a constant factor.

Our method outperforms existing approaches in simulations and a LastFM app user network. We then

compare our method with others on a statistics publication citation network where 30% of nodes are isolated, and our method produces reasonable and balanced results.

Joint work with Yaofang Hu.

Data-Driven Minimax Optimization with Expectation Constraints

Xudong Li

Fudan University

Abstract: Attention to data-driven optimization approaches has grown significantly over recent decades,

but data-driven constraints have rarely been studied.

In this talk, we focus on the non-smooth convex-concave stochastic minimax regime and formulate the data-driven constraints as expectation constraints. Then, we propose a class of efficient primal-dual algorithms to tackle the minimax optimization with expectation constraints, and show that our

algorithms converge at the optimal rate of ? (

1

√?

),

where ? is the number of iterations. We also verify

the practical efficiency of our algorithms by conducting numerical experiments on large-scale real-world

applications.

Joint work with Shuoguang Yang and Guanghui Lan.

Fair Risk Control: A Generalized Framework for

Calibrating Multi-group Fairness Risks

Linjun Zhang

Rutgers University

Abstract: This paper introduces a framework for

post-processing machine learning models so that their

predictions satisfy multi-group fairness guarantees.

Based on the celebrated notion of multicalibration, we

introduce(?, ? , ?)-GMC (Generalized Multi- Dimensional Multicalibration) for multi-dimensional mappings ?, constraint set ?, and a pre-specified threshold level ?. We propose associated algorithms to

achieve this notion in general settings. This framework is then applied to diverse scenarios encompassing different fairness concerns, including false negative rate control in image segmentation, prediction set

conditional uncertainty quantification in hierarchical

classification, and de-biased text generation in language models. We conduct numerical studies on several datasets and tasks.

Optimal and Efficient Thompson Sampling Algorithms for Bandits and Reinforcement Learning

Pan Xu

Duke University

Abstract: Thompson sampling is widely recognized

for its effectiveness in addressing online decision

problems, boasting simplicity in implementation and

superior empirical performance compared to other

methods. However, its theoretical analysis remains

relatively unexplored. In this presentation, I will detail

our recent efforts to achieve asymptotically and minimax optimal regret for Thompson sampling in multi-armed bandits. Additionally, I will explore the development of computationally efficient Thompson

sampling algorithms for contextual bandits, deep reinforcement learning, and cooperative multi-agent

reinforcement learning. These algorithms leverage

approximate sampling techniques to learn the exact

posterior distribution, offering flexibility in design

and ease of implementation in practical applications.

Invited Session IS057: Recent Advances in Statistical Network Analysis - Theory and Methodology

Hypothesis Test of General Network Models

一般网络模型的假设检验

Jianwei Hu

Central China Normal University

Abstract: The network data has attracted considerable

attention in modern statistics. In research on complex

network data, one key issue is finding its underlying

connection structure given a network sample. The

methods that have been proposed in literature usually

assume that the underlying structure is a known model.

124

In practice, however, the true model is usually unknown, and network learning procedures based on

these methods may suffer from model misspecification.

To handle this issue, based on the random matrix theory, we first give a spectral property of the normalized

adjacency matrix under a mild condition. Further, we

establish a general goodness-of-fit test procedure for

the unweight and undirected network. We prove that

the null distribution of the proposed statistic converges in distribution to the standard normal distribution.

Theoretically, this testing procedure is suitable for

nearly all popular network models, such as stochastic

block models, and latent space models. Further, we

apply the proposed method to the degree-corrected

mixed membership model and give a sequential estimator of the number of communities. Both simulation

studies and real-world data examples indicate that the

proposed method works well.

Transfer Learning in High-Dimensional Network

Regression Model

Danyang Huang

Renmin University of China

Abstract: Transfer learning utilizes models from

various source studies to enhance prediction accuracy

in a specific target study. Although it is widely applied

in the biomedical and social sciences for independent

samples, its application to network data is less explored and currently lacks solid theoretical investigation. We introduce a transfer learning algorithm designed for high-dimensional linear regression on network data, with its theoretical properties thoroughly

established. Theoretical analysis demonstrates superior estimation speeds compared to methods that do not

incorporate source studies. Additionally, we propose a

NetTrans-Lasso algorithm, which effectively identifies transferable data across networks, thereby improving learning performance through robust

knowledge transfer. Our theoretical findings are corroborated by extensive simulations and a real-world

social network dataset from Weibo.

Joint work with Huimin Cheng, Debarghya Mukherjee.

Community Detection in Weighted Networks

Binghui Liu

Northeast Normal University

Abstract: In this paper, we address the issue of community detection in weighted networks. Most existing

methods, particularly statistical approaches based on

likelihood optimization, suffer from a notable drawback: the necessity to specify in advance the specific

form of the distribution of edge weights conditional

on the community labels. This requirement forces

algorithms to be custom-tailored exclusively to that

distribution, leading to significant limitations in practical applications where the distribution type is unknown. To overcome this limitation, we propose a

general framework based on expectation profile-pseudo likelihood maximization for community

detection in both undirected and directed weighted

networks. These methods are applicable to various

types of weighted networks and are independent of the

specific form of the conditional distribution. Through

simulation studies, we show significant advantages of

the proposed methods across a wide range of conditional distributions and parameter settings, in terms of

both community detection accuracy and computational efficiency. In practical applications, we demonstrate

the applicability of the proposed methods using three

real-world weighted networks.

Community Detection with Heterogeneous Block

Covariance Model

Yunpeng Zhao

Colorado State University

Abstract: Community detection is the task of clustering objects based on their pairwise relationships. Most

of the model-based community detection methods,

such as the stochastic block model and its variants, are

designed for networks with binary (yes/no) edges. In

many practical scenarios, edges often possess continuous weights, spanning positive and negative values,

which reflect varying levels of connectivity. To address this challenge, we introduce the heterogeneous

block covariance model (HBCM) that defines a community structure within the covariance matrix, where

edges have signed and continuous weights. Further-

125

more, it takes into account the heterogeneity of objects when forming connections with other objects

within a community. A novel variational expectation-maximization algorithm is proposed to estimate

the group membership. The HBCM provides provable

consistent estimates of memberships, and its promising performance is observed in numerical simulations

with different setups. The model is applied to a yeast

gene expression dataset to detect the gene clusters

regulated by different transcript factors during the

yeast cell cycle.

Joint work with Xiang Li, Qing Pan, Ning Hao.

Invited Session IS017: Financial and Macroeconometrics

Panel Quantile GARCH Models under Homogeneity

Qianqian Zhu

Shanghai University of Finance and Economics

Abstract: Empirical evidence indicates that the estimates of GARCH parameters cluster in a panel of

financial assets, potentially due to assets with similar

exposure to common market factors. To capture the

subgroup effect on conditional quantiles of financial

asset returns and improve estimation efficiency by

pooling information across individuals within the

same group, this paper introduces the panel quantile

GARCH model with homogeneous structures in the

coefficient functions. We propose a three-stage estimation procedure to detect the grouping structures

using a binary segmentation algorithm and estimate

the coefficient functions under detected homogeneity

by quantile regression. Asymptotic properties are

established for both group detection and the coefficient estimators. Simulation experiments are conducted to evaluate the finite sample performance of

the proposed estimation procedure. The results indicate that the final estimator, which utilizes group panel information, is more efficient than the initial estimator that relies on individual information alone,

particularly when a subgroup effect exists. An empirical example on a panel of U.S. financial returns is

presented to illustrate the usefulness of the proposed

methodology in pursuing homogeneity, as well as its

superior performance in forecasting value-at-risks at

tail quantiles and inter-quartiles compared to quantile

GARCH models that do not utilize any homogeneous

information in the panel.

Joint work with Wenyu Li, Wenyang Zhang,

Guodong Li.

Sign-Based Tests for Structural Changes in Multivariate Volatility

Jilin Wu

Xiamen University

Abstract: This paper develops two sign-based tests to

detect structural changes in multivariate volatility by

the least absolute deviation approach. We establish

mild conditions under which the new tests have

standard null distributions and are powerful against

any fixed alternatives that deviate from the null, including smooth changes, single or multiple breakpoints in multivariate volatility. In addition, the tests

also have asymptotic unit power against Pitman-type

local alternatives. Simulations are conducted to show

the better finite sample performance of the new tests

relative to other tests in the presence of heavy-tailed

innovations. Finally, two empirical applications to

detection of structural changes in volatilities of financial markets highlight the usefulness of our tests in

real datasets.

Robust M-Estimation for Additive Single-Index

Cointegrating Time Series Models

Chaohua Dong

Zhongnan University of Economics and Law

Abstract: Robust M-estimation uses loss functions,

such as least absolute deviation (LAD), quantile loss

and Huber's loss, to construct its objective function, in

order to for example eschew the impact of outliers,

whereas the difficulty in analysing the resultant estimators rests on the nonsmoothness of these losses.

Generalized functions have advantages over ordinary

functions in several aspects, especially generalized

functions possess derivatives of any order. Generalized functions incorporate local integrable functions,

the so-called regular generalized functions, while the

so-called singular generalized functions (e.g. Dirac

126

delta function) can be obtained as the limits of a sequence of sufficient smooth functions, so-called regular sequence in generalized function context. This

makes it possible to use these singular generalized

functions through approximation. Nevertheless, a

significant contribution of this paper is to establish the

convergence rate of regular sequence to nonsmooth

loss that answers a call from the relevant literature.

For parameter estimation where objective function

may be nonsmooth, this talk first shows as a general

paradigm that how generalized function approach can

be used to tackle the nonsmooth loss functions using a

very simple model. This approach is of general interest and applicability. We further use the approach in

robust M-estimation for additive single-index cointegrating time series models; the asymptotic theory is

established for the proposed estimators. We evaluate

the finite-sample performance of the proposed estimation method and theory by both simulated data and an

empirical analysis of predictive regression of stock

returns.

A Consistent Specification Test for Expectile Models

Xiaojun Song

Peking University

Abstract: In this article, we propose a nonparametric

test for the correct specification of parametric expectile models over a continuum of expectile levels. The

test is based on continuous functionals of a residual

marked empirical process. Its asymptotic null distribution is derived. We then show that the test is consistent and has nontrivial power against a sequence of

local alternatives approaching the null at the parametric rate. Since the asymptotic null distribution of the

test statistic is not pivotal, we propose a simple multiplier bootstrap procedure to approximate the critical

values. A Monte Carlo study shows that the asymptotic results provide good approximations for small sample sizes.

Joint work with Yang Zixin.

Invited Session IS020: Foundation Models in

Modern Industries

Current Status and Prospects of Statistics and AI

in Industry

统计学、AI 在业界的现状和展望

Kaixian Yu

Insilicom

摘要: 在当前的互联网电商和生物医学领域，统计

学和 AI 的应用正展现出巨大的潜力和影响力。统

计学不仅为数据分析提供了坚实的基础，同时也为

AI 的算法开发提供了必要的理论支持。AI 技术，

尤其是机器学习和深度学习，已经在这些领域中实

现了自动化处理和智能决策，极大地提升了效率和

精度。在互联网平台上，AI 通过用户行为数据分析

，能够实现个性化推荐和内容定制，从而增强用户

体验和平台的吸引力。电商行业利用 AI 进行库存

管理、需求预测和客户服务优化，使得供应链更加

高效和响应迅速。生物医学领域则见证了 AI 在疾

病预测、新药研发和个性化治疗方案中的应用，显

著提高了医疗服务的质量和准确性。在这次演讲中

，我将分享一些关于统计学和 AI 在互联网和生物

医学领域中的实际应用案例，以及在当前这个大数

据和大语言模型时代，这些技术可能开拓的新应用

场景。

Applications and Developments of Artificial Intelligence in Medical Imaging and Clinical Practice

人工智能在医学影像及临床中的应用与发展

Xiaohuan Cao

ShanghaiTech University

摘要: AI 应用于医学影像分析，正在向更高效、更

精准、更安全的数字化、智能化方向发展。大模型

技术的引入为医学影像和临床问题分析带了巨大

的变化。通过人工智能及大模型技术，能够快速准

确的实现医学影像分析，目前已经有非常多的医学

影像 AI 产品落地。报告主要介绍人工智能在赋能

临床应用、赋能影像设备、赋能临床研究等方面

AI 技术的研发进展与产品化情况，以及医疗 AI 在

实际落地中的挑战与突破。

AI for Drug Discovery - Virtual Screening and

Foundation Model

Zheng Wang

Alibaba

Abstract: Artificial intelligence, particularly through

the self-supervised pre-training of foundational mod-

127

els, has achieved remarkable success in various application domains, including natural language process

and computer vision. Recently, there has been a

growing trend to adapt this powerful technique to

other scientific domains. Drug discovery is one of

such areas which is reaching a bottleneck with traditional solutions and is in urgent need of innovation. In

this presentation, we address the critical challenge of

ligand screening in the early stages of drug discovery.

We construct a highly efficient large-scale virtual

screening system, including large-scale molecule

management, advanced molecular docking techniques

coupled with acceleration methods, and targeted active screening strategies. Using this system, we could

successfully execute the world's largest virtual

screening test within a feasible timeframe. Additionally, we introduce a novel potential end-to-end solution based on our multi-model biomolecule foundation

model. It is our aspiration that these works will spark

further research and innovation in this field.

Invited Session IS029: Interface of Functional Data

Analysis and Dynamic Models

Privacy-Preserving Community Detection for Locally Distributed Multiple Networks

Shujie Ma

University of California, Riverside

Abstract: In this talk, I will introduce a new efficient

and scalable consensus community detection approach

and distributed learning algorithm in a multi-layer

stochastic block model using locally stored network

data with privacy-preserving. Specifically, we develop

a spectral clustering-based algorithm named ppDSC.

To reduce the bias incurred by the randomized response (RR) mechanism for achieving differential

privacy, we develop a two-step bias adjustment procedure. To reduce the communication cost encountered in distributed learning, we perform the eigen-decomposition locally and then aggregate the local

eigenvectors using an orthogonal Procrustes transformation. We establish a novel bound on the misclassification rate of ppDSC. The new bound reveals the

asymmetric roles of the two edge-flipping probabilities of the RR in the misclassification rate. Through

the bound, we can also find the optimal choices for

the flipping probabilities given a fixed privacy budget.

Moreover, we show that ppDSC enjoys the same statistical error rate as its centralized counterpart, when

the number of machines satisfies a polynomial order

with the sample size on each local machine and the

effective heterogeneity is well controlled.

Dynamic Models Augmented by Hierarchical Data

Le Bao

The Pennsylvania State University

Abstract: Dynamic models have been successfully

used in producing estimates of HIV epidemics at the

national level due to their epidemiological nature and

their ability to estimate prevalence, incidence, and

mortality rates simultaneously. Recently, HIV interventions and policies have required more information

at sub-national levels to support local planning, decision making and resource allocation. Unfortunately,

many areas lack sufficient data for deriving stable and

reliable results, and this is a critical technical barrier

to more stratified estimates. One solution is to borrow

information from other areas within the same country.

However, directly assuming hierarchical structures

within the HIV dynamic models is complicated and

computationally time-consuming. We propose a simple and innovative way to incorporate hierarchical

information into the dynamical systems by using auxiliary data. The proposed method efficiently uses information from multiple areas within each country

without increasing the computational burden. As a

result, the new model improves predictive ability and

uncertainty assessment.

Joint work with Xiaoyue Niu, Tim Brown, Jeffrey

Imai-Eaton.

An SPDE Approach to Variational Inference for

Log-Gaussian Cox Process

Nan Zhang

Fudan University

Abstract: Log-Gaussian Cox process (LGCP) provides a flexible statistical framework for modeling

spatial point patterns. Its computation for large scale

datasets is challenging due to the nested stochastic

128

structure of the Poisson process and the lack of a

closed-form posterior distribution for model parameters. In this work, we take a variational inference perspective for LGCP model fitting and derive an explicit

expression of the evidence lower bound. After reformulating the optimization problem as Poisson likelihood maximization, we develop fast approximate

inference via the stochastic differential equation

(SPDE) approach from INLA. Furthermore, we design

a coordinate ascent algorithm and establish its global

convergence. Finally, the superior performance of our

method is demonstrated in both simulation and real

data, particularly in cases with complex boundaries.

Invited Session IS033: Modeling and Statistical

Inference of Medical Big Data

A Distributed Ensemble Algorithm for Learning

Causal Bayesian Network Structure from

Large-Scale Data

Fuzhong Xue

Shandong University

Abstract: With the advent of the information age, the

volume of data has been increasing exponentially.

When attempting to uncover causal relationships from

such large-scale data, traditional Bayesian network

structure learning methods face limitations, including

insufficient computational power, memory constraints,

and high computational costs. Causal Bayesian networks rely on conditional independence tests, but as

the sample size and number of variables increase, the

speed of these tests decreases, further restricting their

application to large datasets. To address these challenges, we introduce a distributed ensemble-based

causal Bayesian network structure learning algorithm

(PE_CausalBN). This algorithm automatically learns

the appropriate size of data slices, facilitating parallel

learning both within and across multiple physical

machines. It employs a two-stage integration process,

first combining local networks from individual machines into a cohesive local network, and then merging these across multiple machines to form the final

Bayesian network (BN). PE_CausalBN effectively

reduces the computational cost of learning causal

networks from large-scale data. It can also be extended for multi-center integration, significantly reducing

communication costs and protecting data ownership.

Multidimensional Aging Prediction Model Based

on Proteomics

基于蛋白质组学的多维度衰老预测模型研究

Yongyue Wei

Peking University

摘要: 随着生物信息学和蛋白质组学技术的发展，

蛋白质组学被广泛用于探究衰老的机制和预测生

物年龄。衰老是生命过程中不可避免的一部分，而

生物年龄的准确预测对理解衰老机制具有重要意

义。然而，目前没有对生物年龄的统一测量标准。

本研究旨在利用蛋白质组学数据，结合可解释性机

器学习模型和潜在生长曲线模型，深入探究蛋白质

组学对衰老的影响，并构建衰老动态预测模型，以

期为未来抗衰老研究和疾病风险评估提供新的视

角和方法。我们定义生物年龄为未患有衰老相关疾

病（如：癌症，慢性肝病，阿尔茨海默症等）的人

的真实年龄，蛋白质组学数据来自 UK biobank 的

白种人群。通过功能聚类，将蛋白质分为若干亚群

以体现不同器官富集（如脑组织特异性蛋白），或

不同生物学功能（如免疫相关蛋白）。运用

Boruta-SHAP 从各蛋白亚群中筛选出年龄相关蛋白

质标志物，进而构建不同组织和生物学功能特异性

衰老预测模型，并将其综合为统一的多维度衰老预

测模型。将在不同种族的蛋白质组学数据中验证该

模型组的准确性和泛化性。进一步评价预测年龄与

未来多种疾病风险的关联性。

Joint work with Yuxin Song.

Study on Causal Association Between Abnormal

Renal Function Indicators and Carotid Plaque

Formation

肾功能指标异常与颈动脉斑块发生的因果关联研

究

Lixin Tao

Capital Medical University

摘要: 目的：基于反事实框架，应用因果森林联合

目标极大似然估计研究肾功能主要指标——肾小

球滤过率（eGFR）的单时点二分类暴露对颈动脉

斑块发生的因果效应；同时应用因果森林联合逆概

率加权计算 eGFR 及多时点 eGFR 斜率的多分类和

连续型暴露对颈动脉斑块发生的个体及平均因果

129

效应，从而评价单时点 eGFR 与多时点 eGFR 斜率

水平的临床价值，并为因果推断研究提供新思路。

方法：本研究基于北京健康管理队列，共纳入 4173

名研究对象，观察结局为颈动脉斑块的发生。暴露

因素为单时点 eGFR（二分类、多分类和连续型）

和多时点 eGFR 斜率（多分类和连续型）。研究步

骤为：（1）基于 Cox 回归模型和限制性立方样条，

初步探索单时点 eGFR 暴露、多时点 eGFR 斜率与

结局关联性，并在基础模型中分别加入单时点

eGFR 暴露和多时点 eGFR 斜率，比较模型的净重

分类改善指数（NRI）、综合判别改善指数（IDI）

以及 AUC 值，评价单时点以及多时点斜率的模型

预测改善能力；（2）基于反事实框架，在因果森

林中嵌入目标极大似然估计，同时平衡估计值的方

差与偏差，探索二分类 eGFR 暴露与结局的因果效

应；（3）基于因果森林模型联合广义逆概率加权

研究单时点 eGFR 和多时点 eGFR 斜率的多分类及

连续型暴露对结局的因果效应。最后综合比较单时

点 eGFR 和多时点 eGFR 斜率的预测价值和因果效

应。

结果：（1）Cox 回归分析显示单时点 eGFR 值与颈

动脉斑块发生的 HR（95%CI）为 0.808（0.752, 0.867

）；多时点 eGFR 斜率与颈动脉斑块发生的 HR（

95%CI）为 0.974（0.967, 0.983），并且限制性立方

样条分析显示存在剂量反应关系。（2）预测模型

研究显示，在基础模型上增加单时点 eGFR 的 NRI

（95%CI）为 0.1333 （0.0602, 0.2063）、IDI（95%CI

）0.0064（0.0036, 0.0092）、AUC（95%CI）为 0.7109

（0.6927, 0.7290）；在基础模型基础上增加多时点

eGFR 斜率的 NRI（95%CI）为 0.1675（0.0946, 0.2405

）、IDI（95%CI）为 0.0089（0.0055, 0.0123）、

AUC（95%CI）为 0.7121（0.6940, 0.7301）。以增

加单时点 eGFR 的预测模型为参照，增加多时点

eGFR 斜率预测模型的 NRI（95%CI）为 0.0961（

0.0230, 0.1692）、IDI（95%CI）为 0.0025（0.0007,

0.0043）。多时点 eGFR 斜率与单时点 eGFR 相比，

对颈动脉斑块发生有更强的预测能力。（3）因果

森林模型结果显示，二分类单时点 eGFR 暴露对颈

动脉斑块发生的平均因果效应 OR（95%CI）为 1.281

（1.101，1.490）；以 eGFR≥90 mL/min/1.73m2 为

参 照 组 ， 多 分 类 单 时 点 eGFR 暴 露 [60,90)

mL/min/1.73m2 组以及<60 mL/min/1.73m2 组的 OR

（95%CI）分别为 1.046（1.020，1.074）和 1.194

（1.001，1.426）；连续型单时点 eGFR 暴露的 OR

（95%CI）为 0.959（0.946，0.972）。以 Q4 为参

照组，多分类多时点 eGFR 斜率 Q1、Q2 和 Q3 组

对颈动脉斑块发生的平均因果效应 OR（95%CI）

分别为 1.143（1.104, 1.184），1.090（1.054, 1.127）

和 1.056（1.022, 1.090）；连续型多时点 eGFR 斜

率的平均因果效应 OR（95%CI）为 0.994（0.993,

0.996）。个体因果效应值进一步证明了 eGFR 对颈

动脉斑块的保护作用。

结论：本研究利用因果森林模型联合靶向极大似然

估计和广义逆概率加权模型，发现单时点 eGFR 和

多时点 eGFR 斜率对颈动脉斑块的发生均存在因果

关联。因果森林模型可以有效处理高维和复杂的数

据结构，有效避免“维度灾难”，并且该方法具有避

免过拟合和偏差，提高泛化能力的优势，可以更准

确地评估暴露因素的因果效应，在医学研究中有进

一步推广的价值。此外，研究结果表明，在基础模

型上增加多时点 eGFR 斜率会显著改善模型的预测

能力，且优于单时点测量的 eGFR。这提示在临床

实践中要注重长期监测 eGFR 水平，而不能只关注

单时点的 eGFR 水平。长时间保持合理的 eGFR 水

平利于预防颈动脉斑块的发生。

Joint work with Rui Jin, Jinlin Wang.

Mid-to Late-Life Trajectories of Depressive Symptoms and Risk of Cardiovascular Disease Mortality:

A Joint Latent Class Mixed Models

Haibin Li

Beijing Chaoyang Hospital of Capital Medical University

Abstract: Background: Depressive symptoms frequently fluctuate over the life course, but studies examining the association between depressive symptoms

and cardiovascular disease (CVD) are mostly based

on single or average depressive symptoms levels.

However, this approach does not take into account

long-term trajectories of depressive symptoms, which

can vary considerably in the elderly. This study aims

to investigate the association of long-term trajectories

of depressive symptoms with the risk of CVD mortality.

Methods: Within the community-based Beijing Longitudinal Study of Ageing, we examined trajectories

of depressive symptoms in 2,101 participants (50.7%

women) over an age-range from 55 to 95 years and

130

jointly modeled their risk of CVD mortality and competing causes of non-CVD mortality using joint latent

class mixed modeling. Depressive symptoms were

assessed on 9 occasions between 1992 and 2017 using

the Center for Epidemiology Depression Scale and the

Geriatric Depression Scale. Death and underlying

causes of death were ascertained by linkage to death

records through December 31, 2017. Lifetime risks to

75 and 95 years of age were estimated for the assigned

trajectory, with non-CVD mortality as a competing

event. We then calculated hazard ratios (HR) for CVD

mortality by assigned trajectory using cause-specific

Cox regression models.

Results: We identified three trajectories of depressive

symptoms in these 2101 individuals, characterized by

maintained low scores (low trajectory; 1757

[84.43%]); moderately high starting scores, increasing,

but then remitting (decreasing trajectory; 87 [4.18%]);

low starting scores that sharply increased (increasing

trajectory; 237 [11.39%]). During 22,250 person-years

of follow-up, 1365 deaths occurred including 616

deaths from CVD. At 55 years of age, lifetime risks to

75 years of age were 19.10% (95% CI, 13.87 to 24.62)

for low trajectory, 28.52% (95% CI, 13.87 to 59.49)

for decreasing trajectory, and 27.22% (95% CI, 11.03

to 42.60) for individuals with high trajectory of depressive symptoms, respectively. The corresponding

lifetime risks to 95 years of age were 47.93% (95% CI,

43.46 to 53.31), 37.84% (95% CI, 17.55 to 73.62),

and 40.98% (95% CI, 25.68 to 62.31). Compared with

the consistently low trajectory, having a decreasing

and increasing depressive symptom trajectory was

significantly associated with an increased risk of CVD

mortality (HR, 1.45; 95% CI, 1.20-1.76 and 2.76; 95%

CI, 2.46-3.10).

Conclusions: Lifetime Risk of CVD mortality differed

with different courses of depressive symptoms. Older

adults with a longitudinal pattern of increasing depressive symptoms were at high risk for CVD mortality. Trajectories of depressive symptoms provide a

more nuanced understanding of the associations between depressive symptoms and CVD mortality.

Joint work with Xiuhua Guo.

Contributed Session CS025：Complex Data Modeling

Calibration for Computer Models with

Time-Varying Parameter

Yang Sun

Peking University

Abstract: Traditional methods for computer model

calibration have typically focused on achieving an

optimal constant parameter, which assumes that the

parameter remains fixed regardless of changes in input variables. However, in practical applications, calibration parameters can often vary with specific subsets of input variables. Under such circumstances,

existing methods may struggle to provide accurate

predictions of the true underlying process, even if they

do manage to identify an optimal parameter value. In

this article, we consider the calibration problem of

computer models with time-varying parameter. Inspired by the idea of profile least square, we obtain a

pointwise estimator of the calibration parameter via

the local linear smoother. Furthermore, we derive the

theoretical properties including the consistency and

the asymptotic normality of the proposed estimators.

Two numerical examples including simulated and

practical models are employed to illustrate good performance of the proposed method. At last, we apply

the proposed procedure to calibrate forward model in

NASA’s OCO-2 mission, the results further illustrate

the feasibility and effectiveness of our proposed

method.

Joint work with Xiangzhong Fang.

Large-Scale Metric Objects Filtering for Binary

Classification

Shuaida He

Southern University of Science and Technology

Abstract: The classification of random objects within

metric spaces without a vector structure has attracted

increasing attention. However, the complexity inherent in such non-Euclidean data often restricts existing

models to handle only a limited number of features,

leaving a gap in real-world applications. In this talk,

we introduce a data-adaptive filtering procedure to

identify informative features from a large set of ran-

131

dom objects, leveraging a novel Kolmogorov-Smirnov

type statistic defined on the metric space. Our method,

applicable to data in general metric spaces with binary

labels, exhibits remarkable flexibility. It enjoys a

model-free property, as its implementation does not

rely on any specified classifier. Theoretically, it guarantees the sure screening property. Empirically,

equipped with a Wasserstein metric, it demonstrates

superior performance compared to many Euclidean

competitors across various settings.

Joint work with Jiaqi Li, Xin Chen.

Penalized Exponentially Tilted Likelihood for

Growing Dimensional Models with MAR Missing

Data

Xiaoming Sha

Yunnan University

Abstract: We conduct statistical inference on growing

dimensional models with missing at random (MAR)

data and propose penalized exponentially tilted (ET)

likelihood to estimate the parameters of interest and

select variables. In the case of MAR-missing response

variables, we use inverse probability weighted approach to compensate for the missing information and

ensure the consistency of parameter estimation, then

construct an ET likelihood ratio statistic. Under the

assumption of sparse models, a penalty function is

added to the ET likelihood ratio, which is robust to

model misspecification. Under certain regularization

conditions, the consistency, asymptotic properties and

oracle properties of the target parameters are obtained,

and we show that the constrained penalized ET likelihood ratio statistic satisfies Wilks' theorem, that is, it

asymptotically follows the chi-squared distribution.

Simulations are conducted by three different penalty

functions in various models to validate the performance of proposed methodologies and show the stability of our methods. An example of thyroid data

from the First People's Hospital of Yunnan Province is

illustrated.

Joint work with Niansheng Tang, Puying Zhao.

A Novel Integration Method for Single Cell Transcriptomics and Proteomics Data Based on Transformer

基于Transformer的单细胞转录组学与蛋白质组学

数据新型整合方法

Jiankang Xiong

Chinese Academy of Sciences

Abstract: The advent of single-cell RNA sequencing

(scRNA-seq) has revolutionized our understanding of

cellular heterogeneity. However, scRNA-seq captures

only a single layer of the complex regulatory networks

that govern cellular function. Cellular indexing of

transcriptomes and epitopes by sequencing (CITE-seq)

represents a significant advancement, allowing for the

simultaneous quantification of RNA expression and

surface protein abundance in single cells. Despite

these advances, there remains a critical need for sophisticated computational methods that can effectively

integrate and analyze single-cell multi-omics data,

particularly in the context of batch effects and missing

data, to fully harness the potential of CITE-seq.

We introduce scTransporter, a novel single-cell transcriptomics and proteomics data integration model

based on transformer. scTransporter leverages the

power of the Transformer architecture's Attention

mechanism to model the intricate intercellular interactions and the multi-omics data's complexity. Our

model not only integrates RNA and ADT data within

CITE-seq datasets but also effectively mitigates batch

effects when combining multiple datasets. Furthermore, scTransporter demonstrates superior performance in imputing missing ADT data across batches,

outperforming existing methods such as totalVI and

Seurat v5. Particularly, scTransporter's specialized

version for spatial CITE-seq data integrates spatial

coordinates, significantly enhancing the accuracy of

integrating spatially resolved RNA and ADT data. The

results of our study showcase scTransporter's potential

to become a cornerstone in the analysis of single-cell

transcriptomics and proteomics data, paving the way

for new insights into cellular identity and function.

Joint work with Liang Ma, Lin Wan.

Estimation of Spatial Weight Matrix in Spatial

Error Autocorrelation Models with Non-zero Autocorrelation Random Perturbations

132

具有非零自协方差随机扰动的空间误差自相关模

型中的空间权重矩阵的估计

Zhaohui Man

Ningxia University

Abstract: 在空间误差自相关模型中，将条件“随机

扰动关于空间位置独立”一般化，在“未必空间独立”

的情形下，研究将空间权重矩阵视为未知参数进行

估计的方法。分析了空间权重矩阵的自由度，给出

了估计量，提出了将其转化为优化问题求解的具体

方法，分两种情形数值模拟，最后给出了应用实例

。在此过程中，发现了克服估计量计算困难的实用

方法，发现了空间权重矩阵解释的有效性问题并给

出了解决方案。推进了空间权重矩阵内生构建法的

研究。

Joint work with Yanbing Fang.

Contributed Session CS026：Change-Points Detection

Filtrated Common Functional Principal Component Analysis of Multi-group Functional Data

Shuhao Jiao

City University of Hong Kong

Abstract: Local field potentials (LFPs) are signals

that measure electrical activities in localized cortical

regions, and are collected from multiple tetrodes implanted across a patch on the surface of cortex. Hence,

they can be treated as multi-group functional data,

where the trajectories collected across temporal

epochs from one tetrode are viewed as a group of

functions. In many cases, multi-tetrode LFP trajectories contain both global variation patterns (which are

shared by most groups, due to signal synchrony) and

idiosyncratic variation patterns (common only to a

small subset of groups), and such structure is very

informative to the data mechanism. Therefore, one

goal in this paper is to develop an efficient algorithm

that is able to capture and quantify both global and

idiosyncratic features. We develop the novel filtrated

common functional principal components (filt-fPCA)

method which is a novel forest-structured fPCA for

multi-group functional data. A major advantage of the

proposed filt-fPCA method is its ability to extract the

common components in a flexible \"multi-resolution\"

manner. The proposed approach is highly data-driven

and no prior knowledge of \"ground-truth\" data structure is needed, making it suitable for analyzing complex multi-group functional data. In addition, the

filt-fPCA method is able to produce parsimonious,

interpretable, and efficient functional reconstruction

(low reconstruction error) for multi-group functional

data with orthonormal basis functions. Here, the proposed filt-fPCA method is employed to study the impact of a shock (induced stroke) on the synchrony

structure of rat brain. The proposed filt-fPCA is general and inclusive that can be readily applied to analyze any multi-group functional data, such as multivariate functional data, spatial-temporal data and longitudinal functional data.

Joint work with Ron Frostig, Hernando Ombao.

Change-Points Detection and Support Recovery

for Spatiotemporal Functional Data

Decai Liang

Nankai University

Abstract: Large volumes of spatiotemporal data,

including patterns of climatic variables, satellite images and FMRI data, usually exhibit inherent mean

changes. Due to the complicated cross-covariance

structure, the full covariance function is commonly

described as a product of independent spatial covariance and temporal covariance, which is a mathematically convenient yet not always reflective assumption

of the data. To remedy this, we propose a novel hypothesis test based on a more realistic assumption

known as weak separability. We establish solid asymptotic theory to support this approach. Furthermore,

we develop a comprehensive procedure for support

recovery amidst the intricate correlations between

space and time, effectively identifying true signals

(locations with mean change) while controlling the

false discovery rate. This represents the first work of

support recovery within a spatiotemporal framework.

Simulation studies and a Chinese precipitation data

application validate the efficacy and enhanced power

of our methodology on both change point detection

and support recovery.

Moment Selection and Generalized Empirical

133

Likelihood Estimation in High-Dimensional Unconditional Moment Conditions

Zhihuang Yang

Yunnan University

Abstract: This paper investigates the moment selection and parameter estimation problem of

high-dimensional unconditional moment conditions.

First, the authors propose a Fantope projection and

selection (FPS) approach to distinguish the informative and uninformative moments in high-dimensional

unconditional moment conditions. Second, for the

selected unconditional moment conditions, we present

a generalized empirical likelihood (GEL) approach to

estimate unknown parameters. The proposed method

is computationally feasible, and can efficiently avoid

the well-known ill-posed problem of GEL approach in

the analysis of high-dimensional unconditional moment conditions. Under some regularity conditions,

the authors show the consistency of the selected moment conditions, the consistency and asymptotic normality of the proposed GEL estimator. Two simulation

studies are conducted to investigate the finite sample

performance of the proposed methodologies. The

proposed method is illustrated by a real example.

Joint work with Wenjun Wang.

On Robust Estimation of Hidden Semi-Markov

Regime-Switching Models

Shanshan Qin

Tianjin University of Finance and Economics

Abstract: Regime-switching models provide an efficient framework for capturing the dynamic behavior

of data observed over time and are widely used in

economic or financial time series analysis. In this

paper, we propose a novel and robust hidden

semi-Markovian regime-switching (rHSMS) method.

This method uses a general ho-based distribution to

correct for data problems that contain atypical values,

such as outliers, heavy-tailed or mixture distributions.

Notably, the rHSMS method enhances not only the

scalability of the distribution assumptions for all regimes, but also the scalability to accommodate arbitrary sojourn types. Furthermore, we develop a likelihood-based estimation procedure coupled with the use

of the EM algorithm to facilitate practical implementation. To demonstrate the robust performance of the

proposed rHSMS method, we conduct extensive simulations under different sojourns and scenarios involving atypical values. Finally, we validate the effectiveness of the rHSMS method using monthly returns

of the S&P500 Index and the Hang Seng Index. These

empirical applications demonstrate the utility of the

rHSMS approach in capturing and understanding the

complexity of financial market dynamics.

Joint work with Zhenni Tan, Yuehua Wu.

Interaction Tests with Covariate-Adaptive Randomization

Likun Zhang

Renmin University of China

Abstract: Treatment-covariate interaction tests are

commonly applied by researchers to examine whether

the treatment effect varies across patient subgroups

defined by baseline characteristics. The objective of

this study is to explore treatment-covariate interaction

tests involving covariate-adaptive randomization.

Without assuming a parametric data generating model,

we investigate usual interaction tests and observe that

they tend to be conservative: specifically, their limiting rejection probabilities under the null hypothesis do

not exceed the nominal level and are typically strictly

lower than it. To address this problem, we propose

modifications to the usual tests to obtain corresponding valid tests. Moreover, we introduce a novel class

of stratified-adjusted interaction tests that are simple,

more powerful than the usual and modified tests, and

broadly applicable to most covariate-adaptive randomization methods. The results encompass two types

of interaction tests: one involving stratification covariates and the other involving additional covariates

that are not used for randomization. Our study clarifies the application of interaction tests in clinical trials

and offers valuable tools for revealing treatment heterogeneity, crucial for advancing personalized medicine.

Joint work with Wei Ma.

Contributed Session CS027：Nonparametric Sta-

134

tistical Inference

Nonparametric Estimation of Regression Function

and Error Covariance Function with Unknown

Correlation Structure

Sisheng Liu

Hunan Normal University

Abstract: In practice, it is common for errors to be

positively correlated in a nonparametric regression

model. Without considering the positive error correlation, it is well-known that the selected smoothing

parameter can lead to undersmoothing. In this article,

we propose a procedure for selecting the bandwidth in

local linear regression to estimate the regression function in the presence of unknown correlated errors.

Additionally, we nonparametrically estimate the error

covariance function. Theoretical support for the procedure of bandwidth selection and error covariance

estimation is provided. Simulations demonstrate a

clear improvement in the estimation of the regression

function and error covariance when the correlation

function is not strongly dominated by correlations

near zero. We illustrate the methodological applications of our procedure in nonparametric partial derivative estimation, partial linear models, and kernel ridge

regression in the presence of correlated errors. Finally,

we present real data applications of our procedure

using a temperature dataset from Southern China.

Complex Representation Matrix of Third-Order

Quaternion Tensors with Application to Video

Inpainting

Fengsheng Wu

Yunnan University

Abstract: Quaternion tensors have been widely applied in various scientific and engineering fields due

to their ability to capture internal correlations and

structural features in multi-channel and multimodal

data. However, it is undeniable that the multiple imaginary components in quaternion and their

non-commutative multiplication make operations on

high-dimensional quaternion tensors extremely challenging. In this talk, we define a complex representation matrix (CRM) for third-order quaternion tensors

under a given algebraic multiplication and discuss its

relevant properties. Based on the discrete Fourier

transform, the proposed CRM is unique, allowing for

the conversion of quaternion tensor optimization

problems into the complex number domain. This simplifies the computational process while preserving the

advantages of quaternion modeling. Additionally, the

proposed CRM possesses a block-diagonal structure,

implying its capability for efficient and parallel computations for large-scale problems. Quaternion tensor

techniques can be used to restore missing color video

data that may occur during collection and transmission. To this end, we propose two CRM-based quaternion tensor completion methods to evaluate the advantages of CRM in optimizing solutions for video

inpainting problems. Simulation studies and real example analysis were conducted to demonstrate the

effectiveness and efficiency of the proposed methods.

Joint work with Chaoqian Li, Yaotang Li, Niansheng

Tang.

Generalized Rank Regression with Multiplier

Bootstrap

Jiyuan Tu

Shanghai University of Finance and Economics

Abstract: Rank regression (RR) and composite quantile regression (CQR) are two classical regression

methodologies known for their robustness and statistical efficiency. Despite the acknowledge similarity in

their statistical efficiency, the precise relationship

between them has not been extensively investigated in

the existing literature. This article aims to elucidate

the nuanced connection between composite quantile

regression and rank regression. Furthermore, we extend RR to a broader class of loss functions, termed as

the generalized rank regression (GRR), and reveals

the equivalence of asymptotic variance between CQR

and GRR, establishing a direct transformation between them through a simple formula. Additionally,

we delve into a more intricate analysis of GRR, introducing a Bahadur representation and establishing

asymptotic normality under mild conditions. To address efficient solutions for GRR, we propose a

two-stage sub-gradient descent algorithm, which ensures convergence to parameters with desired statisti-

135

cal properties, despite the non-smooth and

non-convex nature of the loss function. Additionally,

we develop a multiplier bootstrap method in conjunction with sub-gradient descent for conducting statistical inference. The superiority of our methods is substantiated through comprehensive simulation studies

and real-world applications.

Joint work with Yichen Zhang, Suqi Wu.

Bayesian Generalized Method of Moments Inference for Clustered Samples with Missing Values

Gen Ye

Yunnan University

Abstract: This paper aims to develop a unified

Bayesian approach to clustered data analysis when

observations are subject to missingness at random. We

consider a more general framework that the parameters of interest are defined through the generalized

estimating equations and the probability of missingness is of general parametric form. The framework of

generalized method of moments is employed to seek

an optimal combination of the inverse-probability

weighted estimating equations for parameter of interest and the score estimating equations for propensity

scores. We utilize this framework to develop a semiparametric Bayesian analysis of clustered samples

with missing values. A unified model selection approach is also developed to compare models defined

by different moment conditions. We systematically

evaluate large sample properties of the proposed quasi-posterior density with any fixed or shrinking priors,

and also establish selection consistency of the proposed model selection criterion. Our results are valid

under very mild conditions and have major advantages on parameters defined through non-smooth

estimating functions. Extensive numerical results

demonstrate that the proposed method works remarkably well for finite samples.

Bayesian INLA Estimation and Application of

Generalized VCSA Model

广义 VCSA 模型的 Bayesian-INLA 估计及应用

Lijun Meng

Xinjiang University of Finance & Economics

摘要: 本文以广义空间自相关变系数模型为研究对

象。考虑了数据的空间自相关性，并允许协变量系

数在空间上变化。尽管对空间数据模型已进行相关

研究，但使用贝叶斯框架下的 Bayesian-INLA 算法

的文章尚不多见。因此，本文使用了最近提出的

Bayesian-INLA 算法，该算法具有高效和精确的计

算性质，能够对模型的参数和非参数进行估计，并

得到它们的后验分布。这为空间数据模型的贝叶斯

分析提供了一个新的方法和框架。通过中国艾滋病

影响因素数据的实证分析，展示了本文所提估计方

法的实用价值。

Contributed Session CS028：Recent Advances in

Large-Scale Data

Moment-Assisted Subsampling Method for Maximum Likelihood Estimator with Large-Scale Data

Miaomiao Su

Beijing University of Posts and Telecommunications

Abstract: The maximum likelihood estimator for a

parametric conditional density model can be computationally cumbersome for large-scale datasets, especially when the likelihood function requires integral

computations. Subsampling offers an efficient solution for this problem. This paper proposes a moment-assisted subsampling method which can improve the estimation efficiency of any estimating

equation-based subsampling estimators. Some sample

moments can be efficiently computed even if the

sample size of the whole dataset is huge. The proposed method incorporates informative sample moments of the whole data through the generalized

method of moments. The resulting subsampling estimator can be computed rapidly and is asymptotically

normal with a smaller asymptotic variance than the

corresponding estimator without incorporating sample

moments of the whole data. The proposed method is

further extended to accommodate model misspecification. The asymptotic variance of the moment-assisted

estimator relies on the moment function of the sample

moment. We derive the optimal moment function that

minimizes the resulting asymptotic variance in terms

of Loewner order. The optimal moment function depends on the data generating process and might be

challenging to calculate. We provide its data-adaptive

136

approximation that is easy to compute. When the approximation is sufficiently accurate, the proposed

estimator can achieve the same efficiency as the

whole data-based maximum likelihood estimator.

Numerical results demonstrate the promising performance of the moment-assisted method across various

scenarios.

Joint work with Qihua Wang, Ruoyu Wang.

Subsampling and Jackknifing: A Practically Convenient Solution for Large Data Analysis with

Limited Computational Resources

Shuyuan Wu

Shanghai University of Finance and Economics

Abstract: Modern statistical analysis often encounters

datasets with large sizes. For these datasets, conventional estimation methods can hardly be used immediately because practitioners often suffer from limited

computational resources. In most cases, they do not

have powerful computational resources (e.g., Hadoop

or Spark). How to practically analyze large datasets

with limited computational resources then becomes a

problem of great importance. To solve this problem,

we propose here a novel subsampling-based method

with jackknifing. The key idea is to treat the whole

sample data as if they were the population. Then,

multiple subsamples with greatly reduced sizes are

obtained by the method of simple random sampling

with replacement. It is remarkable that we do not

recommend sampling methods without replacement

because this would incur a significant cost for data

processing on the hard drive. Such cost does not

exist if the data are processed in memory. Because

subsampled data have relatively small sizes, they can

be comfortably read into computer memory as a

whole and then processed easily. Based on subsampled datasets, jackknife-debiased estimators can be

obtained for the target parameter. The resulting estimators are statistically consistent, with an extremely

small bias. Finally, the jackknife-debiased estimators

from different subsamples are averaged together to

form the final estimator. We theoretically show that

the final estimator is consistent and asymptotically

normal. Its asymptotic statistical efficiency can be as

good as that of the whole sample estimator under very

mild conditions. The proposed method is simple

enough to be easily implemented on most practical

computer systems and thus should have very wide

applicability.

Joint work with Xuening Zhu, Hansheng Wang.

A Case Study on the Share Holder Network Effect

of Stock Market Data: An SARMA Approach

Rong Zhang

Yunnan University

Abstract: One of the key research problems in financial markets is the investigation of inter-stock dependence.

A good understanding in this regard is crucial for

portfolio optimization. To this end, various econometric models have been proposed. Most of them assume

that the random noise associated with each subject is

independent. However, dependence might still exist

within this random noise. Ignoring this valuable information might lead to biased estimations and inaccurate predictions. In this article, we study a spatial

autoregressive moving average model with exogenous

covariates. Spatial dependence from both response

and random noise is considered simultaneously. A

quasi-maximum likelihood estimator is developed,

and the estimated parameters are shown to be consistent and asymptotically normal. We then conduct an

extensive analysis of the proposed method by applying it to the Chinese stock market data.

Study on Fractional Random Drug-Resistant Tuberculosis Kinetic Model

分数阶随机耐药结核病动力学模型研究

Shaoping Jiang

Yunnan Minzu University

摘要: 结核病是目前世界上单一致病菌死亡率最高

的疾病，近 20 来全球结核病疫情有所回升，每年

约有 800 万新发结核病人，有 300 万死于结核病。

面对结核病流行的严峻局面，各国政府已采取了相

应的措施。本文在传染病动力学建模思想的基础上

，考虑了不同因素对结核病传播的影响，建立具有

Atangana-Baleanu 算子的分数阶随机耐药结核病模

型，通过牛顿多项式插值法进行数值模拟，分析改

137

变其中的敏感性参数会给耐药结核病的控制带来

什么样的影响并据此制定合理的干预措施。

Joint work with Hongyan Wang.

Identification and Estimation of the Bi-Directional

MR with Some Invalid Instruments

Zhen Yao

Beijing Technology and Business University

Abstract: We consider the challenging problem of

estimating causal effects from purely observational

data in the bi-directional Mendelian randomization

(MR), where some invalid instruments, as well as

unmeasured confounding, usually exist. To address

this problem, most existing methods attempt to find

proper valid instrumental variables (IVs) for a target

causal effect by expert knowledge or by assuming that

the causal model is a one-directional MR model. As

such, in this paper, we first theoretically investigate

the identification of the bi-directional MR from observational data. In particular, we provide necessary

and sufficient conditions under which valid IV sets are

correctly identified such that the bi-directional MR

model is identifiable, including the causal direction of

a pair of phenotypes. Moreover, based on the identification theory, we develop a cluster fusion-like method

to discover valid IV sets and estimate the causal effects of interest. We demonstrate theoretically that the

proposed algorithm is correct. Experimental results

show the effectiveness of our proposed method for

estimating causal effects in bi-directional MR.

Contributed Session CS029：Interdisciplinary and

Applied Research: Statistical Analysis on Medical

and Economic Data

Dual Cox Model Theory and Its Application in

Oncology Research

对偶 Cox 模型理论及其在肿瘤学研究的应用

Haojin Zhou

Xi'an Jiaotong-liverpool University

摘要: 靶向治疗和免疫治疗在近期的癌症治疗中取

得了突破性进展。它们的成功使得患者的异质性变

得益发重要，患者的异质性在很多临床试验中被证

明为疗效的重要预测因素。但是患者的异质性在很

大程度上导致广泛使用的等比例风险假设在一些

临床试验中被证明无效。 为了解决肿瘤临床试验

中数据分析的挑战，我们提出了一个对偶 Cox 模型

理论，包括一个对偶 Cox 模型和一个基于 EM 的拟

合算法。在常见的假设下，我们证明了模型中效应

参数估计的一致性和渐近正态性。而在数值模拟中

，该理论除了拟合算法上较强的理论性质外，还可

以很好地逼近各类生存模型，对删失率和缓解率的

变化具有相对的稳健性，在亚组分类中具有较高的

预测精度和稳定性，同时具有较快的收敛速度。在

两个特别的临床试验（KM 曲线出现交叉）上，该

理论能够确定很好地区分病人是否受益于治疗。

对偶 Cox 模型理论的提出为解决肿瘤临床试验中

患者对癌症治疗反应的异质性所带来的挑战提供

了一个有价值的工具，为解释患者反应异质性并改

善靶向治疗和免疫治疗时代的治疗疗效评估提供

了一个新的框架。

Joint work with Bowei Chen, Siyin Hu.

A Study on the Light Variation of Active Galactic

Nuclei Based on Probability Theory and Statistical

Methods

基于概率论统计方法对活动星系核光变的研究

Haiyun Zhang

Yunnan University

摘要: 基于天文领域大量观测数据的积累，概率数

值计算的方法和大数据分析技术将会极大促进天

文学的研究。高斯过程方法就是一类概率统计分析

方法，也是机器学习的重要方向。该方法在天文学

领域中已经有了较多应用，比如，凌日系外行星探

测及恒星变化、活动星系核（AGN）光变分析、引

力波探测等。我们关注的是 AGN 的光变，它在各

个电磁波段都被观测到，且光变时间跨度很大（从

分钟到年量级）。对于 AGN 这种光变的机制和起

源问题，人们虽然进行了一系列的研究，也提出了

一些相应的物理理论，但是具体物理起源仍然是未

知的。统计方法在 AGN 光变中的应用可能为这一

问题提供更多的物理信息。我们用高斯过程方法对

几十个 AGN 喷流中多波段长期光变进行了研究。

结果表明，光学、X 射线和伽马射线长期光变模式

是阻尼随机游走（DRW）模式，光变的特征阻尼时

标与吸积盘的热不稳定性时标一致。喷流的长期光

变特征与吸积盘辐射的光变特征一致。由此推断，

喷流的长期光变可能与吸积盘的热不稳定性有关，

建立了喷流和吸积盘的新联系。

138

Joint work with Dahai Yan, Li Zhang.

Static and Dynamic Connectivity Patterns of White

Matter Functional Networks in the Adult Life Cycle

成人生命周期中白质功能网络的静态和动态连通

模式

Zeqiang Linli

Guangdong University Of Foreign Studies

摘要: 人类大脑的成熟和衰老涉及复杂的生物过程，

导致大脑结构和功能的复杂变化。尽管衰老对灰质

（GM）区域内功能连接的影响已经得到了广泛的

研究，但对白质（WM）功能变化的研究仍然有限。

在这项研究中，我们的目的是使用 494 名 19-80 岁

的健康被试的静息态功能磁共振成像（rs-fMRI）数

据来研究 WM 功能动力学中与年龄相关的轨迹。首

先，通过聚类方法识别 GM 和 WM 功能网络（FNs

）。接下来，我们对 WM 功能网络连接性（FNC）

进行静态和动态分析，以探索年龄对 WM FNs 的影

响。此外，我们还研究了动态 FNC 的复发模式。

最后，我们进行了验证分析，以确保研究结果的可

复制性。我们确定了 9 个可靠的 WM 和 12 个 GM

FN。研究结果揭示了年龄相关对 WM FNC 强度和

WM-GM FNC 动力学的影响，主要包括静态 FNC

强度中的线性正和 U 形年龄轨迹，以及 FNC 时

间变异中的线性负和倒 U 形年龄轨迹。此外，我

们确定了三个不同的大脑状态与显着的年龄相关

的差异。上述发现在验证分析中得到了很大程度的

重复。WN-GM FNC 的高整合性和低时间变异性可

能反映出老年人的网络系统效率较低。这些发现增

强了我们对大脑衰老过程的理解，并为正常衰老过

程中 WM 功能动力学的轨迹提供了见解。

Joint work with Shuixia Guo.

The Direct Impact and Spatial Effect of Digital

Economy Driving Rural Residents' Service Consumption under the Background of Rural Revitalization

乡村振兴背景下数字经济驱动农村居民服务消费

的直接影响和空间效应

Jia Liu

Inner Mongolia Finance And Economics College

摘要: 农村居民消费水平的提高对于我国实体经济

与乡村振兴的发展发挥着一定作用。本文基于基准

回归模型和空间计量模型，利用 2011-2020 年的省

级面板数据，实证检验了数字经济驱动农村居民服

务消费的直接影响和空间效应。结果显示：数字经

济对农村居民的服务消费存在直接促进作用；从服

务消费的结构来看，数字经济的发展对农村居民医

疗保健消费水平的驱动作用最大、交通通信消费水

平次之、教育文化消费水平最小；分地区进行检验

可知，不同地区数字经济的发展对农村居民服务消

费的驱动作用大小依次为中部、西部、东部；最后

，通过空间杜宾模型进行检验显示，数字经济对农

村居民服务消费存在显著的正向溢出效应。本文据

此提出相应政策建议以充分发挥数字经济红利，激

发农村消费市场活力。

Joint work with Xin Wei.

Inference for ARMA Time Series with Mildly Varying Trend

Yinghuai Yi

Tsinghua University

Abstract: Statistical inference is studied for ARMA

time series with a mildly varying smooth trend. Following properly calibrated B-spline estimation of the

trend, approximates of the unobserved ARMA series

are used instead to estimate ARMA parameters by

maximum likelihood. The parameter estimates are

shown to be oracally efficient in the sense that they

are asymptotically as efficient as if the true trend

function were known and removed to obtain the true

ARMA series. The mildly varying trend function is

estimated by Nadaraya-Watson method with asymptotically correct simultaneous confidence band (SCB).

The SCB sprawls over an interval that eventually

covers the half real line, instead of a bounded interval

in existing literature, with asymptotic width of the

SCB dependent on ARMA coefficients. Simulation

experiments corroborate the theoretical findings.

Joint work with Zening Song, Lijian Yang.

Contributed Session CS030：Statistical Modeling

and Its Applications

MLEce: Statistical Inference for Asymptotically

Efficient Closed-Form Estimators in R

Jun Zhao

Ningbo University

139

Abstract: Maximum likelihood estimation is a classical method with useful properties like efficiency,

consistency, and asymptotic normality. However, the

maximum likelihood estimator (MLE) cannot be in

closed form in many distributions. Therefore, it is

obtained by iterative methods, such as the Newton–

Raphson algorithm, which is time consuming and

unrobust. For three multivariate distributions (bivariate Weibull, bivariate gamma and multivariate Dirichlet distributions) where the corresponding MLEs

are not in closed forms, the R package MLEce is developed to provide new efficient estimators with asymptotic normality and efficiency like the MLE.

Along with MLE and method of moment estimator

(MME), MLEce package conducts point estimation

(closed-form efficient estimator), random sample

generation, goodness-of-fit test, and confidence interval estimation. Simulations and real data examples are

provided to help us understand how to use MLEce.

Joint work with Yu-Kwang Kim, Yu-Hyeong Jang,

Jae Ho Chang, Sang Kyu Lee, Hyoung-Moon Kim.

Economic Forecasts Using Many Noises

Zhentao Shi

The Chinese University of Hong Kong

Abstract: This paper addresses a fundamental question in economic forecasting: Can noises help economic forecasts? Economists typically conduct variable selection to eliminate noises from predictors.

However, we prove a compelling result: in most economic forecasts, including noises in predictions yields

greater benefits than its exclusion. Furthermore, if the

total number of predictors is not sufficiently large,

intentionally adding more noises yields superior forecast performance, outperforming benchmark predictors relying on dimension reduction. The intuition lies

in economic predictive signals being densely distributed among regression coefficients, maintaining modest forecast bias while diversifying away overall variance. Therefore, economic forecasts can significantly

benefit from the \"benign overfitting\" even if a significant proportion of predictors constitute pure noises.

One of our empirical demonstrations shows that intentionally adding 300~6000 pure noises to the Welch

and Goyal (2008) comprehensive dataset achieves a

noteworthy 10% out-of-sample R2 accuracy in forecasting the annual U.S. equity premium. The performance surpasses the majority of sophisticated machine learning models.

Joint work with Yuan Liao, Xinjie Ma, Andreas

Neuhierl.

Standard Representation of Order of Addition

Design and Rapid Isomorphism Comparison

Bing Guo

Sichuan University

Abstract: Experimental design aims to obtain data

through scientifically designed experiments, maximizing information acquisition with limited resources,

thus achieving efficient statistical inference. Isomorphic designs are equivalent in terms of statistical inference capabilities, but how to quickly identify representative designs that are non-isomorphic among

numerous designs has always been a research focus.

For order of addition experimental design, in this talk,

we propose to use the minimum representation under

lexicographic order as the standard design of the isomorphic class. This representation is both unique and

effective in determining isomorphism, providing an

efficient method for isomorphism verification in order

of addition experimental design.

Survival Mixed Membership Block Model

Fangda Song

The Chinese University of Hong Kong (Shenzhen)

Abstract: Whenever we send a message via a channel

such as E-mail, Facebook, WhatsApp, WeChat, or

LinkedIn, we care about the response rate—the probability that our message will receive a response—and

the response time—how long it will take to receive a

reply. Recent studies have made considerable efforts

to model the sending behaviors of messages in social

networks with point processes. However, statistical

research on modeling response rates and response

times on social networks is still lacking. Compared

with sending behaviors, which are often determined

by the sender’s characteristics, response rates and

response times further depend on the relationship

140

between the sender and the receiver. Here, we develop

a survival mixed membership blockmodel (SMMB)

that integrates semiparametric cure rate models with a

mixed membership stochastic blockmodel to analyze

time-to-event data observed for node pairs in a social

network, and we are able to prove its model identifiability without the pure node assumption. We develop a

Markov chain Monte Carlo algorithm to conduct posterior inference and select the number of social clusters in the network according to the conditional deviance information criterion. The application of the

SMMB to the Enron E-mail corpus offers novel insights into the company’s organization and power

relations. Supplementary materials for this article are

available online.

Joint work with Jing Chu, Shuangge Ma, Yingying

Wei.

Cox Proportional Hazards Regression with Compositional Covariates

协变量含成分变量时 Cox 比例风险回归模型的统

计推断

Xiaobo Wang

Yunnan University

摘要：成分变量在实际应用中十分常见，因其存在

多重共线性、非正态性以及定和约束，从而为统计

推断带来许多挑战。本文主要研究了 Cox 比例风险

回归模型下部分协变量为成分变量时的统计推断

问题，提出了线性约束矩阵化的方法。先将回归系

数的线性约束条件等价地转化为对数成分变量的

约束转换矩阵，再基于转换后的协变量提出回归系

数的约束估计。建立了约束估计的渐近正态性及渐

近方差的估计方法。模拟实验结果表明所提方法可

以有效地克服成分变量所带来的挑战。进一步将该

方法应用于美国健康和营养调查中成人的身体活

动模式对其死亡风险的影响研究。

Joint work with Yuanshan Wu.

July 14, 10:30-12:10

Invited Session IS038: New Statistical Methods for

Complex Imaging and Genetics Data

Cortical Surface Alignment Based on Continuous

Structural Connectivity

Zhengwu Zhang

University of North Carolina, Chapel Hill

Abstract: Most current registration procedures for the

brain cortical surface rely solely on anatomical features, such as sulcal depth and cortical folding patterns. Very few consider using the cortical surface's

connectivity information. Incorporating connectivity

information can offer more functionally relevant

alignment, overcoming the limitations and variability

inherent in methods relying solely on anatomical

landmarks. However, most existing network alignment methods are formulated as computationally demanding node-matching problems, making it impractical to align large networks. This paper proposes a

novel network alignment solution based on a new

network representation called Continuous Connectivity (ConCon). Under the ConCon representation, network registration becomes a task of finding the optimal diffeomorphisms to match cortical surfaces based

on their ConCon profiles. We develop an efficient

optimization algorithm for this purpose, expanding on

techniques from the field of functional data alignment.

Applying it to the Human Connectome Project data,

we demonstrate that the proposed network alignment

method can align cortical surfaces based on their

connectivity profiles and significantly improve downstream network analysis tasks.

TCRpred: Incorporating T-Cell Receptor Repertoire for Clinical Outcome Prediction

Qianchuan He

Fred Hutchinson Cancer Center

Abstract: T-cell receptor (TCR) plays critical roles in

recognizing antigen peptides and mediating adaptive

immune response against disease. The TCR sequences

provide important information about patients' adaptive

immune system, and have the potential to improve

clinical outcome prediction. However, it is challenging to incorporate the TCR repertoire data for prediction, because the data is unstructured, highly complex,

and TCR sequences vary widely in their compositions

and abundances across different individuals. We introduce TCRpred, an analytic tool for incorporating

TCR repertoire for clinical outcome prediction. The

TCRpred is able to utilize features that can be ex-

141

tracted from the TCR amino acid sequences, as well

as features that are hidden in the TCR amino acid

sequences and are hard to extract. Simulation studies

and real data analysis show that the proposed approach has good performance and adds extra power in

predicting clinical outcomes.

Joint work with Meiling Liu, Yang Liu and Li Hsu.

Brain Aging Chart for Understanding Complex

Aging Process via Multimodal Data Integration

Haochang Shou

University of Pennsylvania

Abstract: With the increasing need for big data analytics in medical imaging, integrating data from multiple studies and various biological domains has become critical to better understanding and differentiating complex human diseases such as aging and Alzheimer's disease (AD). In this talk, we will discuss

several most recent developments in multisite imaging

harmonization, multimodal data integration for aging

research and our efforts to build a brain aging chart

through the NeuroImaging Computational Harmonization and Artificial intelligence Toolbox (niCHART).

Our methods are motivated by the iSTAGING (Imaging-based coordinate SysTem for AGing and NeurodeGenerative diseases) consortium. The brain aging

chart utilizes machine-learning based structural and

functional brain age, AD-like neurodegeneration,

white matter hyperintensities (WMHs) as well as cognition and molecular biomarkers to quantify and display the complex and heterogeneous process of normal aging and AD pathology. We proposed a distance-based regression model to enable simultaneous

regression of multiple data modalities of varying

properties and dimensionalities and detect associations with various clinical covariates. Subsequently,

we proposed a framework for removing confounding

effects from arbitrary distance-based dimension reduction methods via partial embedding (PARE). We

demonstrate that the PARE framework could highlight

biological patterns of interest while effectively removing confounders in complex data objects.

Joint work with Andrew Chen, Zheng Ren, Christos

Davatzikos and Russell Shinohara.

LP-Micro: Explainable Longitudinal Prediction

Using Machine Learning Methods on Disease

Outcomes From Microbiome Data

Di Wu

University of North Carolina, Chapel Hill

Abstract: Longitudinal microbiome captures disease

progression. We developed a robust and interpretable

framework, LP-Micro, to identify disease-related

microbiome taxa and detect temporal dependencies

between longitudinal microbiome sequencing data and

subsequent disease outcomes. We explicitly define

two types of longitudinal prediction problems which

has not been previously well explored previously. The

LP-Micro framework leverages group lasso for feature

selection, machine learning methods for prediction,

and permutation feature importance testing (PermFit)

for explainable features. Simulations show that

LP-Micro identifies disease-related microbiome taxa,

resulting in improved prediction accuracy, comparing

to other prediction methods. Application in (i) early

childhood caries (ECC) data from the VicGen cohort

study and (ii) weight loss data following bariatric

surgery (BS) yields >70% prediction accuracy. The

critical predictive time points we identified are consistent with clinical knowledge. Importantly,

LP-Micro offers new insights of the prediction power

from the longitudinal microbiome data to a later clinical outcome.

Joint work with Yifan Dai.

Invited Session IS035: Network Analysis and

Cluster Analysis

Multi-Modal Data Integration via Orchestrated

Approximate Message Passing

Zongming Ma

Yale University

Abstract: The need to integrate multiple sets of features for understanding latent states of subjects arises

often in data analysis. For example, in single-cell

studies, one can leverage simultaneous sequencing of

mRNAs and proteins for delineation of cell subpopulations and single-cell atlas construction at fine granularity. In this talk, we present an orchestrated ap-