192

University of New Mexico

Abstract：There are many practical issues such as

logistics support, duration of study, and potential high

drop-out rate in clinical trial study. In particular, when

there are 4 or more treatments under consideration.

To overcome these issues, researchers use incomplete

block crossover design when patients receive only a

subset of treatments under comparison.

In this article, we propose a Bayesian approach to test

treatment effects for binary data in a 4X4 Latin-square.

We use several approaches such as dta augmentation,

scaled mixture of normals, and parameter expansion

to improve efficiency. The approach is illustrated with

a simulation study and a real data example.

Asymptotic Inference Theory of the Hawkes Process with Time-Varying Baseline Intensity and a

General Excitation Kernel

Jeffrey Kwan

University of New South Wales

Abstract：The Hawkes process is a popular point

process model for event sequences that exhibit a temporal clustering behavior. A natural way of measuring

the rate of occurrence of points of a point process is

via its intensity function. The intensity of a Hawkes

process consists of two components, the baseline intensity and the excitation kernel. The classical

Hawkes process assumes a constant baseline intensity

and an exponential excitation kernel. This results in an

intensity process that is Markovian. However, the

assumptions imposed by the classical Hawkes process

can be restrictive and unrealistic for certain modelling

purposes. By allowing the baseline intensity to vary

with time, and the excitation kernel to be

non-exponential, such a setup expands the modelling

capacity of the classical Hawkes process. However,

asymptotic inference under this setup is substantially

more difficult since the resulting intensity process is

non-Markovian, thus rending standard techniques for

asymptotic inference of Markov processes futile. To

overcome this challenge, we devised an approximation procedure to show the intensity process is asymptotically ergodic. This allows for the identification

of an ergodic limit to the likelihood function. Consequently, by taking a parametric approach and under

minimal regularity conditions, asymptotic results for

likelihood based statistical inference, for example,

consistency and asymptotic normality of the maximum likelihood estimator, can be achieved.

Joint work with Feng Chen, William Dunsmuir.

Contributed Session CS043: Recent Advances in

Differentially Private and Complex Data Model

Causal Inference in Randomized Experiments for

Dyadic Data

Yilin Li

Peking University

Abstract： Estimating the total average treatment

effect on a network could be considerably biased due

to spillover effects in the presence of unknown network interference. We consider novel dyadic outcomes in the presence of interference. Such outcomes

are common in many social network sources, such as

forwarding a message or sharing a link. We first introduce the setting of network interference with dyadic outcomes, which is of particular interest in online

experimentation. Then we manifest that the unbiased

estimator for the total average treatment effect based

on the conventional outcomes does not exist under

heterogeneous treatment effects. We provide subsequently unbiased estimators based on dyadic outcomes for randomized experiments. We show the

possible variance bounds of our proposed estimators

and provide a variance estimator to quantify the uncertainty. We illustrate the above phenomenon with a

variety of numerical experiments. We utilize our

method and discuss further applications scenarios.

Joint work with Lu Deng, Yong Wang, Wang Miao.

Optimal Locally Private Nonparametric Classification with Public Data

Yuheng Ma

Renmin University of China

Abstract：In this work, we investigate the problem of

public data-assisted non-interactive LDP (Local Differential Privacy) learning with a focus on

non-parametric classification. Under the posterior drift

assumption, we for the first time derive the mini-max

193

optimal convergence rate with LDP constraint. Then,

we present a novel approach, the locally differentially

private classification tree, which attains the mini-max

optimal convergence rate. Furthermore, we design a

data-driven pruning procedure that avoids parameter

tuning and provides a fast converging estimator.

Comprehensive experiments conducted on synthetic

and real datasets show the superior performance of

our proposed methods. Both our theoretical and experimental findings demonstrate the effectiveness of

public data compared to private data, which leads to

practical suggestions for prioritizing non-private data

collection.

Joint work with Hanfang Yang.

A Data-Driven Weighted Combination Approach

for Wind Speed Distribution Estimation Based on

Curve Clustering

Dan Zhuang

Fujian Normal University

Abstract：Developing wind energy can not only

promote economic development but also reduce CO2

emissions and control environmental pollution. Estimating the probability distribution of wind speed is

critically important to evaluate wind energy production for turbine design and site planning. In contrast to

existing literature, this paper introduces a data-driven

weighted combination approach to assess the probability distribution of annual wind speed. First, a spectral clustering method based on dynamic time warping

and daily kernel density function is used to categorize

daily wind speed curves to account for various probabilistic behaviors of different daily wind speeds. Second, the density function of each category is fitted

with a three-parameter Weibull distribution or mixed

Weibull distribution, determined by the morphological

characteristics of its histogram. Third, the annual wind

speed probability distribution is obtained by weighing

all classification distribution functions, with each

category weighted in proportion to its number of days.

Fourth, Monte Carlo simulation is conducted, demonstrating that our proposed method outperforms other

commonly used methods. Finally, the proposed

method is applied to illustrate a wind speed dataset

from a height of 50m above the ground in Southwest

China.

Joint work with Shihan Zhou, Jianbao Chen, Xiaoping Zhan, Shuangzhe Liu.

Two-Sample Distribution Tests in High Dimensions

via Max-Sliced Wasserstein Distance and Bootstrapping

Xiaoyu Hu

National University of Singapore

Abstract: Two-sample hypothesis testing, challenging

in high dimensions, is a fundamental statistical problem for inference about two populations. In this paper,

we construct a novel test statistic to detect

high-dimensional distributional differences based on

the max-sliced Wasserstein distance to mitigate the

curse of dimensionality. By exploiting an intriguing

link between the distance and suprema of empirical

processes, we develop an effective bootstrapping procedure to approximate the null distribution of the test

statistic. One distinctive feature of the proposed test is

the ability to construct simultaneous confidence intervals for the max-sliced Wasserstein distances of

projected distributions of interest. This enables not

only detecting global distributional differences but

also identifying significantly different marginal distributions between the two populations without the

need for additional tests, setting our approach apart

from existing methods. We establish consistency

through Berry-Esseen type bounds for Gaussian and

bootstrap approximations of the proposed test, based

on which we show that the test is valid and powerful

as long as the considered max-sliced Wasserstein distance is adequately large. The superior performance of

our approach is illustrated via simulated and real data

examples.

Joint work with Zhenhua Lin.

Differentially Private Top-k Selection and Its Application

Yaxuan Wang

Sichuan University

Abstract：The utilization of the k most important

(top-k) query values over a large domain universe

194

occurs widely in the medical, financial, statistical

fields, etc. With the release of these data leading to the

disclosure of personal privacy, Differential Privacy

(DP), as a novel and reliable privacy-preserving

framework, has been applied to the top-k query procedure called differentially private top-k selection.

How to select and output the most important value

and its index with high efficiency and accuracy is of

great concern where Report Noisy Max/Min (RNM)

and exponential mechanism are most commonly employed privacy techniques. Furthermore, the pealing

and one-shot approaches generalizing the aboved

method are used to search the top-k. In this work, we

propose a new RNM algorithm with Gumbel mechanism which output the most important value and its

corresponding index at the same time compared to the

former more efficiently. And generalizing it to top-k

using the peeling algorithm yields tighter privacy

upper bounds. It is applied in the Benjamini-Hochberg

procedure (BHq) to control the false discovery rate in

the multiple hypothesis testing problem, denoted as

the new privateBHq. Numerical experiments show

that that the proposed privateBHq method outperforms the original privateBHq.

Joint work with Jie Zhou.

Contributed Session CS044：Complex Data Model

Distributed Linear Regression with Compositional

Covariates

Xuejun Ma

Soochow University

Abstract: With the availability of extraordinarily

huge data sets, solving the problems of distributed

statistical methodology and computing for such data

sets has become increasingly crucial in the big data

area. In this paper, we focus on the distributed sparse

penalized linear log-contrast model in massive compositional data. In particular, two distributed optimization techniques under centralized and decentralized

topologies are proposed for solving the two different

constrained convex optimization problems. Both two

proposed algorithms are based on the frameworks of

Alternating Direction Method of Multipliers (ADMM)

and Coordinate Descent Method of Multipliers(CDMM, Lin et al., 2014). It is worth emphasizing

that, in the decentralized topology, we introduce a

distributed coordinate-wise descent algorithm based

on Group ADMM(GADMM, Elgabli et al., 2020 ) for

obtaining a communication-efficient regularized estimation. Correspondingly, the convergence theories of

the proposed algorithms are rigorously established

under some regularity conditions. Numerical experiments on both synthetic and real data are conducted to

evaluate our proposed algorithms.

A Study on the Causal Relationship Between Abnormal Renal Function Indicators and the Occurrence of Carotid Artery Plaques

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

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。

195

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.

Integration of Longitudinal Physical Activity Data

from Multiple Sources

Jingru Zhang

Fudan University

Abstract: As various devices have been developed

for collecting physical activity data, a critical challenge arises in integrating datasets across different

conditions to gain a better understanding of physical

activity characterization. The key issue lies in effectively removing site effects while preserving common

features. However, the continuous minute-by-minute

recording of physical activity by wearable sensor

devices over multiple days introduces a longitudinal

time-dependent structure that complicates integration.

To address this challenge, we propose a novel method

for integrating longitudinal physical activity datasets.

This method models shared information using common eigenvalues and eigenfunctions while allowing

for site-specific scale and rotation. We applied our

proposed method to NHANES datasets collected with

different types of wearable sensors. The results

demonstrate the superiority of our approach in removing site effects while preserving biological signals

compared to existing methods. This study establishes

a framework for integrating longitudinal

time-dependent datasets and offers insights into the

analysis of physical activity data.

Joint work with Hongzhe Li, Haochang Shou.

Transfer Learning Accounting for Time-Varying

Heterogeneity in Mixture Cox Model

Wei Zhao

Shandong University

Abstract：Mixture Cox model offers a flexible and

powerful tool to characterize complex time-varying

heterogeneity in risk estimation problem. In this paper,

we consider estimation and prediction of a mixture

Cox model from the perspective of transfer learning,

where except for the target cohort, auxiliary samples

from different but related mixture Cox models are

available. To allow for data-driven information bor-

196

rowing, we impose Lasso penalties on the discrepancies in mixing proportions, class-specific regression

coefficients and baseline hazard functions between

two cohorts. Therefore time-varying heterogeneity can

be successfully addressed via accounting for both

within-cohort mixture and inter-cohort differences. As

shown in the numerical studies, the proposed method

improves the estimation and prediction accuracy in

terms of coefficient estimates and risk prediction

compared to its competitors.

How to Analyze Adverse Events Data from Different Trials to Support a New Drug Application

Yankun Gong

Pfizer

Abstract: This talk will answer what statistical

method should be used for different kind of adverse

event data pooled from different studies in a new drug

application. We will first give a summary from ICH

guideline on what conclusion we want to reach when

conduct safety analysis, the requirement of sample

size for safety analysis, the commonly used analysis

approaches for safety analysis. Then we will use hypothetical examples to introduce the problem and to

provide the appropriate methods under different assumptions. The descriptive methods introduced including crude incidence proporiton, crude incidence

rate, weighted incidence proportion and incidence rate,

time to event analysis methods and so on.

July 14, 16:00-17:40

Invited Session IS064: Research on the Statistics

and Development of Networked Economic and

Social Systems in the Context of Digital Intelligence Technology

Research on the Construction of Theoretical

Framework of Intelligent Academic Evaluation

智能学术评价的理论框架构建研究

Liping Yu

Zhejiang Gongshang University

Abstract：随着人工智能技术发展与学术评价的现

实需求，学术评价的人工智能化必将出现。本文在

分析人工智能参与学术评价的方式以及学术评价

对人工智能现实需求的基础上，建立学术评价智能

化的研究框架，将人工智能技术嵌入到多属性评价

各环节，并分析了各环节中人工智能的作用机制。

研究发现：学术评价智能化将逐渐成为一种新的评

价模式；人工主导将是学术评价智能化的本质方式

；需要加强学术评价智能化的理论与方法研究；评

价数据库建设是学术评价智能化的基础；人工智能

为评价方法创新提供了有效的方式；人工智能会给

评价方法带来一场冲击；评价方法库整理是学术评

价智能化的重要环节；需要专门开发部分学术评价

智能化的专用软件。

Government Attention Index for National Governance Modernization with Its Application —

Measurement Based on News Text

国家治理现代化政府注意力指数构建及其应用

Kuangnan Fang

Xiamen University

摘要：国家治理现代化被誉为中国的第五个\"现代

化\"，是中国式现代化的重要组成部分。为了实现

国家治理效能的提升，需要政府长期集中政策注意

力于治理建设，而政府注意力具有不可观测性的问

题，基于官方新闻媒体测度政府注意力是一种有效

的方法。针对基于规则的文本提取方法具有主观性

强和传统 LDA 模型无法提取特定文本主题信息等

问题，本文提出了一种用于提取文本特定主题信息

的定向 LDA 模型，以人民日报、光明日报和新闻

联播文字稿等官方媒体为语料库，首次构建了国家

治理现代化政府注意力指数，并基于该指数研究了

政府注意力的治理效应与信息效应。研究发现：定

向 LDA 模型可以提炼得到更纯净的定向主题的关

注度，而不需要给定主题参数、不过度依赖于人工

构建的词典以及不需人工审核；基于新闻文本构建

的国家治理现代化政府注意力指数可以较好地反

映政府对国家治理现代化关注度的变化情况；政府

对国家治理现代化的重视可以有效转化为治理效

能的提升以及政府透过官方媒体的宣传也能有效

带动公众对国家治理现代化的关注度。

Joint work with Mingxiao Dai, Tingguo Zheng,

Hongwei Lin.

Statistical Analysis Methods and Applications in

Financial Networks under the Background of Digital Intelligence

数智化背景下金融网络统计分析方法与应用研究

197

Zisheng Ouyang

Hunan Normal University

摘要：随着人工智能和大数据技术的普及，金融数

据的来源得到进一步拓展，其不再局限于传统的经

济指标等结构化数据，还包括图像、音频、文本等

非结构化数据，这为金融统计分析带来了新的挑战

和困难。本文从复杂网络与金融风险传染、高阶网

络与金融数据分析、图神经网络与金融风险预警等

视角出发，系统梳理了网络分析在金融统计领域中

的应用，并归纳总结了前沿的网络分析方法。此外

，本文通过三个具体例子，展示了多层金融网络如

何与现实数据结合，并阐述了其在金融风险建模中

的应用。最后，本文对后续研究进行展望，以期为

网络分析在金融统计中的未来应用提供启示。

Network Statistics of Economic and Social Systems

in the Age of Data Intelligence

数智化时代经济社会系统网络统计

Guobin Fang

Anhui University of Finance & Economics

摘要：数字化智能化背景下经济社会系统网络统计

及其应用对数智化社会发展具有基础性作用，梳理

数字化智能化技术数据处理方法的发展才能明确

经济统计的未来发展方向。统计学研究的主要范畴

是通过数据的搜集整理与分析探索经济社会的客

观规律性，传统上主要讨论时序数据、截面数据和

面板数据的处理方法，互联网时代网络化文本流媒

体等非结构化数据处理是基于数学和计算机的广

义统计问题，如何把统计方法引入这类新型数据分

析和处理当中，相较于传统经济统计各个流程比较

分散（指标体系、统计调查、数理统计方法应用等

），微观到宏观整体性的系统网络统计及其分析研

究将起着至关重要的作用，类似于 ChatGPT 建立健

全的数据和文本资源基础和大模型应用，把现实领

域平台构建成统计应用方向。针对复杂系统网络统

计目标，把经济统计总体最大化、总体单位颗粒化

和数智化生态系统网络统计逻辑作为理论基础，从

实际出发研究产品和服务统计分类体系支撑的数

据资源基础设施，为行业平台提供个性化服务和智

能化技术服务，促进统计学及相关学科之间交叉融

合发展。

Invited Session IS055: Recent Advances in Statistical Machine Learning: Theory and Applications

Combinatorial Inference on the Optimal Assortment in Multinomial Logit Models

Junwei Lu

Harvard University

Abstract: Assortment optimization has been extensively explored in recent decades due to its practical

significance. Despite the substantial literature on optimization algorithms and latent score estimation,

there remains a need to explore uncertainty quantification for the optimal assortment, which holds great

practical importance. Instead of estimating and recovering the complete optimal offer set, decision-makers

may be interested in testing whether a given property

holds for the optimal assortment. For instance, they

may wish to determine whether certain products of

interest should be included in the optimal set or ascertain how many categories of products the optimal set

should contain. This paper proposes a novel inferential framework for testing such properties within the

context of the widely adopted multinomial logit

(MNL) model. In this model, each customer is assumed to purchase an item within the offered products

with a probability proportional to the underlying preference score associated with the product. We reduce

inferring a general optimal assortment property to

quantifying the uncertainty associated with the sign

change point detection of the marginal revenue gaps.

We establish the asymptotic normality of the marginal

revenue gap estimator and construct a maximum statistic using the gap estimators to detect the sign

change point. By approximating the distribution of the

maximum statistic through multiplier bootstrap techniques, we propose a valid testing procedure. Numerical experiments are also conducted to assess the performance of our method.

A Statistical Perspective of LLMs: Geometry of

Embeddings in Transformers

Yiqiao Zhong

University of Wisconsin-Madison

Abstract: Transformers are neural networks that underpin the recent success of large language models.

They are often used as black-box models and building

blocks of complex AI systems. Yet, it is unclear what

198

information is processed through layers of a transformer, which raises the issue of interpretability. In

this talk, I will present an empirical study of transformers by examining various pretrained transformer

models via a factor analysis inspired decomposition. A

surprisingly consistent geometry pattern emerges in

hidden states (or intermediate-layer embeddings)

across layers, models, and datasets. Our study underscores two statistical notions-smoothness and incoherence-as two key reasons behind the success of

LLMs.

Joint work with Jiajun Song.

Tightness of SDP and Burer-Monteiro Factorization for Phase Synchronization in High Noise Regime

Anderson Zhang

University of Pennsylvania

Abstract：We study the phase synchronization problem in the presence of Gaussian noise. For this problem, the maximum likelihood estimation (MLE) is

computationally challenging, and hence is relaxed to a

semi-definite programming (SDP). When the noise is

small, existing literature shows that the solution of

SDP is exactly equal to the MLE. However, it remains

unclear what happens when the noise is large. In this

work, we investigate the distance between the SDP

solution and the MLE, that is, quantifying the tightness of the SDP relaxation compared to the MLE. We

establish that, in the high-noise regime, the distance

between them is exponentially small, with the signal-to-noise ratio appearing in the exponent. By increasing the signal-to-noise ratio, our result shows

they coincide with each other, recovering the existing

results in the low-noise regime. We further extend our

analysis to the Burer-Monteiro factorization of the

SDP and establish similar results.

Dyadic Reinforcement Learning

Shuangning Li

University of Chicago

Abstract：Mobile health aims to enhance health

outcomes by delivering interventions to individuals as

they go about their daily life. The involvement of care

partners and social support networks often proves

crucial in helping individuals managing burdensome

medical conditions. This presents opportunities in

mobile health to design interventions that target the

dyadic relationship---the relationship between a target

person and their care partner---with the aim of enhancing social support. In this paper, we develop dyadic RL, an online reinforcement learning algorithm

designed to personalize intervention delivery based on

contextual factors and past responses of a target person and their care partner. Here, multiple sets of interventions impact the dyad across multiple time intervals. The developed dyadic RL is Bayesian and

hierarchical. We formally introduce the problem setup,

develop dyadic RL and establish a regret bound. We

demonstrate dyadic RL's empirical performance

through simulation studies on both toy scenarios and

on a realistic test bed constructed from data collected

in a mobile health study.

Invited Session IS044: Progress in Best Subset Selection

Best-Subset Selection in Generalized Linear Models: A Fast and Consistent Algorithm via Splicing

Technique

Junxian Zhu

Sun Yat-sen University

Abstract: In high-dimensional generalized linear

models, it is crucial to identify a sparse model that

adequately accounts for response variation. Although

the best subset section has been widely regarded as

the Holy Grail of problems of this type, achieving

either computational efficiency or statistical guarantees is challenging. In this article, we intend to surmount this obstacle by utilizing a fast algorithm to

select the best subset with high certainty. We proposed

and illustrated an algorithm for best subset recovery in

regularity conditions. Under mild conditions, the

computational complexity of our algorithm scales

polynomially with sample size and dimension. In

addition to demonstrating the statistical properties of

our method, extensive numerical experiments reveal

that it outperforms existing methods for variable selection and coefficient estimation. The runtime analy-

199

sis shows that our implementation achieves approximately a fourfold speedup compared to popular variable selection toolkits like glmnet and ncvreg.

Joint work with Jin Zhu, Borui Tang, Xuanyu Chen,

Hongmei Lin, Xueqin Wang.

Minimax Optimal Methods for High-Dimensional

Doubly Sparse Linear Regression

高维双稀疏线性回归的极小极大最优方法

Yanhang Zhang

Renmin University of China

Abstract: In this paper, we focus our attention on the

high-dimensional double sparse linear regression, that

is, a combination of element-wise and group-wise

sparsity. To address this problem, we propose an

IHT-style (iterative hard thresholding) procedure that

dynamically updates the threshold at each step. We

establish the matching upper and lower bounds for

parameter estimation, showing the optimality of our

proposal in the minimax sense. More importantly, we

introduce a fully adaptive optimal procedure designed

to address unknown sparsity and noise levels. Our

adaptive procedure demonstrates optimal statistical

accuracy with fast convergence. Additionally, we

elucidate the significance of the element-wise sparsity

level ?0 as the tradeoff between IHT and group IHT,

underscoring the superior performance of our method

over both. Leveraging the beta-min condition, we

establish that our IHT-style procedure can attain the

oracle estimation rate and achieve almost full recovery of the true support set at both the element and

group levels. Finally, we demonstrate the superiority

of our method by comparing it with several

state-of-the-art algorithms on both synthetic and real-world datasets.

Joint work with Zhifan Li, Shixiang Liu, Jianxin Yi.

Subset Selection EM Algorithm: Statistical Analysis and Applications

Ning Wang

Beijing Normal University

Abstract: We study the finite mixture of linear regression models, which is frequently used in many

fields, such as biology, genetics, engineering, and

marketing. In modern applications, the number of

predictors can be much larger than the sample size.

The heterogeneity and the high dimensionality bring a

challenging variable selection problem. There have

been some literature studies on the high dimensional

mixture linear regression problem, among which the

lasso penalized approach and its variants are the most

popular and promising. However, the existing solutions have the following limitations: the method analyses the maximum of a penalized likelihood function

but does not provide an algorithm that can identify the

(local) maximum due to then on-convexity; the method analyses the output of a proposed algorithm but

usually provides a biased estimator and lacks variable

selection consistency results. In this article, we propose a subset-selection approach for the

high-dimensional mixture linear regression model to

some of those limitations. We established the

non-asymptotic convergence rate and the variable

selection consistency results for the direct output of

the proposed algorithms. Besides the theoretical advantages, numerical studies show that the proposed

algorithm has encouraging performance and fast

computation.

Simultaneous Dimension Reduction and Variable

Selection for Multinomial Logistic Regression

Canhong Wen

University of Science and Technology of China

Abstract: Multinomial logistic regression is a useful

model for predicting the probabilities of multiclass

outcomes. Because of the complexity and high dimensionality of some data, it is challenging to fit a

valid model with high accuracy and interpretability.

We propose a novel sparse reduced-rank multinomial

logistic regression model to jointly select variables

and reduce the dimension via a nonconvex row constraint. We develop a block-wise iterative algorithm

with a majorizing surrogate function to efficiently

solve the optimization problem. From an algorithmic

aspect, we show that the output estimator enjoys consistency in estimation and sparsity recovery even in a

high-dimensional setting. The finite sample performance of the proposed method is investigated via

200

simulation studies and two real image data sets. The

results show that our proposal has competitive performance in both estimation accuracy and computation time.

Joint work with Zhenduo Li, Ruipeng Dong, Yijin Ni,

Wenliang Pan.

Invited Session IS098: Spatial and Network

Econometrics

Estimation and Selection of High Order Spatial

Autoregressive Models with Adaptive Elastic Net

Tuo Liu

Xiamen University

Abstract：Spatial autoregressive (SAR) model with

SAR errors is frequently used to analyze spatially

correlated data. This paper proposes a penalized

maximum likelihood method with adaptive elastic net

penalty to estimate a high order SARAR model. It

allows for simultaneous model selection and parameter estimation. We allow the order of spatial correlations and the number of exogenous regressors to grow

with sample size. To resolve the computational problem, we generalize the method of least squares approximation (Wang and Leng, 2007) to our model. We

show that the approximate adaptive elastic net estimator enjoys the oracle property as long as the initial

estimator has proper rate of convergence. We extend

the generalized two-stage least squares procedure in

Kelejian and Prucha (1988) to increasing dimension

setting and show that the estimator possess the desired

rate of convergence. To select the tuning parameter,

we propose a modified Bayesian information criterion

and prove that the tuning parameter selected can consistently identify the true model. We also carry out

Monte Carlo experiments to examine the performance

of our estimation method and apply our approach to

study peer effect on academic performance using a

social network data set from Taiwan.

Applications of Functional Dependence to Spatial

Econometrics

Xingbai Xu

Xiamen University

Abstract：In this paper, we generalize the concept of

functional dependence from time series (Wu, 2005)

and stationary random fields (El Machkouri, Volný

and Wu, 2013) to nonstationary spatial processes.

Within conventional settings in spatial econometrics,

we define the concept of spatial functional dependence measure and establish a moment inequality, an

exponential inequality, a Nagaev-type inequality, a

law of large numbers, and a central limit theorem. We

show that the dependent variables generated by some

common spatial econometric models, including spatial

autoregressive models, threshold spatial autoregressive models and spatial panel data models, are functionally dependent under regular conditions. Furthermore, we investigate the properties of functional dependence measures under various transformations,

which are useful in applications. Moreover, we compare spatial functional dependence with the spatial

mixing and spatial near-epoch dependence proposed

in Jenish and Prucha (2009, 2012), and we illustrate

its advantages.

Testing the Number of Network Communities Using the Eigengap Ratio

Wei Lan

Southwestern University of Finance and Economics

Abstract：Extensive research has been conducted on

the community structure of network data using various block models, including the stochastic block

model (SBM), degree-corrected stochastic block

model (DCSBM), mixed membership block model

(MM), and degree-corrected mixed membership block

model (DCMM). Accurately determining the number

of network communities is crucial for assessing the

adequacy of these models. To the best of our

knowledge, most of the existing approaches in detecting the number of network communities cannot be

compatible with sparse networks and diverging number of communities simultaneously. Furthermore,

these approaches rely on estimating network distribution parameters and are tailored to specific block

models. To address these limitations, we introduce an

eigengap ratio test statistic. This statistic is straightforward to compute, applicable to a wild range of

block models, avoids parameter estimations, and

201

adapts to both dense and sparse networks with divergent community numbers. We demonstrate theoretically that the proposed test statistic converges to a

function of the type I Tracy-Widom distributions under the null hypothesis and derive its asymptotic power under alternatives. Simulation studies and two empirical examples with dense and sparse networks indicate that the proposed method performs satisfactorily.

Dynamic Spatial Panel Data Models with Interactive Fixed Effects: M-Estimation and Inference

under Fixed or Relatively Small T

Liyao Li

East China Normal University

Abstract：We propose an M-estimation method for

estimating dynamic spatial panel data models with

interactive fixed effects based on (relatively) short

panels. Unbiased estimating functions (EF) are obtained by adjusting the concentrated conditional quasi

scores, given the initial values and with the factor

loadings being concentrated out, to account for the

effects of conditioning and concentration. Solving the

estimating equations gives the M-estimators of the

common parameters and common factors. Under fixed

?, √?-consistency and joint asymptotic normality of

the two sets of M-estimators are established. Under

? = ?(?), the M-estimators of the common parameters are shown to be √??-consistent and asymptotically normal. For inference, difficulty lies in the estimation of the variance-covariance (VC) matrix of the

EF. We decompose the EF into a sum of ? nearly

uncorrelated terms. Outer products of these ? terms

together with a covariance adjustment lead to a consistent estimator of the VC matrix under both fixed ?

and ? = ?(?). Important extensions of the methods,

allowing for unknown heteroskedasticity,

time-varying spatial weight matrices, high-order dynamic and spatial effects, are critically discussed.

Monte Carlo results show that the proposed methods

perform well in finite sample.

Invited Session IS100: Doctoral Dissertation in

Statistical Machine Learning

Decoupled Convergence of Two-Time-Scale Stochastic Approximation

Yuze Han

Peking University

Abstract：Two-time-scale stochastic approximation is

a variant of the classic stochastic approximation (SA),

devised to find the roots of systems with two interconnected equations based on noisy observations. In

this approach, two iterates are updated at varying

speeds using different step sizes, with each update

influencing the other. Previous studies in linear

two-time-scale SA have found that the convergence

rates for these updates depend solely on their respective step sizes, leading to what is termed decoupled

convergence. However, the possibility of achieving

this decoupled convergence in the nonlinear case remains less understood.

In this talk, we explore the potential for decoupled

convergence of nonlinear two-time-scale SA. First, we

relax the standard Lipschitz condition into a nested,

star-shaped Lipschitz condition. Under this condition,

we not only reproduce existing convergence results

but also construct a counterexample illustrating the

infeasibility of decoupled convergence through numerical experiments. Additionally, by introducing a

nested local linearity assumption, we demonstrate that

finite-time decoupled convergence can be achieved

with appropriate step size selection. Building upon

this, we further characterize the asymptotic behavior

of two-time-scale SA and establish a decoupled functional central limit theorem.

Trustworthy Reinforcement Learning: Safety

Constraints and Uncertainty Quantification

Liangyu Zhang

Peking University

Abstract：When applying RL in high-stakes scenarios,

researchers and practitioners aim to develop RL systems with not only good average performance but

trustworthiness. In the first part of this talk, we focus

on RL problems with a continuum of safety constraints. Our work can be viewed as a generalization

of the conventional framework of safe RL, which only

considers a finite number of safety constraints. Spe-

202

cifically, we propose a new modeling framework for

RL problems with a continuum of safety constraints

called the semi-infinitely Markov decision process.

We also devise efficient algorithms for solving RL

problems with a continuum of safety constraints and

provide their sample complexity bounds and iteration

complexity bounds. In the second part of this talk, we

discuss uncertainty quantification in RL. Uncertainties

in RL include intrinsic uncertainties and statistical

uncertainties. In the framework of distributional RL,

we go beyond expected return and consider the full

return distributions. This approach enables us to address intrinsic uncertainty in RL. Our exploration of

statistical aspects in distributional RL allows us to

address both intrinsic and statistical uncertainties in

RL simultaneously. From the non-asymptotic perspective, we present sample complexity bounds for distributional RL. From the asymptotic perspective, we

prove that the estimation error of the return distribution converges weakly to a Gaussian random element.

Based on the asymptotic theory, we propose asymptotically valid statistical inference procedures for a

large family of statistical functionals of return distributions.

Distributed Statistical Inference under Heterogeneity

Jia Gu

Peking University

Abstract: In this talk, we consider distributed statistical optimization and inference in the presence of

heterogeneity among distributed data blocks. A

weighted distributed estimator is proposed to improve

the statistical efficiency of the standard \"split-andconquer\" estimator for the common parameter shared

by all the data blocks. The weighted distributed estimator is at least as efficient as the would-be full sample and the generalized method of moment estimators

with the latter two estimators requiring full data access. A bias reduction is formulated for the weighted

distributed estimator to accommodate much larger

numbers of data blocks (relax the constraint from

? = ?(?1/2

) to ? = ?(?2/3

) , where ? is the

number of blocks and ? is the total sample size) than

the existing methods without sacrificing the statistical

efficiency at the same time. The mean square error

bounds, the asymptotic distributions, and the corresponding statistical inference procedures of the

weighted distributed and the debiased estimators are

derived, which shows an advantageous performance

of the debiased weighted estimators when the number

of data blocks is large.

Joint work with Song Xi Chen.

Invited Session IS088: Special Topic on Data Asset

Accounting

数据资产核算专题

The Value of Data -- Exploratory Measurement

Based on Platform Companies

数据的价值--基于平台公司的探索性测算

Jingping Li

Renmin University of China

摘要：数字经济时代，数据创造价值。平台公司创

造数据驱动的新商业模式，深入挖掘数据价值，从

中获取收益。本文采用数据价值链思想，选取五家

不同类型的平台公司，基于组织资本和永续盘存法

测算平台公司的数据资产，其中折旧率的估计使用

了前瞻利润模型。研究的主要结论包括：第一，五

家平台公司的数据资产价值占纳入数据资产的总

资产价值的比例均高于 10%；第二，五家平台公司

数据资产均呈现高速增长，但不同公司的数据资产

折旧率存在较大差异；第三，与传统资产只会发生

折旧不同，数据资产存在\"再增值\"的可能性；第四，

技术进步和市场竞争是平台公司数据资产折旧率

的重要影响因素。

Joint work with Zijun Xie.

Estimation of Data Asset Value: International Experience and China's Exploration

数据资产价值估算：国际经验与中国探索

Xiuhua Xiao; Weiying Ping

Jiangxi University of Finance and Economics

摘要：随着数字技术的迅猛发展和持续升级，数据

已逐渐成为国家的战略资源和新型生产要素，在优

化产业内分工、增强产业间协同和推动数实孪生等

方面发挥着重要作用。准确评估数据资产的价值，

深刻理解数据资产在国民经济中的地位和作用，已

成为当前的迫切需求。本文旨在系统探讨数据资产

203

价值核算问题，全面梳理关于在国民账户体系中记

录数据的修订历程，深入总结国际范围内数据资产

价值核算的实践经验。在此基础上，结合我国当前

数据资产的现实状况，参考国民账户咨询专家组的

有关建议以及国际上其他国家的经验做法，提出一

套兼具合理性和可操作性的数据资产流量和存量

核算框架和实践路径，指出当前数据资产价值核算

面临的问题与挑战，并提出针对性的对策建议。本

文的研究可以为我国政府部门开展数据资产核算

实践提供一定的理论依据和方法支撑。

Research on the Reconstruction of Production

Function under Data Asset Accounting

数据资产核算下的生产函数重构问题研究

Kewei Ma

Shanxi University of Finance and Economics

摘要：数据被列为生产要素后，数据资产也必将被

纳入国民经济核算范围，相应的经济生产范围及产

出口径就需要进行调整，否则将会出现生产要素与

产出配比的严重失调，即数据资产越多，其生产率

反而越低的不合理现象。本研究尝试在原有生产函

数中引入数据要素，分析数据要素自身，及其通过

作用于传统要素进而影响经济产出的路径机制，以

实现新要素构成下的生产函数重构。构造过程中，

以我国经济核算指标有关 R&D 支出资本化的调整

经验为参考，将数据要素下的数据资产相关投入从

“中间消耗”修改为“数据要素资产形成”。同时

，考虑数据资产自身的特殊性，即不存在折旧和资

产价值可能伴随资产规模的积累而增值等问题。鉴

于数据资产在现代经济中的关键地位，力图构建一

套具备现实操作性的可行方法，为宏观视角下开展

数据资产核算应用研究提供一些借鉴。

How the Capitalization of Data Elements Affects

GDP: Theoretical Basis and Real Evidence

数据要素资本化核算何以影响 GDP：理论基础与现

实证据

Kaike Wang

Shandong University of Finance and Economics

摘要：随着信息通信技术的快速发展和全球范围内

计算机通信类设备价格下行趋势，数据参与生产活

动的形式日益多样，对社会生产的影响程度也愈发

突出，并逐渐成为了与土地、技术、管理等并行的

生产要素。那么，如何衡量数据要素在生产中的贡

献也就成了各界普遍关注的问题。国民经济核算领

域应对数据要素化，首要的是处理生产边界扩展的

影响问题，这其中的关键更是体现在对 GDP 的核

算影响问题。对此，本文从基本核算理论层面阐明

了数据要素纳入国民账户体系的路径，搭建出数据

要素资本化核算影响 GDP 统计的基本框架，在此

基础上，利用相关统计资料开展了数据要素资本化

核算影响经济增长率和要素贡献率的实证研究，以

期为验证本文搭建的数据要素资本化核算理论框

架合理性和科学性，提供现实证据。这一研究，一

方面是对当前数据要素价值测度理论研究热点的

积极回应，另一方面也期望通过较为系统的核算理

论设计和统计测算，进一步规范增长动能分析和要

素投入测度的实证研究范式。

Joint work with Qiang He.

Invited Session IS086: Business Big Data Analysis

and Application

商务大数据分析与应用

Streamer Dynamics in Live Streaming Commerce

Shaohui Wu

Harbin Institute of Technology

Abstract：Live streaming commerce (LSC) has received a massive success worldwide, drawing increasing attention from academics and practitioners

alike. While the prior work documents several factors

that influence viewer engagement, little is known

about the dynamics among competing streamers on a

LSC platform. Conventional wisdom suggests that

competitor entry leads to demand substitution due to

cannibalization, whereas more recent work shows that

demand spillovers can arise possibly because of market expansion. In this paper, we aim to add to the

\"substitution or spillovers\" debate by examining the

impacts of competitor entry on other streamers’ performance (as measured by viewing traffic and sales

conversion). Our identification strategy exploits an

exogenous event wherein a top streamer, commanding

a massive following and playing a central role in attracting viewers and driving sales on live streaming

platforms, unexpectedly returned to the platform after

a long departure. The results from a difference-in-differences model reveal that whether one

effect dominates the other is situational, depending on

204

the stage at which viewers are along the customer

journey (viewing vs. purchasing). On the one hand,

the top streamer’s return increased viewing traffic of

competing streamers, supporting the emergence of a

spillover effect at the viewing stage. The same event,

on the other hand, led to decreases in competing

streamers’ sales conversion, endorsing the dominance

of a substitution effect at the purchasing stage. Our

findings generate sought-after insights into how LSC

platform managers can effectively allocate resources

in recruiting, managing, and promoting live streamers.

Data-Driven Intelligence: Promoting the Evaluation and Optimization of the Business Environment

数据智能驱动：促进营商环境的评估与优化

Liangke Cao

吉林省格远市场调研咨询有限公司

摘要：在当前全球经济一体化和技术飞速发展的双

重环境下，优化营商环境已成为提升地区竞争力的

核心举措。数据智能驱动的营商环境评估与优化策

略，借助了大数据分析、机器学习等先进技术，通

过精细化和动态化的评估方式，有效提升了政府服

务的精准性和效率，降低了企业运营的不确定性，

为经济的高质量发展注入了新动力。本研究报告深

入研究了数据智能在营商环境评估与优化中的理

论基础与实践应用，通过对营商环境优化案例的系

统梳理，分析了数智技术在政府服务效率的提升、

企业开办流程的简化、税务管理的智能化等方面的

应用，剖析数据智能技术在优化营商环境中的巨大

潜力。本报告坚持定性与定量研究的有机结合，运

用文献综述、案例分析、实证研究等多维度的研究

工具，确保了研究成果的科学性与实用性。研究结

论不仅证实了数据智能技术在增强政府服务效能、

减少企业运营成本等方面的显著优势，也揭示了在

数据治理、隐私保护等方面面临的挑战。展望未来

，数据智能驱动的营商环境评估与优化将是不可逆

转的趋势，政府、企业和社会各界应共同努力，积

极推动数据智能技术的应用，共同促进经济社会的

可持续发展。

Measurement of Node Importance in Social Networks

社会网络中节点的重要性度量

Li Zhai

Jilin University of Finance and Economics

摘要：社会网络遍布在我们的社会和经济生活中。

网络中的节点往往具有高度的异质性，不同节点在

网络结构、地位及其影响上的差异巨大。因而，挖

掘社会网络中的重要节点是社会网络分析的重要

研究内容之一。本研究结合合作网络、引文网络、

社会媒体用户信息交互网络等不同的网络背景，建

立了多种基于网络结构的节点重要性度量方法，并

对所提出的方法进行了理论探讨和实证分析。本研

究为网络中节点重要性评价、重要节点的识别以及

网络划分提供了新方法。

Joint work with Xiangbin Yan.

Contributed Session CS013: Statistical Inference

for Functional/Time Series Data

Unified Principal Components Analysis of Irregularly Observed Functional Time Series

Zerui Guo

Sun Yat-sen University

Abstract: Irregularly observed functional time series

(FTS) are increasingly available in many real-world

applications. To analyze FTS, it's crucial to account

for both serial dependencies and the irregularly observed nature of functional data. However, existing

methods for FTS often rely on specific model assumptions in capturing serial dependencies, or cannot

handle the irregular observational scheme of functional data. To solve these issues, one can perform dimension reduction on FTS via functional principal component analysis (FPCA) or dynamic FPCA. Nonetheless, these two methods may either be not theoretically optimal or too redundant to represent serially dependent functional data. In this article, we introduce a

novel dimension reduction method for FTS based on

the framework of dynamic FPCA. Through a new

concept called optimal functional filters, we unify the

theories of FPCA and dynamic FPCA, providing a

parsimony and optimal representation for FTS adapting to its serial dependence structure. This framework

is referred to as principal analysis via dependency-adaptivity (PADA). Under a hierarchical Bayesian

model, we establish an estimation procedure for dimension reduction via PADA. Our method can be

205

used for both sparsely and densely observed FTS, and

is capable of predicting future functional data. We

investigate the theoretical properties of PADA and

demonstrate its effectiveness through extensive simulation studies. Finally, we illustrate our method via

dimension reduction and prediction of daily PM2.5

data.

Joint work with Jianbin Tan, Hui Huang.

Factor Modelling for Matrix-Variate Functional

Time Series in High Dimensions

Zihan Wang

Tsinghua University

Abstract: Nowadays, the analysis of interconnected

systems is crucial across various fields, including

transportation and social networks. To address this

challenge, this paper introduces factor modelling for a

new data-type known as matrix-variate functional

time series, which competes with existing factor modelling for tensor-time series by treating intraday observations as random functions instead of random

vectors. Theoretical results on the consistency of the

estimated quantities under mild conditions have been

provided, and its finite-sample performances have

been illustrated through extensive simulations under

both fully and partially observed scenarios. Real data

examples about dynamic transportation networks have

been exercised to demonstrate the advantages of our

proposed method in terms of flexibility, interpretability and forecasting performance compared to the tensor-based method.

Joint work with Dong Li, Xinghao Qiao.

Functional Data Analysis with Covariate- Dependent Mean and Covariance Structures

Chenlin Zhang

Southwestern University of Finance and Economics

Abstract: Functional data analysis has emerged as a

powerful tool in response to the ever-increasing resources and efforts devoted to collecting information

about response curves or anything that varies over a

continuum. However, limited progress has been made

with regard to linking the covariance structures of

response curves to external covariates, as most functional models assume a common covariance structure.

We propose a new functional regression model with

covariate-dependent mean and covariance structures.

Particularly, by allowing variances of random scores

to be covariate-dependent, we identify eigenfunctions

for each individual from the set of eigenfunctions that

govern the variation patterns across all individuals,

resulting in high interpretability and prediction power.

We further propose a new penalized quasi-likelihood

procedure that combines regularization and B-spline

smoothing for model selection and estimation and

establish the convergence rate and asymptotic normality of the proposed estimators. The utility of the developed method is demonstrated via simulations, as

well as an analysis of the Avon Longitudinal Study of

Parents and Children concerning parental effects on

the growth curves of their offspring, which yields

biologically interesting results.

Joint work with Huanzhen Lin, Li Liu, Jin Liu, Yi Li.

Differential Inference for scATAC-Seq Data

Jiasheng Li

The Chinese University of Hong Kong (Shenzhen)

Abstract: Chromatin-accessible states play a critical

role in regulating gene expression. The development

of single-cell Assays for Transposase-Accessible

Chromatin using sequencing (scATAC-seq) technology allows us to measure the genome-wide chromatin

states at single-cell resolution. With the wide adoption

of scATAC-seq, researchers have begun to perform

scATAC-seq experiments for cells collected from

different biological conditions, like prefrontal cortex

cells from patients with Alzheimer's disease and

healthy controls. However, rigorous statistical methods for identifying chromatin regions with different

accessible states between conditions in a specific cell

type are still lacking. The main challenge comes from

unknown cell type labels of each cell and severe batch

effects between different scATAC-seq experiments.

Moreover, scATAC-seq data is ultrahigh-dimensional

with hundreds of thousands of features and extremely

sparse with over 95% zero counts. To tackle these

challenges, we develop an interpretable Bayesian

hierarchical model to simultaneously cluster cell types,

206

correct for batch effects, and identify

cell-type-specific differential region accessibility.

Both simulation studies and real data analysis show

that our proposed method can accurately cluster cell

types and identify differentially accessible regions

between conditions for each cell type. As a result, our

method can provide more valuable insights into the

gene regulatory mechanism.

Joint work with Fangda Song.

Revisiting the Poisson Autoregressive Model:

Structure and Statistical Inference

重访泊松自回归模型：结构与统计推断

Chenxiao Dai

Tsinghua University

Abstract: The first-order stationary Poisson autoregression (PAR) is one of the most classical count

time series models and has been widely studied.

However, few researchers pay attention to nonstationary PAR. The paper revisits PAR and provides some

novel results on asymptotical behaviors of the intensity process under nonstationarity. Further, the maximum likelihood estimation is considered in a unified

framework of stationary and nonstationary cases, and

its asymptotics is established. Monte Carlo simulation

studies are conducted to assess the finite-sample performance of the MLE.

Joint work with Yuxin Tao, Dong Li.

Contributed Session CS041 ： Biostatistics and

Industrial Statistics

Deep Neural Network for Partially Linear Subdistribution Hazard Model

Nengjie Zhu

Shanghai Jiao Tong University

Abstract：Subdistribution hazard models are widely

used for competing risk data analysis. Partially linear

regression for subdistribution hazard models in competing risk analysis is still yet to be studied. We propose the deep partially linear subdistribution hazard

model (DPLSHM), which utilizes deep neural networks (DNN) to estimate the nonlinear component

and circumvent curse of dimensionality issue. To

evaluate the predictive performance of the model, we

further develop a time-dependent AUC method specifically tailored for competing risk data and establish

its relationship with the C-index. Theoretical results

demonstrate the asymptotic normality of the parameter component at a rate of √? and provide the convergence rate of the nonparametric component, which

achieves the minimal limit convergence rate (multiplicative logarithmic factors). Subsequently, the paper

validates the excellent performance of DPLSHM in

estimation and prediction through numerical simulations and real-world datasets.

Joint work with Zhangsheng Yu.

Fault Diagnosis of Rolling Bearings Based on

WOA-ART and Improved DenseNet

基于 WOA-ART 和改进的 DenseNet 的滚动轴承故

障诊断

Cheng Shi

Chengdu University of Technology

摘要：滚动轴承作为工业生产中的核心部件，其健

康与否直接关乎工业的生产效率。目前基于时频分

析的卷积神经网络轴承诊断方法具有高精度的故

障表征以及出色的故障诊断能力。然而，目前许多

时频分析方法核心参数仍需要人为设定，出色的故

障诊断能力背后伴随着网络参数冗余、计算资源占

用严重。为了解决上述不足，本文提出了一种基于

鲸鱼算法（WOA）的自适应重新分配时频分析方法

（ART）和改进的 DenseNet 的滚动轴承故障诊断。

首先，通过 WOA 对 ART 的参数进行寻优，以 Rényi

熵作为适应度函数，实现轴承故障的自适应高精度

时频表征。然后，利用深度可分离卷积替换

DenseNet 中部分卷积模块，以降低模型参数量。最

后，利用轴承故障的时频表征作为特征输入到改进

的 DenseNet 模型，达到故障诊断的目的。本文通

过模拟信号和实际资料验证了所提出方法的有效

性和可行性，实验结果表明，WOA-ART 方法在提

高了时频表征精度的同时解决了许多时频分析方

法存在参数需要人工选取、不具备全局适应性的问

题。此外相比传统 DenseNet，改进的 DenseNet 故

障识别率达到 98.97%，参数量减少了 26%，具有

更高的故障诊断准确率以及更少的计算压力。

Joint work with Hui Chen, Yanjie Fan, Yuanwei

Song, Xuping Chen.

207

Economic Policy Uncertainty, Systemic Financial

Risks, and High-Quality Economic Development

经济政策不确定性、系统性金融风险与经济高质量

发展

Zhixin He

Xinjiang University of Finance & Economics

摘要：经济政策不确定性是系统性金融风险的重要

来源，进而会影响经济高质量发展，在中国经济转

型的新发展阶段，这种影响特征和强度更值得深入

探讨。通过 TVP-VAR 模型，探究 1997 年第 1 季

度至 2022 年第 4 季度期间中国的经济政策不确

定性、系统性金融风险与经济高质量发展之间的动

态关系。重点探讨了经济政策不确定性如何通过直

接与间接两条路径对经济高质量发展产生影响。研

究发现：在直接传导渠道方面，经济政策不确定性

在短期和长期均制约经济高质量发展，且具有时变

特征。短期制约作用随时间推移增强，而长期制约

随时间收敛。在间接传导渠道方面，经济政策不确

定性在短期内降低了系统性金融风险，中长期增加

了系统性金融风险。系统性金融风险的增加则可能

会先助力再抑制经济高质量发展。

Joint work with Jian He, Yu Liu.

Research on Patent Value Evaluation Based on

Entity Recognition

基于实体识别的专利价值评估研究

Ruoyu Yao

Shanghai University of Engineering Science

摘要：随着全球创新活动的日益加速，专利作为技

术进步和创新成果的重要象征，其数量呈指数级增

长，这种增长使得专利评估面临高人力成本和评估

难度增大的挑战。现有的专利评估方法往往只使用

词向量的方式对专利的文本信息进行提取，难以捕

捉到专利文本中的关键信息，从而影响了专利价值

的准确评估。本研究以新能源汽车领域的专利为例

，首先从技术、法律、经济三个维度选取代表性指

标，构建专利价值的评估体系。在广泛阅读相关专

利和文献的基础上，确定了包括发明结构类、物质

量类、功效类在内的七种实体类型。接着，本文采

用 Bert-bi-Lstm-CRF 模型来识别专利摘要中的这七

种实体，并将其作为实体指标纳入评估体系中。随

后，运用主成分分析法对指标进行特征筛选，并使

用随机森林算法对专利价值进行评估。研究结果显

示，通过将实体指标纳入传统指标体系，不仅显著

提高了模型的预测准确率，而且能有效提取专利摘

要中的关键实体，有助于分析该领域的技术发展趋

势和识别高价值特征。

Joint work with Qianguo Wang.

Contributed Session CS053：Complex Statistical

Models and Its Applications

Parsimonious Generative Machine Learning for

Non-Gaussian Tail Modeling and Risk-Neutral

Distribution Extraction

Nan Yang

Renmin University of China

Abstract: In financial modeling problems,

non-Gaussian tails exist widely in many circumstances. Among them, the accurate estimation of

risk-neutral distribution (RND) from option prices is

of great importance for researchers and practitioners.

A precise RND can provide valuable information

regarding the market's expectations, and can further

help empirical asset pricing studies. This paper presents a parsimonious parametric approach to extract

RNDs of underlying asset returns by using a generative machine learning model. The model incorporates

the asymmetric heavy tails property of returns with a

clever design. To calibrate the model, we design a

Monte Carlo algorithm that has good capability with

the assistance of modern machine learning computing

tools. Numerically, the model fits Heston option prices well and captures the main shapes of implied volatility curves. Empirically, using S&P 500 index option

prices, we demonstrate that the model outperforms

some popular parametric density methods under mean

absolute error. Furthermore, the skewness and kurtosis

of RNDs extracted by our model are consistent with

intuitive expectations. More generally, the proposed

methodology is widely applicable in data fitting and

probabilistic forecasting.

Joint work with Qi Wu, Zhonghao Xian and Xing

Yan.

A Generalization Sample Learning Method of

Deep Learning for Semantic Segmentation of Remote Sensing Images

Jingying Li

208

Henan University

Abstract: Deep learning methods have been widely

studied in the semantic segmentation field of the remote sensing image. Training images play an important role in these methods; however, each training

image usually contains not only the generalization

information of each land category but also the specific

interclass context between different categories. The

specific interclass context prevents deep learning

methods from focusing on generalization information

learning during training and limits the performance on

different data distributions. This article proposes a

generalization sampling learning method of deep

convolutional neural network (GSL-CNN) to emphasize generalization information learning for the semantic segmentation of remote sensing images. The

proposed method develops a new CBR sampling

strategy that contains three modules: category grouping (C), basic unit extraction (B), and random combination (R). Module C collects each land category map

and strips away the specific interclass context from

the raw annotated image. Module B extracts basic

units with different granularities from each land category map, and each basic unit can keep the generalization information of this category. Module R aims to

enhance the robustness against different data distributions by randomly picking basic units of different

categories and randomly generating their interclass

context. The new GSL-CNN method integrates the

CBR sampling strategy with the convolutional neural

network (CNN) model for semantic segmentation.

Experiments on different remote sensing datasets and

15 state-of-the-art CNN models validated that the

proposed method has the potential of improving the

generalization ability of the CNN method from a

sampling perspective.

Joint work with Chen Zheng.

EWMA Control Chart for Simultaneously Detecting Dual Parameters of Beta Distribution

同时检测 Beta 分布双参数的 EWMA 控制图

Qiuhan Chen

Liaoning University

摘要: 传统控制图通常假设数据服从正态分布，而

描述产品缺陷率、广告转化率等实际生产生活相关

问题的数据常常服从 Beta 分布，目前关于 Beta 分

布的控制图研究成果极少。由于 Max-EWMA 控制

图比 EWMA 控制图操作更加简单，检测效果显著，

且可同时监控两个参数，因此本文首先给出 Beta

分布参数的传统矩估计、对数矩估计和广义极大似

然估计，并基于估计量设计了三个新的 EWMA 控

制图：M-MAX-EWMA 控制图、L-MAX-EWMA 控

制图和 R-MAX-EWMA 控制图。然后，进行数值模

拟对比三个新控制图在 Beta 分布参数不同时的

ARL 和 SDRL 值，结果显示三个新控制图对参数小

漂 移 均 比 较 敏 感 ， L-MAX-EWMA 图 和

R-MAX-EWMA 图 的 控 制 效 果 优 于

M-MAX-EWMA 图，而 L-MAX-EWMA 图 和

R-MAX-EWMA 图之间差距不大。最后，将三个新

控制图应用于岩土黏聚力数据控制验证了三个控

制图在真实事例中的可行性和有效性。

Tensor Quantile Regression Based on Elastic Net

Penalty with Its Applications

Yan Gao

Liaoning University

Abstract: Quantile regression is an indispensable tool

in statistical learning, and variable selection is a crucial step in model building. We have now entered an

era of vast and diverse data, giving rise to tensor data.

Tensors pose challenges such as sparsity, low rank,

and multicollinearity. Direct vectorization analysis can

lead to information loss, disruption of spatial structure,

and overfitting, severely impacting the estimation

performance of models. Traditional vector covariate

estimation methods and variable selection methods

face formidable challenges. This paper proposes a

quantile regression model with elastic net penalty

based on Tucker decomposition with tensors as covariates, and provides corresponding estimation

methods and properties for large samples. On one

hand, this model simultaneously employs l1 and l2

regularization, combining the advantages of Ridge

regression and LASSO penalty functions. It not only

achieves coefficient compression and variable selection but also effectively addresses multicollinearity

and strong correlation issues. On the other hand, using

quantile loss function instead of traditional least

209

squares loss function enables the model to achieve

robust estimation when dealing with outliers or data

presenting peaked or thick-tailed distributions. Utilizing an improved ADMM algorithm to solve the

aforementioned model, numerical simulations validate

the superiority of the model. Analysis and research on

attention deficit hyperactivity disorder (ADHD) data

using the proposed model and algorithm yield the

following findings: the model can serve as a prognostic tool for assessing the future development risk of

ADHD; using the model, it is possible to identify the

most active brain subregions associated with ADHD

Index quantiles, with the thalamus, caudate nucleus,

and anterior cingulate cortex being highly correlated

with the 25th, 50th, and 75th quantiles, respectively.

Research on the Impact of Financial Accessibility

on the Value of Ecosystem Services

金融可得性对生态系统服务价值的影响研究

Wanmeng Gui

Anhui University of Finance and Economics

Abstract: 随着城市化进程加快和社会经济发展，

生态系统服务功能出现退化现象，使得生态安全面

临诸多威胁和挑战。目前大多研究只关注某一特定

区域生态服务价值的评估、时空演变等问题。本研

究基于 2007-2020 年我国 284 个地级市的 9 类土地

利用数据，运用基于主要农作物的经济价值系数修

订当量因子，计算得到了 14 年的生态系统服务价

值（ESV）。实证分析了金融可得性（FA）与 ESV

之间的关系。研究结果表明：首先，FA 能够显著

提升 ESV。其次，从异质性结果可以看出，FA 对

ESV 的促进作用受城市经济发展水平、规模大小以

及资源丰富度的影响。最后，通过动态面板门槛模

型估计发现，FA 对 ESV 的影响受到市场化水平（

Market）、科技投入（RD）和产业结构高级化（ISA

）的调节，Market、RD 和 ISA 均能够提高 FA 对

ESV 的积极影响。在金融市场快速发展背景下，本

研究为政府如何通过金融活动维护生态系统服务

功能、建设生态文明提供了实证启示。

Contributed Session CS045： Bayesian and Casual

Inference

The Empirical Bayes Estimators of the Variance

Parameter of the Normal Distribution with a Conjugate Inverse Gamma Prior under Stein's Loss

Function

Yingying Zhang

Yunnan University

Abstract：For the hierarchical normal and inverse

gamma model, we calculate the Bayes posterior estimator of the variance parameter of the normal distribution under Stein's loss function which penalizes

gross overestimation and gross underestimation

equally and the corresponding Posterior Expected

Stein's Loss (PESL). We also obtain the Bayes posterior estimator of the variance parameter under the

squared error loss function and the corresponding

PESL. Moreover, we obtain the empirical Bayes estimators of the variance parameter of the normal distribution with a conjugate inverse gamma prior by two

methods. In numerical simulations, we have illustrated

five aspects: The two inequalities of the Bayes posterior estimators and the PESLs; the moment estimators

and the Maximum Likelihood Estimators (MLEs) are

consistent estimators of the hyperparameters; the

goodness-of-fit of the model to the simulated data; the

numerical comparisons of the Bayes posterior estimators and the PESLs of the oracle, moment, and MLE

methods; and the plots of the marginal densities for

various hyperparameters. The numerical results indicate that the MLEs are better than the moment estimators when estimating the hyperparameters. Finally,

we utilize the bodyfat data of 250 men of various ages

to illustrate our theoretical studies.

Local Causal Structure Learning and Causal Effect Estimating in the Presence of Latent Variables

Feng Xie

Beijing Technology and Business University

Abstract ： Discovering causal relationships from

observational data, particularly in the presence of

latent variables, poses a challenging problem. While

numerous effective methods have been proposed, it is

often unnecessary and wasteful to find the global

structures when our interest lies solely in the local

structure of one target variable. Current local structure

learning methods largely assume causal sufficiency,

meaning that all the common causes of the measured

210

variables are observed. This paper first investigates

how to locally identify potential parent and child

nodes of a target node within models that include

latent variables. Specifically, we provide a principled

method for determining whether a variable is a cause

or non-cause of a target, based solely on the local

structure rather than the entire graph. Next, we

demonstrate how to locally estimate the causal effects

of parent nodes on the target node. Theoretically, we

demonstrate the correctness of our approach under the

assumptions of faithfulness, and the accurate checking

of independencies.

Joint work with Zheng Li, Peng Wu, Zeng Yan,

Chunchen Liu, Zhi Geng.

Multiply Robust Causal Inference in the Presence

of An Error-Prone Treatment

Shaojie Wei

Beijing Technology and Business University

Abstract：Numerous estimation procedures employed

in causal inference often rely on accurately measured

data. However, the prevalence of measurement errors

in practical studies may yield biased effect estimates.

It is common to employ validation samples to rectify

such biases in the measurement error literature. This

paper focuses on the estimation of the average causal

effect with a misclassified binary treatment in a primary population of interest. By leveraging a validation sample with covariates, an error-prone version of

treatment and a true treatment recorded, we provide

identifiability results under certain conditions. Building on identifiability, we explore three classes of estimators, each demonstrating consistency and asymptotic normality within distinct model sets. Furthermore, we propose a multiply robust estimation approach for the treatment effect based on the semiparametric theory framework. The multiply robust estimator retains consistent under any one of the listed

model sets and achieves the semiparametric efficiency

bound, provided all models are correct. We demonstrate the satisfactory performance of the proposed

estimators through simulation studies and a real data

analysis.

Joint work with Qinpeng He, Wei Li, Zhi Geng.

Causal Effect of Functional Treatment

Ruoxu Tan

Tongji University

Abstract：Functional data often arise in the areas

where the causal treatment effect is of interest. However, research concerning the effect of a functional

variable on an outcome is typically restricted to exploring the association rather than the casual relationship. The generalized propensity score, often used to

calibrate the selection bias, is not directly applicable

to a functional treatment variable due to a lack of

definition of probability density function for functional data. Based on the functional linear model for the

average dose-response functional, we propose three

estimators, namely, the functional stabilized weight

estimator, the outcome regression estimator and the

doubly robust estimator, each of which has its own

merits. We study their theoretical properties, which

are corroborated through extensive numerical experiments. A real data application on electroencephalography data and disease severity demonstrates the practical value of our methods.

Joint work with Zheng Zhang, Wei Huang, Guosheng Yin.

Statistical Learning Estimation of Treatment Effect Under Covariate-Adaptive Randomization

Jun Wang

Yunnan University

Abstract: Estimation of the treatment effect parameter is one of the crucial problems with clinical trials

for two or multiple treatments. We propose the statistical learning estimator of average treatment effects

under covariate adaptive randomization. The proposed

statistical learning estimator is consistent and asymptotically normally distributed. Simulation studies

show that the proposed statistical learning estimators

have some advantages over the Bugni’s estimator and

Ye’s estimator when the outcome model is nonlinear

models. Finally, we apply the proposed methodology

to a data set that studies the treatment effect of insurance provision on tobacco production for 12 tobacco

producing counties in Jiangxi Province, China during

211

2003.

Contributed Session CS046： Recent Advances in

Large Models and Artificial Intelligence

大模型和人工智能的最近研究

Shupai Big Model Platform (A Large Model Platform Based on AI Computing Power Management,

Helping You Create Private Large Models In 30

Minutes)

数派大模型平台（一款基于 AI 算力管理的大模型

平台，帮您 30 分钟创建私有大模型）

Jieyi Liao

Yunnan Shupai Technology Co., Ltd

摘要：数派大模型平台，如何降低大模型融入到行

业应用过程中的技术门槛，加快大模型技术在各行

业领域的推广和普及，有效促进产业升级和创新发

展。

Computing Power Drive, Data Empowerment,

Build a New Pattern of Digital Economy Development

算力驱动,数据赋能,构建数字经济发展新格局

Qiqiang Fan

Yunnan Provincial Digital Economy Industry Investment Group Co., Ltd

摘要：数字经济健康发展，有利于推动构建新发展

格局，有利于推动建设现代化经济体系，有利于推

动构筑国家竞争新优势。近年我国数字经济持续高

速增长，总量长期居世界第二，充分体现数字经济

是新质生产力重要抓手，是全球竞争制高点。算力

是数字经济时代的新型生产力，云南应发挥区位、

能源、成本“三重优势”，主动融入国家“东数西

算”战略，建立立足云南、面向国内和两亚的算力

中心，为数字经济发展提供算力支撑；数据是数字

经济的关键要素，云南应基于信创技术，加快治理

和特色产业数字化转型进度，加强智慧城市建设，

抢抓低空经济风口，汇据数据，设立数据交易场所

，推进数据资产入表，落地人工智能应用场景，谱

写好数字经济时代的云南篇章。

Application Practice and Reflection of Artificial

Intelligence Technology in Life Sciences

人工智能技术在生命科学的应用实践及思考

Yuedong Gao

Kunming Institute of Zoology, Chinese Academy of

Sciences

摘要：近年来人工智能快速发展，正融入千行百业

。随着计算能力的显著增强、大数据的积累以及深

度学习等机器学习技术的重大突破，人工智能成为

加快发展新质生产力的重要引擎。通过数据驱动和

自我学习，利用模型进行大量数据训练，模拟人脑

的学习机制，实现了在语音识别、图像识别、自然

语言处理等诸多领域超越人类的表现。

Industry Empowerment Based on the Nine Day

Model

基于九天大模型的行业赋能

Chen Yan

Chinese Mobile Research Institute

摘要：九天基座大模型是中国移动完全自主研发，

打造的千亿参数多模态大模型，实现数据构建、预

训练、微调、推理加速等全链路核心技术自主掌控

。通过叠加专项训练和优化，服务企业\"量体裁衣\"

快速构建行业大模型和打造智能化应用。

Ascending AI Accelerates the Deepening of Artificial Intelligence Towards Reality

昇腾 AI 加速人工智能走深向实

Zhipeng Yang

Huawei

摘要：介绍在人工智能蓬勃发展以及国际竞争日趋

激烈的背景下，国产人工智能的产业形式，以及华

为在智能计算领域的发展策略及产业实践。

Contributed Session CS047： Statistical Inference

in Complex Data

Accommodating Space Heterogeneity in Geographically Weighted Regression with Group Penalty

Tengdi Zheng

Beijing University of Technology

Abstract：Space-varying heterogeneity of data has

attracted increasing attention in recent years for adaptively borrowing information across different locations. Our method is motivated by the survey data

collected from multiple locations, which contain a lot

of valuable information but also involve some noises.

However, there is no literature on how to combine

212

borrowing information across different locations and

variable selection while maintaining the similarity

across all locations based on almost unified policies.

Our goal is to address this statistical challenge by

accommodating space heterogeneity under the geographically weighted regression strategy. To allow

borrow information across different locations and

variable selection, we propose a novel \"geographically

weighted regression + penalized variable selection\"

approach. Another significant development is that

comprehensively statistical properties are established.

Numerical results demonstrate the satisfactory competitive advantages of the proposed method compared

with alternative methods. Finally, we apply the proposed method to \"One Thousand and One Hundred

Villages\" social survey in China, and investigate risk

factors of inpatient treatment cost, outpatient treatment cost, and self-treatment cost.

Joint work with Chenjin Ma.

A Conditional Distribution Function-based Measure for Independence and K-sample Test in Multivariate Data

Li Wang

Tsinghua University

Abstract：We introduce a new index to measure the

degree of dependence and test for independence between two random vectors. The index is obtained by

generalizing the Cramer-von Mises distances between

the conditional and marginal distribution functions via

the projection-averaging technique. If one of the random vectors is categorical with K categories, we propose a slicing estimator to estimate the proposed index.

We conduct an asymptotic analysis for the slicing

estimator, allowing K to increase with the sample size,

which differs from previous studies that have mostly

focused on the classical univariate setting with a fixed

number of categories. We derive the asymptotic distribution of the estimator under the null hypothesis

and show that it converges to a normal distribution.

When both random vectors are continuous, we introduce a kernel regression estimator for the proposed

index, and also demonstrate that the asymptotic null

distribution follows a normal distribution. The proposed tests are studied via simulation, with two real

data applications presented to illustrate our methods.

Joint work with Hongyi Zhou, Weidong Ma, Ying

Yan.

Copula-Based Nonparametric Estimation of

Hellinger Correlation and Distance

Wenjing Liu

University of Macau

Abstract：In this paper, we study the estimation of the

Hellinger correlation proposed by Geenens and

Lafaye de Micheaux (2022), where they show that the

Hellinger correlation can be written as a function of

the integral of square-rooted copula density. They

provide two estimators, one consistent estimator via

the power function of a density function and another

based on the probit-transformation, which deals with

boundary bias. Motivated by these, we propose novel

estimators of the Hellinger correlation via a

resampling approach. We first use the kernel method

(luckily, see that the beta kernel also works here) to

estimate the copula density and then construct the

estimator for the Hellinger correlation by employing

the external uniform random variables generated artificially. We consider both the classic kernel method

and the beta kernel approach in the density estimation.

The asymptotic properties of the suggested estimators

of copula density, Hellinger distance, and Hellinger

correlation are derived sequentially, and a

cross-validation method for selecting the critical

smoothing parameters is discussed. We also address

the problem of testing dependence with the copula

function of two given random variables. Finally, simulation studies and a real data analysis evidence their

excellent performance compared to their competitors.

Joint work with Zhi Liu.

Bayesian Optimization via Exact Penalty

Jiangyan Zhao

East China Normal University

Abstract：Constrained optimization problems pose

challenges when the objective function and constraints

are nonconvex and their evaluation requires expensive

black-box simulations. Recently, hybrid optimization

213

methods that integrate statistical surrogate modeling

with numerical optimization algorithms have shown

great promise, as they inherit the properties of global

convergence from statistical surrogate modeling and

fast local convergence from numerical optimization

algorithms. However, the computational efficiency is

not satisfied by practical needs under limited budgets

and in the presence of equality constraints. In this

article, we propose a novel hybrid optimization

method, called exact penalty Bayesian optimization

(EPBO), which employs Bayesian optimization within

the exact penalty framework. We model the composite

penalty function by a weighted sum of Gaussian processes, where the qualitative components of the constraint violations are smoothed by their predictive

means. The proposed method features (i) closed-form

acquisition functions, (ii) robustness to initial designs,

(iii) the capability to start from infeasible points, and

(iv) effective handling of equality constraints. We

demonstrate the superiority of EPBO to

state-of-the-art competitors using a suite of benchmark synthetic test problems and two real-world engineering design problems.

Joint work with Jin Xu.

Optimal Model Average Prediction Based on Semi

Parametric Mixed Effects Model

基于半参数混合效应模型的最优模型平均预测

Baoqun Chang

Kunming University of Science and Technology

摘要：在对纵向数据的预测过程中,通常会遭遇模型

的不确定问题。为了解决模型不确定性问题，文章

提出一种基于半参数混合效应模型的最优模型平

均预测方法。文章提出采用删组交叉验证的方法获

得最优权重向量的估计，进而获得最终的模型平均

预测值。理论上，文章证明了当所有候选模型均误

设定时，所提出的模型平均估计具有渐近最优性。

即，提出的模型平均估计量的二次损失渐近地达到

了用该集合中权重进行加权的所有模型平均估计

的二次损失下确界。另一方面，当候选模型中包含

正确模型时，证明了提出的权重估计方法，在大样

本意义下能够将权重分配给正确模型。模拟和实际

数据分析表明，所提出的模型平均估计方法与一些

常用的方法相比具有较好的预测性能。

Joint work with Liucang Wu, Na Li.

Contributed Session CS048： Factor Model and

Community Network

A Weighted Iterative Projection Estimation Procedure for Robust Tensor Factor Model with

Tucker Decomposition

Xuemei Hu

Chongqing Technology and Business University

Abstract：Tensor Factor Models (TFM) have been

proposed as appealing dimension reduction tools for

tensor-valued time series. In this paper we firstly replace the ordinary least square loss by the Huber-exponential squared loss (H-ESL), i.e., a hybrid of

squared loss for relatively small errors and exponential squared loss for relatively large ones), and creatively propose a fire-new robust TuckerTFM to handle

some possible outliers or heavy-tailed cases from

tensor observations. Then, we review the two existing

estimation methods: the initial mode-wise PCA estimation (IE) and projection estimation under least

square loss (PE-LS) for Tucker-TFM, further develop

a fire-new weighted iterative projection estimation

method under H-ESL(WIPE-HESL) for loading matrices, tensor factors and signal parts from the robust

Tucker-TFM, and exhibit a specific algorithm and its

pseudo-code for WIPE-HESL. The proposed robust

procedure can efficiently deal with outliers or

heavy-tailed cases. Monte Carlo simulation data are

analyzed using four estimation methods: IE, PE-LS,

the robust weighted projection estimator under the

Huber loss (WPE-HL) and the robust WIPE-HESL.

Simulation results verify that WIPE-HESL performs

best, WPE-HL performs second, PE-LS performs third,

and IE performs worse. Finally, we apply the proposed WIPE-HESL method to analyze Tartu population flow data, and obtain results that are consistent

with reality.

Nonasymptotic Performance Analysis of ESPRIT

and Spatial-Smoothing ESPRIT

Zai Yang

Xi'an Jiaotong University

Abstract：This paper is concerned with the problem

214

of frequency estimation from multiple-snapshot data.

It is well-known that ESPRIT (and spatial-smoothing

ESPRIT in presence of coherent sources or given

limited snapshots) can locate the true frequencies if

either the number of snapshots or the signal-to-noise

ratio (SNR) approaches infinity. In this paper, we

analyze the nonasymptotic performance of ESPRIT

and spatial-smoothing ESPRIT with finitely many

snapshots and finite SNR. We show that the absolute

frequency estimation error of ESPRIT (or spatial-smoothing ESPRIT) is bounded from above by

C

max{σ,σ

2

}

√L

with overwhelming probability, where σ

2

denotes the Gaussian noise variance, L is the number

of snapshots and C is a coefficient independent of L

and σ

2

, if and only if the true frequencies can be localized by ESPRIT (or spatial-smoothing ESPRIT)

without noise or with infinitely many snapshots. Our

results are obtained by deriving new matrix perturbation bounds and generalizing the classical Schur

product theorem, which may be of independent interest. Extensions to MUSIC and spatial-smoothing

MUSIC are also made. Numerical results are provided

corroborating our analysis.

Statistics, Data Science, and Artificial IntelligenceA Comparative Analysis Based on Quantitative

History

统计学、数据科学与人工智能——基于量化历史的

比较分析

Leping Liu

Tianjin University of Finance and Economics

摘要：统计是动态的历史，历史是静态的统计。本

文基于量化历史的比较分析，以时间为轴，按公元

前统计计数\"三大难题\"、近代统计学\"四大发现\"和

人工智能的\"三次浪潮\"和\"两次寒冬\"，纵向简述公

元前 450—2023 年人类历史长河中的与数据分析相

关的年代大事；然后，以统计科学发现为标志，从

汇总、信息测度、似然、相互比较、回归、实验设

计、残差和因果推断等方面，横向比较 Stigler 于

2016 年提出的构筑\"统计智慧\"殿堂的七大支柱。

Joint work with Rui Xu, Liyuan Liu.

Community Detection in Directed Networks Based

on Co-clustering and Node Similarity

Yang Jiao

Guizhou University of Finance and Economics

Abstract：Currently, there are relatively few methods

for community detection in directed networks. Moreover, a common practice in research is to either ignore

the directionality of edges or directly apply methods

from undirected networks to directed ones, thereby

disregarding the differences between node receiving

and sending states. To address these issues, this paper

proposes a directed network community detection

method based on the ideas of co-clustering and node

similarity, called DI-CCNS. This method aims to partition directed networks into communities of outgoing

and incoming nodes. Through empirical studies on

nine real network datasets, we have demonstrated the

superiority of the DI-CCNS algorithm over the

DI-SIM algorithm in terms of modularity results. Additionally, our method can identify communities of

nodes with similar sending states, further validating its

effectiveness. Therefore, our research provides new

research directions for the field of community detection in directed networks.

Joint work with Guihai Yu.

The Unique Solution of Factor Analysis and the

Significance Test of Solutions

因子分析的唯一解与解的显著性检验

Haiming Lin

Guangzhou Huashang College

摘要：因子分析中，因子模型的目标、唯一解、最

好的方法、Heywood 情况、公共因子个数的检验等

重要内容，受到了著名统计学家 Kandall、Rao、

Anderson、Johnson 和 Wichern 等的关注。本文由此

提出 5 个问题，用因子方差贡献降序法、标准化主

成分法、变异特殊因子解法，修正了因子模型中的

3 个设定偏差，开发了一种有唯一解的因子最小误

差模型，在因子分析目的约束下，该模型及其解是

因子分析的最好方法，且与近期应用最多的主成分

法和回归法的估计相同。用该模型及其解，开发了

一种新的检验准则，能进行因子解的显著性检验，

解决了本文提出的问题。

Contributed Session CS049： Bayesian and Ma-

215

chine Learning

Bayesian Optimization with Pareto-Principled

Training for Economical Hyperparameter Optimization

Yang Yang

Nankai University

Abstract: The specification of hyperparameters plays

a critical role in determining the practical performance

of a machine learning method. Hyperparameter Optimization (HPO), i.e., the searching for optimal specification of hyperparameters, however, often faces

critical computational challenges due to the vast

searching space and the high computational cost on

model training under a given hyperparameter specification. In this paper, we propose BOPT-HPO, a systematic approach for efficient HPO by leveraging

Bayesian optimization with Pareto-principled training,

based on the observation that the training procedure of

a machine learning method under a given hyperparameter specification often follows the Pareto principle (the 80/20 rule) that about 80% of the total improvement in the objective function is achieved in 20%

of the training time. By introducing two levels of

training corresponding to the Pareto principle, i.e., the

eighty-percent training (ET) and the complete training

(CT), and establishing a joint surrogate model for CT

runs and ET runs, BOPT-HPO reduces the computational cost of HPO significantly under the framework

of Bayesian optimization with multi-fidelity measurements. A wide range of experimental studies confirm that the proposed approach achieves economical

HPO for various machine learning models, including

support vector machines, fully connected networks,

and convolutional neural networks.

A Unified EM Framework for Estimation and Inference of Normal Ogive Item Response Models

Xiangbin Meng

Northeast Normal University

Abstract：Normal Ogive (NO) models have made

substantial contributions to the development of Item

Response Theory (IRT) and have become popular

measurement models in educational and psychological

measurement. However, the estimation and inference

associated with NO models often present statistical

and computational challenges. The purpose of this

study is to present an efficient and reliable computational method for fitting NO models. Specifically, we

introduce a novel and unified EM algorithm for the

marginal maximum likelihood or maximum a posteriori estimations of NO models, including two–

parameter, three–parameter, and four–parameter NO

models. Key improvements of our EM algorithm include the formulation of the complete data likelihood

within an exponential family, simplifying numerical

computations and ensuring convergence, and the introduction of a two-step expectation procedure in the

E-step, which effectively reduces the dimensionality

of integration. Moreover, we develop statistical inference procedures for estimating the standard errors

(SEs) of the estimated parameters. Simulation results

demonstrate the superior performance of our algorithm in terms of recovery accuracy and robustness.

To further validate our methods, we apply them to real

data from the Programme for International Student

Assessment (PISA). The results affirm the reliability

of the parameter estimates using our method.

A Spatial-Temporal Graph Neural Network Model

for Multi-site Temperature Forecasting

Zhouping Li

Lanzhou University

Abstract：Temperature is an important meteorological

factor that closely related to people's daily lives, and

influences the public health, cardiovascular diseases,

the growth and yield of crops, etc. It is known that

numerical weather prediction (NWP) models, which

are based on physical laws to predict the future state

of the weather, may be inaccurate because the atmospheric physical processes may be incomplete. In this

paper, we study the temperature prediction problem

with data from multiple meteorological stations, taking into account of the interactions between meteorological factors and different terrains, we propose a

spatial-temporal graph neural network model that

utilizes the Gated Recurrent Unit (GRU), Convolutional Neural Network (CNN) and Graph Attention

Network (GAT). The experimental results show that

216

our proposed method significantly improves the temperature prediction for the meteorological observation

sites in Lanzhou city and outperforms its competitors.

Joint work with Jiali Xu, Xinwei Liu and Wubin

Huang.

Enhancing Missing Data Imputation through

Combined Bipartite Graph and Complete Directed

Graph

Ziqi Chen

East China Normal University

Abstract：In this paper, we confront a significant

challenge in the field of missing data imputation:

identifying and leveraging the critical interdependencies among features to improve the precision of feature imputation. We propose a novel framework,

known as the bipartite and complete directed graph

neural network (BCGNN). Within BCGNN, observations and features are differentiated as two distinct

node types. The values of observed features are converted into attributed edges, linking nodes representing features with those representing observations. The

bipartite segment of our framework inductively creates embedding representations for nodes, efficiently

utilizing the comprehensive information encapsulated

in the attributed edges. In parallel, the complete directed graph segment adeptly outlines and communicates the complex interdependencies among features,

ensuring a deep understanding of their connections.

When compared to contemporary leading imputation

methodologies, BCGNN consistently outperforms

them, achieving a noteworthy average reduction of 15%

in mean absolute error for feature imputation tasks

under different missing mechanisms. Our extensive

experimental investigation confirms that an in-depth

grasp of the interdependence structure substantially

enhances the model's feature embedding ability. We

also highlight the model's superior performance in

label prediction tasks involving missing data, and its

formidable ability to generalize to novel, unseen data

points.

Joint work with Zhaoyang Zhang, Hongtu Zhu,

Yingjie Zhang, Hai Shu.

Contributed Session CS020：Optimal Subsampling

Distributed Subsampling for Multiplicative Regression

Xiaoyan Li

Chongqing University

Abstract: Multiplicative regression is a useful alternative tool in modeling positive response data. This

paper proposes two distributed estimators for multiplicative error model on distributed system with

non-randomly distributed massive data. We first present a Poisson subsampling procedure to obtain a

subsampling estimator based on the least product

relative error (LPRE) loss, which is effective on a

distributed system. Theoretically, we justify the subsampling estimator by establishing its convergence

rate, asymptotic normality and deriving the optimal

subsampling probabilities in terms of the L-optimality

criterion. Then, we provide a distributed LPRE estimator based on the Poisson subsampling (DLPRE-P),

which is communication-efficient since it needs to

transmit a very small subsample from local machines

to central site, which is empirically feasible, together

with the gradient of the loss. Practically, due to the use

of Newton-Raphson iteration, the Hessian matrix can

be computed more robustly using the subsampled data

than using one local dataset. We also show that the

DLPRE-P estimator is statistically efficient as the

global estimator, which is based on putting all the

datasets together. Furthermore, we propose a distributed regularized LPRE estimator (DRLPRE-P) to

consider the variable selection problem in high dimension. A distributed algorithm based on the alternating direction method of multipliers (ADMM) is

developed for implementing the DRLPRE-P. The

oracle property holds for DRLPRE-P. Finally, simulation experiments and two real-world data analyses are

conducted to illustrate the performance of our methods.

Joint work with Zhimin Zhang and Xiaochao Xia.

Jackknife Empirical Likelihood with Complex

Surveys

Mengdong Shang

Shaanxi Normal University

217

Abstract: We propose a novel method called the

jackknife empirical likelihood approach for analyzing

survey data obtained from general unequal probability

sampling designs. This method applies to parameters

defined by U-statistics. Theoretical proofs establish

that the jackknife pseudo empirical likelihood ratio

statistic follows an asymptotic chi-square distribution.

And this statistic can be utilized to construct confidence intervals for complex survey samples. In this

paper, we explore scenarios involving the presence or

absence of auxiliary information and the utilization of

design weights or calibration weights. We conduct

numerical studies to evaluate the performance of the

jackknife pseudo-empirical likelihood ratio confidence intervals in terms of coverage probability and

tail error rates. Our findings demonstrate that the proposed methods outperform those based on the normal

approximation.

Joint work with Xia Chen.

Alpha Optimal Subsampling of Joint Mean and

Variance Models under Heteroscedasticity Big

Data

异方差大数据下联合均值与方差模型的 alpha-最

优子抽样

Zhengyu Xiong

Kunming University of Science and Technology

摘要: 随着信息技术的发展, 经济、金融、工业等

领域产生了异常庞大的数据, 这些数据往往具有异

方差特性, 传统统计模型和统计方法难以解决该类

大数据的建模问题. 子抽样是处理大数据的重要方

法. 本文针对联合均值与方差模型, 在异方差大数

据环境下研究了子抽样问题. 本文主要贡献如下：

对具有异方差特性的大数据建立联合均值与方差

模型, 在一定条件下，基于 A-最优准则和 L-最优准

则讨论了子样本参数估计的一致性和渐近正态性;

首次提出了异方差大数据下联合均值与方差模型

的 alpha-最优子抽样算法. 数值模拟和实证分析的

结果表明, 该抽样算法能提高估计的精确性、减少

计算成本.

Joint work with Liucang Wu and Lanjun Yang.

Optimal Linear Combination of Biomarkers by

Weighted Youden Index Maximization

Sizhe Wang

East China Normal University

Abstract: In medical research, it's a common practice

to combine various biomarkers to improve the accuracy of disease diagnosis. The weighted Youden index

(WYI), which assigns diverse weights to sensitivity

and specificity based on their relative importance,

serves as an important and flexible evaluation metric

of diagnostic tests. However, no existing methods

have been designed specifically to identify the optimal

linear combination of biomarkers that maximizes the

WYI. In this paper, we propose a novel method to

construct a diagnosis score and determine the optimal

cut-off point by directly maximizing the WYI. The

optimal combination coefficients and cut-off point are

shown to have cube root asymptotics and their joint

limiting distribution is established subsequently. Further, the asymptotic normality of the optimal

in-sample WYI is established and out-sample inference for score distribution and comparison is investigated. These results provide deep theoretical insights

for methods about Youden index maximization probably for the first time. Computationally, an iterative

marginal optimization algorithm that's different from

the literature is adopted to deal with the neither continuous nor smooth objective function. Simulation

studies support the theoretical results and show superiority of the proposed method over the commonly

used logistic regression. A real-world example of

coronary disease diagnosis is presented for illustration.

Multidimensional Distribution-to-Distribution

Regression via Optimal Transport Maps

Jianyang Tong

Yunnan University

Abstract: The distribution-to-distribution regression

models have drawn more and more attention in statistics in recent years. Nowadays the studies of distribution-to-distribution regression models mainly focus on

univariate distributions, and extending the existing

methods to the multivariate setting is difficult, because the corresponding asymptotic analysis will be

more complicated and computation will be full of

218

numerical difficulties. In practical applications the

multivariate distributions are more common, so we

aim to develop the multidimensional distribution-todistribution regression model where the covariate and

response are both multidimensional probability distributions. The regression model is based on the theory of optimal transportation, and links the conditional

Fréchet mean of the response to the covariate via an

optimal transport map. The Fréchet-least-square estimator of the regression map is defined and the identifiability, consistency and rate of convergence to the

true map are established by the theory of empirical

process. Guided by the optimal transport theory,

computation of the estimator can be done by learning

the optimal Kantorovich potential which induces the

optimal transport map. This involves learning two

convex functions, by solving a minimax optimization.

Using the input convex neural networks (ICNNs), we

estimate the optimal transport map as the gradient of a

convex function that is trained via minimax optimization. We compare our method with other

state-of-the-art methods: Barycentric-OT, W1-LP and

W2-GAN on simulated and real data to show the superiority of our method.

Joint work with Jian Sun and Niansheng Tang.

Contributed Session CS050： Mathematical Statistics and Industrial Statistics

Leveraging Cross-Population Fine-Mapping to

Strengthen cis-Mendelian Randomization

Mingxuan Cai

City University of Hong Kong

Abstract：By integrating GWASs and expression

quantitative trait loci (eQTL) mapping studies,

cis-Mendelian randomization (cis-MR) seeks to determine the causal effect of gene expression on human

complex traits. However, two key challenges have

hampered the accurate identification of causal genes.

First, eQTLs inherited at a low recombination rate

often harbor multiple causal variants in linkage disequilibrium (LD), making it difficult to identify independent instrumental variables (IVs), and limiting the

power of cis-MR. Second, eQTL variants can affect

the outcome trait through pathways other than the

target gene (i.e. horizontal pleiotropy), which violates

the MR assumption and leads to false positive results.

Here, we introduce a statistical method called XMR

that leverages cross-population fine-mapping to effectively identify causal SNPs of gene expression as IVs

for MR analysis, which helps maximize the MR power. At the same time, we explicitly correct for the horizontal pleiotropy by using genome-wide information,

effectively controlling the type-I error in cis-MR.

Planning for Efficient Accelerated Life Testing

Experiments

Xiaojian Xu

Brock University

Abstract: Accelerated life testing (ALT) enables rapid data collection on product lifetime and reliability by

exposing products to elevated stress levels. In this

paper, we present methods for obtaining optimal designs, with time-censoring, for step-stress ALTs to

efficiently estimate product reliability under normal

usage conditions. We assume a Weibull life distribution and log-linear life-stress relationships, with a

potentially non-constant shape parameter. We consider

several design optimality criteria, with the primary

criterion being the minimization of the asymptotic

variance or mean squared error (MSE) of the maximum likelihood estimator (MLE) of the reliability

estimator. Secondary criteria include the maximization

of either the determinant of the Fisher information

matrix or its trace. Additionally, we address potential

deviations from the model assumptions to ensure that

the recommended planning is robust.

Addressing Dispersion in Mis-measured Multivariate Binomial Outcomes: A Novel Statistical Approach for Detecting Differentially Methylated

Regions in Bisulfite Sequencing Data

Kaiqiong Zhao

York University

Abstract：Motivated by a DNA methylation application, this work addresses the problem of fitting and

inferring a multivariate binomial regression model for

outcomes that are contaminated by errors and exhibit

extra-parametric variations, also known as dispersion.

219

While dispersion in univariate binomial regression has

been extensively studied, addressing dispersion in the

context of multivariate outcomes remains a complex

and relatively unexplored task. The complexity arises

from a noteworthy data characteristic observed in our

motivating dataset: non-constant yet correlated dispersion across outcomes. To address this challenge

and account for possible measurement error, we propose a novel hierarchical quasi-binomial varying coefficient mixed model, which enables flexible dispersion patterns through a combination of additive and

multiplicative dispersion components. To maximize

the Laplace-approximated quasi-likelihood of our

model, we further develop a specialized two-stage

Expectation-Maximization (EM) algorithm, where a

plug-in estimate for the multiplicative scale parameter

enhances the speed and stability of the EM iterations.

Simulations demonstrated that our approach yields

accurate inference for smooth covariate effects and

exhibits excellent power in detecting non-zero effects.

Additionally, we applied our proposed method to investigate the association between DNA methylation,

measured across the genome through targeted custom

capture sequencing of whole blood, and levels of anti-citrullinated protein antibodies (ACPA), a preclinical marker for rheumatoid arthritis (RA) risk. Our

analysis revealed 23 significant genes that potentially

contribute to ACPA-related differential methylation,

highlighting the relevance of cell signaling and collagen metabolism in RA. We implemented our method

in the R Bioconductor package called \"SOMNiBUS\".

Graphical Principal Component Analysis of Multivariate Functional Time Series

Jianbin Tan

Duke university

Abstract：In this paper, we consider multivariate

functional time series with a two-way dependence

structure: a serial dependence across time points and a

graphical interaction among the multiple functions

within each time point. We develop the notion of dynamic weak separability, a more general condition

than those assumed in literature, and use it to characterize the two-way structure in multivariate functional

time series. Based on the proposed weak separability,

we develop a unified framework for functional graphical models and dynamic principal component analysis, and further extend it to optimally reconstruct signals from contaminated functional data using graphical-level information. We investigate asymptotic

properties of the resulting estimators and illustrate the

effectiveness of our proposed approach through extensive simulations. We apply our method to hourly

air pollution data that were collected from a monitoring network in China.

Joint work with Hui Huang.

Decision Support System of Juvenile Hacking

Classification

Siying Guo

Wayne State University

Abstract：In the digital age with data and artificial

intelligence booming, juvenile hacking, which involves adolescents gaining unauthorized access to and

manipulating computer systems, has become a major

concern. This research tends to develop a more precise decision support system of juvenile hacking classification via the use of a stacking ensemble learning

(SEL) that combines 20 machine learning (ML) models using a meta-learner. The base models are trained

based on a complete training set of teenager hacking

data, then the meta-model is trained on the outputs of

the base models to make a final classification of

hacking behavior. This method harnesses the predictive power of a series of models, thereby enhancing

the generalization capability and mitigating the risk of

selection bias associated with individual algorithms.

The proposed SEL model ensures the flexibility and

achieves greater accuracy than the general and generic

ML models, which is more transparent than the black

box model inherent in the realm of ML. More importantly, the factors triggering hacking behavior are

ranked via the SEL model and then the selected features get to decide the maximum depth of the decision

tree (DT). The pathway from data processing, feature

engineering, SEL to DT interpretation model (DTIM)

elucidates the relationship and decisive mechanism

among the risk factors. The research framework is

220

able to interpret the decision chain of hacking behavior through teenagers’ risk factors.

Contributed Session CS051： Quantile Regression

and Dimension Reduction

Tree-guided Compositional Variable Selection

Analysis of Microbiome Data

Yicong Mao

Peking University

Abstract：Studies of microbial communities, represented by the relative abundances of taxa at various

taxonomic levels, have underscored the significance

of microbiota in numerous aspects of human health

and disease. A pivotal challenge in microbiome research lies in pinpointing microbial taxa associated

with disease outcomes, which could play crucial roles

in prevention, detection, and treatment of various

health conditions. Alongside these relative abundance

data, the presence of a phylogenetic tree offers a

unique lens to explore the impact of shared evolutionary histories on patterns of microbial abundance. In

pursuit of this goal, we utilize the tree structure to

more flexibly identify taxa associated with disease

outcomes. To enhance the accuracy of our selection

process, we introduce auxiliary knockoff copies of

microbiome features designated as noise. This approach allows for the assessment of false positives in

the selection process and aids in refining it towards

more precise outcomes. Extensive numerical simulations demonstrate that our methodology outperforms

two existing methods in terms of selection accuracy.

Furthermore, we demonstrate the practicality of our

approach by applying it to a widely used gut microbiome dataset, identifying microbial taxa linked to

body mass index.

Joint work with Zhiwen Jiang,Tianying Wang,Xiang

Zhan.

Heterogeneous Predictability on Mutual Fund Alphas: A Sparse Clustering GMM Approach

Jiangshan Yang

City University of Hong Kong

Abstract：Mutual fund managers’ skills, alphas for

risk-adjusted performance, are predictable and

time-varying with a large number of fund characteristics using machine learning (e.g., Kaniel et al., 2023;

DeMiguel et al., 2023). However, this recent literature

only focuses on the aggregate predictability and neglects the potential heterogeneous data structure in the

cross section of mutual fund estimated alphas: most

do not have positive alphas. Furthermore, the predictability of fund characteristics in generating alphas

may be heterogeneous and time-varying between fund

managers with or without skills. This paper introduces

a novel nonparametric clustering method for identifying and estimating latent group structures to distinguish this small set of skillful fund managers from

others. The fund managers are grouped by their market timing skills, so managers within the same group

share group-specific parameters on aggregate market

predictors. In general, our approach sparse clustering

GMM (SCGMM) enables the simultaneous identification and estimation of group-specific parameters,

heterogeneous functional forms, and sparsity in a

potentially high-dimensional set of covariates without

pre-specifying the functional forms or the number of

groups. Finally, our empirical study in the U.S. mutual

fund data shows that only a small proportion of funds

have predictable alphas, which depend on a few aggregate market predictors.

Sliced Inverse Regression with Large Structural

Dimensions

Dongming Huang

National University of Singapore

Abstract：The central space of a joint distribution

(x, y) is the minimal subspace ? such that Y ⊥

X |P? where ?? is the projection onto ?. Sliced inverse regression (SIR), one of the most popular

methods for estimating the central space, often performs poorly when the structural dimension d =

dim (?) is large (e.g., ≥ 5 ). In this paper, we

demonstrate that the generalized signal-noiseratio

(gSNR) tends to be extremely small for a general

multiple-index model when d is large. Then we determine the minimax rate for estimating the central

space over a large class of high dimensional distributions with a large structural dimension d (i.e., there is

221

no constant upper bound on d) in the low gSNR regime. This result not only extends the existing minimax rate results for estimating the central space of

distributions with fixed d to that with a large d, but

also clarifies that the degradation in SIR performance

is caused by the decay of signal strength. The technical tools developed here might be of independent

interest for studying other central space estimation

methods.

Joint work with Songtao Tian,Qian Lin.

Estimation of Average Treatment Effect for

High-Dimensional Panel Data via Random Forests-Based Variable Selection

Xuehong Luo

Xiamen University

Abstract：：It is challenging to conduct controlled

experiments to assess the impacts of social policy. To

address this, Hsiao et al. (2012) propose a panel data

approach using factor models to estimate average

treatment effects. The selection of control units is a

critical step to balance the goodness of fit within-sample with the post-treatment forecasting error

when the number of observed potential control units is

sufficiently large. In this study, we propose using

random forests, an ensemble learning method, which

offers robustness and requires fewer candidate models

compared to existing methods. We demonstrate that

our approach effectively selects almost all relevant

control units, and we provide rigorous asymptotic

normality results and significance tests for policy

interventions. Extensive simulations confirm the

method's superior performance. Applying our approach to evaluate the impact of Brexit on the United

Kingdom's GDP growth, we find an average quarterly

decline of 1.86 percentage points between 2017 and

2019, primarily attributed to reduced private consumption and investment.

Joint work with Wei Long, Guannan Liu.

A Likelihood Function Perspective on Quantile

Regression Model

Zhe Jiang

Kunming University of Science and Technology

Abstract：Quantile regression is one of the classical

regression methods, which is loved by statisticians

because of its robustness and ability to interpret different quantiles. However, most of the current discussion of quantile regression models is limited to the

loss function of least absolute deviation, which leads

to problems such as the not differentiable of the objective loss function and the inability to model heteroskedasticity. In this paper, the quantile regression

model is studied from the perspective of likelihood

function, and a quantile regression model for likelihood function (QRLF) is proposed. When the residuals are normal distributed, t-distributed, and

skew-normal distributed, N-type QRLF, T-type QRLF,

and SN-type QRLF are discussed, respectively. For

heteroskedasticity, an N-type joint quantile and variance regression model (N-Type JQVM) was proposed.

Finally, the proposed method is applied to the prediction of market value at risk (VAR), and the simulation

research and case analysis show the effectiveness of

the proposed method.

Joint work with Liucang Wu.

Contributed Session CS052： Statistical Models

and Methods in Economics and Finance

Earning Rate and Financial Risk Prediction Model

Based on Transformer and WGAN

基于 Transformer 和 WGAN 的收益率与金融风险预

测模型

Haiqing Chen

Nanjing University of Finance and Economics

摘要：本文在 Transformer 模型基础上，改变部分

结构，充分融合数值特征、股民评论、新闻资讯、

专家研报等多源信息，创新性地构建 Trans-WGAN

生成对抗网络模型，该模型同时具备 Transformer

提取特征和生成对抗网络学习收益率分布的能力。

通过模型的生成器对收益率分布进行抽样，并利用

样本的经验分布对收益率、在险价值 VaR 以及预

期损失 ES 的进行估计。实证研究表明，将各类指

标合理分类后，嵌入到同一浅层的特征空间，通过

Transformer 模块构建生成器，模型预测效果显著提

高。 最终选取行业具有代表性的 8 支股票进行实

证研究，并与其他常见模型对比，本文所提出收益

率与金融风险预测模型具有显著性优势，进一步丰

222

富了深度学习与多源数据融合技术在金融科技中

的应用，为金融预测和风险评估提供了新思路。

Ising Network Analysis with Missing Data

Siliang Zhang

East China Normal University

Abstract：The Ising model has become a popular

psychometric model for analyzing item response data.

The statistical inference of the Ising model is typically

carried out via a pseudo-likelihood, as the standard

likelihood approach suffers from a high computational

cost when there are many variables (i.e., items). Unfortunately, the presence of missing values can hinder

the use of pseudo-likelihood, and a listwise deletion

approach for missing data treatment may introduce a

substantial bias into the estimation and sometimes

yield misleading interpretations. This paper proposes

a conditional Bayesian framework for Ising network

analysis with missing data, which integrates a pseudo-likelihood approach with iterative data imputation.

An asymptotic theory is established for the method.

Furthermore, a computationally efficient Polya-Gamma data augmentation procedure is proposed

to streamline the sampling of model parameters. The

method's performance is shown through simulations

and a real-world application to data on major depressive and generalized anxiety disorders from the National Epidemiological Survey on Alcohol and Related

Conditions (NESARC).

Joint work with Yunxiao Chen.

Least Squares Model Averaging for Distributed

Data

Haili Zhang

Shenzhen Polytechnic University

Abstract：Divide and conquer algorithm is a common

strategy applied in big data. Model averaging has the

natural divide-and-conquer feature, but its theory has

not been developed in big data scenarios. The goal of

this paper is to fill this gap. We propose two divide

and conquer type model averaging estimators for linear models with distributed data. Under some regularity conditions, we show that the weights from Mallows model averaging criterion converge in L2 to the

theoretically optimal weights minimizing the risk of

the model averaging estimator. We also give the

bounds of the in-sample and out-of-sample mean

squared errors and prove the asymptotic optimality for

the proposed model averaging estimators. Our conclusions hold even when the dimensions and the number

of candidate models are divergent. Simulation results

and a real airline data analysis illustrate that the proposed model averaging methods perform better than

the commonly used model selection and model averaging methods in distributed data cases. Our approaches contribute to model averaging theory in

distributed data and parallel computations, and can be

applied in big data analysis to save time and reduce

the computational burden.

Joint work with Zhaobo Liu and Guohua Zou.

Spatiotemporal Joint Analysis of PM2.5 and Ozone

in California with INLA Approach

Chengxiu Ling

Xi'an Jiaotong-Liverpool University

Abstract：The substantial threat of concurrent air

pollutants to public health is increasingly severe

under climate change. To identify the common drivers

and extent of spatiotemporal similarity of PM2.5 and

ozone (O3), this paper proposed a log Gaussian-Gumbel Bayesian hierarchical model allowing for

sharing a stochastic partial differential equation and

autoregressive model of order one (SPDE-AR(1))

spatiotemporal interaction structure. The proposed

model, implemented by the approach of integrated

nested Laplace approximation (INLA), outperforms in

terms of estimation accuracy and prediction capacity

for its increased parsimony and reduced uncertainty,

especially for the shared O3 sub-model. Besides the

consistently significant influence of temperature (positive), extreme drought (positive), fire burnt area (positive), gross domestic product (GDP) per capita (positive), and wind speed (negative) on both PM2.5 and

O3, surface pressure and precipitation demonstrate

positive associations with PM2.5 and O3, respectively.

While population density relates to neither. In addition,

our results demonstrate similar spatiotemporal interactions between PM2.5 and O3, indicating that the

223

spatial and temporal variations of these pollutants

show relatively considerable consistency in California.

Finally, with the aid of the excursion function, we see

that the areas around the intersection of San Luis

Obispo and Santa Barbara counties are likely to exceed the unhealthy O3 level for USG simultaneously

with other areas throughout the year. Our findings

provide new insights for regional and seasonal strategies in the co-control of PM2.5 and O3. Our methodology is expected to be utilized when interest lies in

multiple interrelated processes in the fields of environment and epidemiology.

July 14 8:30-12:00

Poster 001: Hierarchical Semi-Implicit Variational

Inference with Application to Diffusion Model Acceleration

Longlin Yu

Peking University

Abstract: Semi-implicit variational inference (SIVI)

has been introduced to expand the analytical variational families by defining expressive semi-implicit

distributions in a hierarchical manner. However, the

single-layer architecture commonly used in current

SIVI methods can be insufficient when the target posterior has complicated structures. In this paper, we

propose hierarchical semi-implicit variational inference, called HSIVI, which generalizes SIVI to allow

more expressive multi-layer construction of

semi-implicit distributions. By introducing auxiliary

distributions that interpolate between a simple base

distribution and the target distribution, the conditional

layers can be trained by progressively matching these

auxiliary distributions one layer after another. Moreover, given pre-trained score networks, HSIVI can be

used to accelerate the sampling process of diffusion

models with the score matching objective. We show

that HSIVI significantly enhances the expressiveness

of SIVI on several Bayesian inference problems with

complicated target distributions. When used for diffusion model acceleration, we show that HSIVI can

produce high quality samples comparable to or better

than the existing fast diffusion model based samplers

with a small number of function evaluations on various datasets.

Joint work with Tianyu Xie, Yu Zhu, Tong Yang,

Xiangyu Zhang and Cheng Zhang.

Poster 002: Rerandomization for Covariate Balance Mitigates p-hacking in Covariate Adjustment

Xin Lu

Tsinghua University

Abstract: Rerandomization is commonly employed in

randomized controlled trials (RCTs) to achieve balance across treatment groups. Despite its intuitive

appeal, the theoretical justification for rerandomization remains insufficient and warrants further exploration. In this study, we address the question of why

rerandomization should be employed in practice, particularly in the context of mitigating false discoveries

resulting from p-hacking, the practice of strategically

selecting covariates in covariate adjustment to get

better p-values. We present evidence suggesting that

rerandomization mitigates the issue of p-hacking.

Moreover, when implemented with a sufficiently

stringent threshold, it is able to solve the problem of

p-hacking. Additionally, we offer guidance on choosing such threshold in practice. Our discussion begins

with an exploration of rerandomization using Mahalanobis distance in RCTs and subsequently extends

to other rerandomization measures and stratified randomized experiments. Through these discussions, we

demonstrate the broad applicability of rerandomization in controlling false discoveries. The insights

gained and recommendations provided in this study

offer valuable guidance to practitioners involved in

conducting RCTs.

Joint work with Peng Ding.

224

Poster 003: Evaluating High-Quality Port Economy Development: A Yunnan Case Study

口岸经济高质量发展水平评价指标体系构建与测

度——基于云南的实证研究

Aixia Fan

Yunnan Minzu University

摘要： 加快口岸由“通道经济”转型为“产业经

济”，是“一带一路”建设高质量发展的重要内容

。基于高质量发展目标要求，对口岸经济高质量发

展的内涵、特征进行学理性探讨，依托层次分析思

想构建测度口岸经济高质量发展水平的评价指标

体系。以边境口岸数量多、类型丰富的云南省为研

究对象，选取 2012-2021 年数据， 运用 CRITIC

权重法，对云南口岸经济高质量发展水平进行测度

并分析其趋势特征。研究发现: 样本期内云南口岸

经济高质量发展水平总体呈现上升趋势，口岸经济

协调发展和开放发展提升显著，创新发展有待加强

， 共享作用显现，绿色发展不平稳，岸城协调性

不足。推动云南口岸经济高质量发展要转变发展理

念， 扩大高水平对外开放，建立岸岸、岸城协调

发展机制，完善口岸基础设施，加强口岸开放发展

能力， 以此推进云南口岸现代化建设进程。

Joint work with Hailan Pan and Xiaoqin Wang.

Poster 004: Angle Covariance-Based Independence

Tests for Multivariate Functional Data

Jiangyuan Bian

Beijing University of Technology

Abstract: This article is focused on the problem to

measure and test pairwise and mutual independence

for multivariate functional data. When testing for

mutual independence, we proposed two aggregation

measures based on angle covariance and give their

asymptotic properties. And both measures have

good properties such as each equals zero if and only

if the random functional elements are mutually independent. The proposed measures of mutual independence are simple in form and easy to compute using the

given algorithm which based on permutation. Finally,

we evaluate the performance of the proposed measures through some simulation studies and

a real data example.

Joint work with Zhongzhan Zhang.

Poster 005: Dynamic Bayesian Model for Earthquake Loss Frequency: Index Insurance Applications

基于动态贝叶斯模型的地震巨灾损失频率分析—

地震指数保险中的应用

Xinmei Yang

Yunnan University of Finance and Economics

摘要：我国是世界上地震灾害最严重的国家之一，

给我国人民生命财产安全造成了较大危害。为此，

完善地震巨灾保险，建立健全重大风险防范机制极

为迫切。地震巨灾保险的完善依赖于对地震巨灾风

险的准确评估，包括地震灾害频率与程度。本文将

针对地震发生频率进行分析，引入动态贝叶斯框架

下的广义泊松模型，对全国地震损失数据进行分析

。与传统的损失频率模型相比，例如零膨胀泊松模

型，动态贝叶斯模型具有如下优点：第一、能够刻

画损失频率的动态变化，符合地震灾害特征；第二

、贝叶斯方法能够缓解小样本下估计偏差问题。在

对地震灾害损失频率进行估计后，即可对地震巨灾

保险费率进行较为准确的评估，完善当前的地震巨

灾保险，提升我国处置地震灾害事故的能力。

Joint work with Tai An.

Poster 006: Influences of Cognitive and Social

Factors on Insurance Demand in the Elderly

认知能力、社会互动与中老年家庭商业保险需求

Dezhi Zhao

Yunnan University of Finance and Economics

摘要：本文基于中国家庭追踪调查( CFPS) 数据，

从行为经济学的视角出发，采用结构方程模型，实

证检验了中老年人认知能力、社会互动对家庭商业

保险需求的影响。结合现实情况，主要区分了传统

社会互动(线下社会互动)和互联网使用（线上社会

互动）这两大信息途径。结果表明，认知能力的提

升可以增加中老年家庭商业保险的需求，线下社会

互动和线上社会互动这两大信息获取途径在其中

起到部分中介作用。在认知能力对中老年家庭商业

保险需求的影响中，互联网使用这一途径对中老年

家庭商业保险需求的影响是不显著的，即互联网的

使用，主要作用于对线下信息渠道社会互动的增强

，中老年家庭在购买商业保险时，可能更倾向于相

信熟人朋友的线下互动交流，而不会直接通过网上

信息就决定购买商业保险。作为对现有研究的补充

225

，本文的研究结果有助于推动商业保险在实现\"积

极老龄化\"过程中发挥重要作用。

Poster 007: Effects of Environmental Liability

Insurance on Environmental Quality

环境污染责任险的实施对提升环境质量水平的实

证研究

Zhilan Zi

Yunnan University of Finance and Economics

摘要: 双碳背景下，绿色保险作为绿色金融的重要

组成部分，在我国绿色改革的过程承担着保驾护航

的职责。本文采用实证分析方法，探究了我国环境

污染责任险与环境质量之间的作用关系，使用了我

国 30 个省份 2011 年-2020 年的面板数据，进行了

多维固定效应回归分析和分组回归。研究结果表明

环境污染责任险的推广确实对促进环境质量提升，

但正向推动作用不明显，这与我国政策设计的初衷

不相符。另外研究发现，我国环境污染责任险的发

展程度地区之间差距较大。本研究丰富了绿色发展

的研究内容，同时对我国环境污染责任险相关政策

的制定和执行提供一定的决策依据。

Poster 008: Meta-Propagation Dynamics Model for

COVID-19 Transmission Monitoring

基于 Meta-传播动力学模型的新型冠状病毒感染传

播监测

Wenhui Zhu

Sichuang University

摘要：

目的：基于新发急性呼吸道传染病流行初期数据存

在高度不确定性以及样本量少的特点，以及算法在

求解过程面临的收敛速度较慢、参数估计准确性较

低、计算成本高、容易出现离群值等方法学上的困

难，如何在不确定性参数空间中确保获得关键流行

病学参数的稳定解成为首要问题。本研究通过统计

模拟的思路为该类问题提供求解策略，采用不同的

算法估计确定传播动力学和随机传播动力学模型

的参数，探究不同算法分别在估计确定和随机传播

动力学模型参数的准确性、计算效率等，从而评估

其在新发急性呼吸道传染病流行初期传播情况实

时监测中的适用性。

方法：本研究分别采用优化算法和蒙特卡洛类算法

估计确定传播动力学和随机传播动力学模型的参

数，即 Nelder-Mead 单纯形算法、共轭梯度算法、

拟 牛 顿 算 法 、 Limited-memory-Broyden-FletcherGoldfarb-Shanno-bound (L-BFGS-b)算法、马尔科夫

链蒙特卡洛算法、粒子马尔科夫链蒙特卡洛(Particle

Markov Chain Monte Carlo, PMCMC)算法。首先，

通过模拟研究探索不同算法在新发急性呼吸道传

染病流行初期的适用性能，分别设置确定和随机传

播动力学模型构建模拟数据集；然后采用不同算法

估计模型参数并从估计准确度、计算效率等方面比

较算法性能的优劣，将参数代入模型评估预测的准

确性，为新发急性呼吸道传染病流行初期的参数估

计方法提供选择依据。

实证研究分为风险评估（模拟不同干预措施并探索

大型国际活动中 COVID-19 的有效干预措施）、估

计时变再感染率（SARS-CoV-2 再感染报告数显著

增加，但再感染率的变化趋势尚不清楚，本研究基

于全球证据探索再感染率的变化趋势）两部分。基

于易感-暴露-有症状感染-无症状感染-康复-住院

(SEIARH)模型，模拟不同干预措施并计算风险评估

指标。检索 PubMed、Web of Science、Medline、

Embase、Cochrane Controlled Trials Register、知网

和万方数据库，对截至 2023 年 3 月 16 日进行全球

SARS-CoV-2 再感染率研究进行 meta 分析，并使用

meta 回归估计再感染率的时变特征。

结果：（1）模拟研究探索了不同参数估计方法在

新发急性呼吸道传染病流行初期场景下的适用性

能。在确定传播动力学模型框架下，PMCMC 算法

估计结果的感染者仓室人数的估计偏差、平均绝对

误差、均方根误差、离差平方和最小。在随机传播

动力学模型框架下，PMCMC 算法估计结果中的易

感者、感染者以及新增感染者人数的离差平方和最

小，其它算法的离差平方和均高出 PMCMC 较多。

总体而言，PMCMC 算法在随机传播动力学模型估

计的准确性上优于其他算法，尤其是识别感染者人

数、新增感染者人数的变化。

（2）由日本奥委会提出的 COVID-19 预防措施需

要加强，大规模接种疫苗将有效控制 COVID-19 的

传播，当疫苗的保护效力为 78.1%或 89.8%，并且

运动员的接种率达到 80%时，就可以建立一道免疫

屏障。基于 55 项研究进行的 meta 分析显示综合

SARS-CoV-2 再 感 染 率 为 0.94%(95% CI:

0.65-1.35%)。Meta 回归分析发现再感染率随时间波

动，总体再感染率随时间先增加后减少，接着出现

了一个平稳期，然后出现了先增加后减少的趋势，

但第二波再感染率的峰值低于第一波。

226

结论：确定传播动力学模型框架下，虽然采样算法

的参数估计准确度低于优化算法，但是采样算法的

预测准确性高于优化算法；随机传播动力学模型框

架下，除了估计耗时，采样算法的整体表现优于优

化算法，而采样算法中的 PMCMC 算法在采样有效

率、估计精度等方面要优于 MCMC 算法。综上，

在对新发急性呼吸道传染病流行初期的传播参数

进行估计时，建议构建随机传播动力学模型框架并

采用 PMCMC 算法估计模型参数、评估传播情况。

Joint work with Yin Chen and Tao Zhang.

Poster 009: Association between TB Delay and TB

Treatment Outcomes in HIV-TB Co-infected Patients: A Study Based on the Multilevel Propensity

Score Method

Lin Hu

Sichuang University

Abstract: Background: HIV-tuberculosis (HIV-TB)

co-infection is a significant public health concern

worldwide. TB delay, consisting of patient delay, diagnostic delay, treatment delay, increases the risk of

adverse anti-TB treatment (ATT) outcomes. Except

for individual level variables, differences in regional

levels have been shown to impact the ATT outcomes.

However, few studies appropriately considered possible individual and regional level confounding variables. In this study, we aimed to assess the association

of TB delay and treatment outcomes in HIV-TB

co-infected patients in Liangshan Yi Autonomous

Prefecture (Liangshan Prefecture) of China, using a

causal inference framework while taking into account

individual and regional level factors.

Methods: We conducted a study to analyze data

from 2068 patients with HIV-TB co-infection in

Liangshan Prefecture from 2019 to 2022. To address

potential confounding bias, we used a causal directed

acyclic graph (DAG) to select appropriate confounding variables. Further, we controlled for these confounders through multilevel propensity score and

inverse probability weighting (IPW).

Results: The successful rate of ATT for patients

with HIV-TB co-infection in Liangshan Prefecture

was 91.2%. Total delay (OR=1.411, 95% CI: 1.015,

1.962), diagnostic delay (OR=1.778, 95% CI: 1.261,

2.508) , treatment delay (OR=1.749, 95% CI: 1.146,

2.668) and health system delay (OR=1.480 95% CI:

(1.035, 2.118) were identified as risk factors for successful ATT outcome. Sensitivity analysis demonstrated the robustness of these findings.

Conclusions: HIV-TB co-infection prevention

and control policy in Liangshan Prefecture should

prioritize early treatment for diagnosed HIV-TB

co-infected patients. It is urgent to improve the health

system in Liangshan Prefecture to reduce delays in

diagnosis and treatment.

Joint work with Tao Zhang.

Poster 010: Reinforcement Learning Using Neural

Networks in Estimating An Optimal Dynamic

Treatment Regime in Patients with Sepsis: A Retrospective Cohort Study Based on the MIMIC-III

Database

Weijie Liang

Peking University

Abstract: Sepsis presents as a clinical syndrome

characterized by systemic inflammation and infection,

and it stands as one of the leading causes of mortality

encountered in emergency departments (EDs) and

intensive care units (ICUs). Fluid resuscitation is considered a cornerstone early intervention for sepsis, but

the appropriate dosage of fluids in clinical practice is

still debated. This study aims to estimate the optimal

dynamic treatment regimes (DTRs) for sepsis patients

using the reinforcement learning framework. One of

the primary limitations encountered with basic

Q-learning resides in its sensitivity to the specification

of the stage-specific conditional mean outcomes

(called Q-function) since it usually assumes linear

parametric models. We propose a reinforcement

learning algorithm combined with neural networks

(RL-NN), using the flexibility of neural networks to

approximate any form of Q-function. Simulations

show that the proposed method outperforms other

methods in terms of both the percentage of correctly

classified optimal treatments and the predicted counterfactual mean outcome. We applied the proposed

method to the sepsis study cohort from the Medical

Information Mart for Intensive Care III (MIMIC-III),

227

providing recommendations for the fluid resuscitation

strategy during medical ICU treatment.

Joint work with Jinzhu Jia.

Poster 011: Construction of Cell-Cell Interaction

Network Based on Spatial Transcriptome Data

Dongyu Li

Tsinghua University

Abstract: Spatial transcriptome is informative to infer

cell cell communications. Existing methods focus on

inference of cell communications at cell type level,

but do not identify individual cells involved in the

communications. We propose a computational

framework based on graphical modeling to Infer

Cell-Cell Communication (IC3) network at single cell

level from spatial transcriptomic data. Our method is

built upon the heuristic that physically adjacent cells

are more likely to communicate with each other. We

have demonstrated in simulations that IC3 outperforms existing methods in accuracy, and improves

sensitivity to identify interactions involving rare cell

types. Applying IC3 to a study of mouse brain, IC3

reveals the spatial heterogeneity of cell communication network, thus providing additional insight of the

underlying biology.

Joint work with Lin Hou.

Poster 012: Stochastic Comparisons of SecondOrder Statistics from Dependent and Heterogenous Modified Proportional Hazard Rate Observations

Jiale Niu

Northwest Normal University

Abstract: This paper investigates the properties of

stochastic orders between second-order statistics of

two groups of dependent random variables with a

proportional hazard model. It provides sufficient conditions for the establishment of the usual stochastic

order between second-order statistics of two groups of

dependent heterogeneous random variables and establishes the hazard rate order and reversed hazard rate

order between second-order statistics of two groups of

independent heterogeneous random variables. Furthermore, it studies the hazard rate order and reversed

hazard rate order between second-order statistics of

random variables with a multiple-outlier model and

obtains the hazard rate order and reversed hazard rate

order between them. Finally, the main conclusions of

the paper are demonstrated through numerical examples.

Joint work with Rongfang Yan.

Poster 013: Network Model Averaging Prediction

for Latent Space Models by K-Fold Edge

Cross-Validation

Yan Zhang

Xiamen University

Abstract: In complex systems, networks represent

connectivity relationships between nodes through

edges. Latent space models are crucial in analyzing

network data for tasks like community detection and

link prediction due to their interpretability and visualization capabilities. However, when the network size

is small, and the true latent space dimension is large,

the latent space can not be estimated very well. To

overcome this, we propose a Network Model Averaging (NetMA) method tailored for latent space models

with varying dimensions, specifically focusing on link

prediction in networks. For both single-layer and multi-layer networks, we first establish the asymptotic

optimality of the proposed averaging prediction in the

sense of achieving the lowest possible prediction loss.

Then we show that when the candidate models contain

the correct models, our method assigns all weights to

the correct models. Furthermore, we demonstrate the

consistency of the NetMA-based weight estimator

tending to the optimal weight vector. Extensive simulation studies show that NetMA performs better than

simple averaging and model selection methods, and

even outperforms the \"oracle'' method when the real

latent space dimension is large. Evaluation on collaboration and virtual event networks further emphasizes

the competitiveness of NetMA in link prediction performance.

Poster 014: Gold Closing Price Prediction Model

Analysis

黄金收盘价预测模型分析

228

Zeying Wang

Inner Mongolia University of Finance and Economics

摘要：在金融投资领域，准确预测黄金价格对投资

者制定战略、降低风险至关重要。本文以黄金未来

四个月的收盘价为研究对象，通过构建时间序列预

测模型、机器学习模型和深度学习模型，对黄金价

格进行了全面而深入的预测。结果表明：机器学习

模型在预测精度上优于传统的时间序列模型，深度

学习作为机器学习研究中的一个新的领域，其预测

精度更高。

为了进一步提高预测精度，本文运用经验模态

分解（EMD）技术对黄金期货价格序列进行分解，

并将其与相关影响变量输入LSTM 模型。引入EMD

可以更细致地分析黄金价格的振荡特征，将信号分

解为不同频率的固有模态函数（IMF）。实证研究

显示，EMD 分解对于提高预测模型的准确性起到

了积极作用，进一步强化了对黄金价格未来趋势的

洞察力。

综上所述，本文通过多层面的对比分析，为金

融投资者提供了更全面、深入的黄金价格预测方案

。通过结合传统模型、机器学习模型、深度学习模

型和 EMD 分解技术，为投资者提供了更为准确和

可靠的市场分析工具，也为金融投资决策提供了更

具前瞻性的支持。

Poster 015: Urban Sprawl's Impact on Carbon

Emissions: Evidence from 260 Cities

城市蔓延对碳排放强度影响研究—基于260个地级

市的实证检验分析

Menghuan Gao

Anhui University of Finance and Economics

摘要：城市蔓延是经济发展过程中不可避免的趋势

，而双碳目标也是生产改革的重要一环。因此，如

何在城市蔓延的过程中降低碳排放强度成为亟需

解决的问题。文章基于 2011-2021 年中国 260 个城

市面板数据，使用动态面板模型、调节效应和中介

模型检测城市蔓延影响碳排放强度的效果及作用

机制。实证结果显示：城市蔓延有助于抑制碳排放

强度，这一结果在多种稳健性检验后依然成立；调

节效应表明，数字经济发展水平越高，城市蔓延对

碳排放强度抑制作用越强。机制检验表明，城市蔓

延可通过加强环境规制和较少工业用地来抑制碳

排放强度；异质性分析表明，在中西部地区和小中

城市地区，城市蔓延降低城市碳排放强度效果更显

著。

Poster 016: Digital Economy and Regional Economic Integration in Beijing-Tianjin-Hebei and

Inner Mongolia

数字经济驱动京津冀蒙经济高质量一体化发展的

路径分析

Chunying Liu

Inner Mongolia University of Finance and Economics

摘要：目前，提高城乡融合发展的质量与效率是实

现区域协调发展战略的必然要求，也是促进我国经

济高质量发展的关键。同时，数字经济也日益成为

推动区域发展的\"新引擎\"，也是推动区域高质量一

体化发展的重要路径。通过对数字经济在经济高质

量一体化发展中的驱动效应的分析，可以把这两个

方面进行有机地融合起来，这对优化经济结构、深

化数字经济发展与实体经济改革、深化高质量一体

化发展为新发展格局服务都有着十分重要的意义。

基于此，本文以京津冀蒙城市群作为研究区域，探

究京津冀蒙城市群数字经济对经济高质量一体化

发展的驱动效应，为数字经济背景下提升区域高质

量一体化发展质量提供一种分析方案，同时也为我

国实现高质量一体化发展提供区域借鉴。

研究发现：在 2017 年至 2021 年，京津冀蒙城市

群各省市数字经济水平整体呈现逐年上升趋势；京

津冀蒙城市群数字经济与高质量一体化发展的基

尼系数总体均呈上升趋势，同时差异变动趋势存在

一定的相关性；高质量一体化发展的莫兰指数呈现

出先下降后上升的趋势，空间正相关性呈现逐渐增

强的现象。莫兰散点图显示发展水平没有明显的象

限迁移特征；从对京津冀蒙城市群高质量一体化发

展的赋能作用来看，核心变量数字经济的系数为

0.7516，存在明显的正向赋能效应，直接效应与溢

出效应系数均通过 1%显著水平检验。

Poster 017: Value Co-Creation and Destruction in

Sharing Contexts

共享情境下价值共创和价值共毁间的关系研究

Wenjun Yang

Henan University of Economics and Law

摘要：以往关于价值共创的研究过多地探讨了正面

效应，忽视其潜在负面影响，如价值共毁。本研究

基于社会认知理论，从房东价值共创公民行为角度

，采用多时点、多源的实地问卷调查法，收集某短

229

租平台 360 份房东—房客配对数据，探讨共享情境

下用户接触阶段房东价值共创公民行为何时以及

如何诱发房客的价值共毁行为。研究结果表明：房

东价值共创公民行为会引发房客的心理权利感，进

而增加其价值共毁行为，房客权力距离在这一过程

中起调节作用。具体来说，房客的权力距离导向较

高时，房东志愿服务通过房客心理权利感影响其价

值共毁行为的正向中介效应显著。当房客的权力距

离导向较低时，房客价值共创公民行为通过房客心

理权利感产生的房客价值共毁行为相对减弱。

Joint work with AoXiang Li.

Poster 018: Digital Economy's Impact on Regional

Innovation

数字经济推动区域创新能力研究

Zhiqing Zhang

Anhui University of Finance and Economics

摘要：二十大提出加快实施创新驱动发展战略，加

快实现高水平科技自立自强，为实现这一目标，数

字经济对科技创新的驱动作用不言而喻。本文基于

2012-2022 年中国省际面板数据构建数字经济发展

水平评价体系，并采用固定效应模型对数字经济是

否促进了创新进行了回归估计。研究表明，数字经

济发展对区域创新能的提升有正向作用，但其潜力

尚未充分发挥。此外通过了一系列内生性和稳健性

检验，在异质性分析中，对全国不同区域数字经济

的分组回归中发现，东部地区数字经济水平高，其

区域创新所受影响也越大。专利异质结果表明，数

字经济对于实用型专利的积极贡献最为显著，而发

明型专利是衡量一个地区创新能力的重要指标，是

推动科学技术进步和产业升级的关键。鉴于此，应

全面推动数字经济量质齐升，各区域应立足自身实

际精准施策，在推进新型基础设施建设的同时，主

攻新型数字技术与当地制造业的融合，加速创新理

论与应用的转化，缩小地区发展差异，最终建设\"

创新型国家\"。

Joint work with Chao Li.

Poster 019: Conditional Independence Test Based

on a Classifier

Xiaodong Wang

Shanxi Normal University

Abstract: Testing conditional independence for continuous variables is a fundamental but challenging

task in statistics. Especially in the absence of distribution or structural assumptions, conditional independence testing becomes more challenging. We first

transform the problem of conditional independence

testing between variables into a problem of independence testing between variables. Based on the

transformation, we propose a general framework for

independent testing by fitting a classifier that distinguishes between joint distribution and product distribution, and testing the significance of the fitted classifier. Our test statistic has a universal, fixed Gaussian

null distribution that is independent of the underlying

data distribution. The proposed method is computationally efficient and easy to implement. Moreover,

the effectiveness of the method is illustrated through

extensive simulations.

Joint work with Xia Chen.

Poster 020: Digital Inclusive Finance's Effects on

Income: China's Provincial Panel Data Analysis

双循环视角下数字普惠金融对居民收入水平影响

的异质性研究——来自中国省级面板数据的经验

证据

Chenyi Zhu

Anhui University of Finance and Economics

摘要：随着双循环格局的提出，各国的经济发展水

平都得到了显著的提升。目前，互联网在多个行业

领域都得到了迅速的发展，在此基础下，数字普惠

金融已逐渐成为经济发展的活力源泉。本文以中国

31 个省份在 2011 年—2020 年的面板数据，研究在

双循环视角下数字普惠金融对居民收入水平的影

响，并从地区差异、城乡差异、收入水平效果三个

方面展开研究分析。

Poster 021: Carbon Reduction Effectiveness in the

Yellow River Urban Areas: Policy Impact Analysis

黄河流域城市碳减排成效评估研究——低碳城市

试点政策的时空动态政策效应视角

Aorigele Wu

Inner Mongolia University of Finance and Economics

摘要：基于黄河流域 2009—2022 年 97 个地级市面

板数据，以低碳城市试点政策的实施视为准自然实

验，运用渐进双重差分模型评估了该政策对黄河流

域城市碳减排成效的作用机制及其时空动态变化。

230

研究发现，低碳城市试点政策显著降低了试点城市

的碳排放强度并提升了碳排放效率。随着时间的推

移，试点政策效果逐渐增强，并且碳排放强度抑制

效应可传播至 0-150km 范围内的邻接非试点城市，

碳排放效率提升效应可传播至300-400km范围内的

邻接非试点城市。机制分析表明，低碳城市试点政

策可以通过强化环境规制、提升人力资本以及激发

城市创业活力来促进黄河流域城市碳减排成效。异

质性分析表明，低碳城市试点政策显著降低了黄河

上游以及金属矿型城市的碳排放强度，显著提升了

黄河中游、成长型以及煤炭型城市的碳排放效率。

Joint work with Chunzhi Wang.

Poster 022: Stock Index Forecasting Using Social

Media Sentiment

基于网络社交媒体情绪的股指预测研究

Jiayuan Li

Inner Mongolia University of Finance and Economics

摘要：市场情绪对股票指数的变化有重要影响。为

构建新型股票指数量化指标，提高量化模型对股票

指数的预测精度，本文运用网络爬虫技术从网络社

交媒体采集海量金融文本数据，使用经过人工标注

的样本数据训练 BERT-BILSTM 模型来判断金融文

本情感。结合成分股情绪信息和指数权重数据计算

新型股指量化指标——加权情感得分，在此基础上

构建了 WS-LSTM 模型预测沪深 300 指数、上证 50

指数和上证 180 指数。实证结果显示：WS-LSTM

在股指预测任务中表现出色，验证了本文设计的加

权情感得分对股票指数有较强的预测作用。与其他

模型相比，WS-LSTM 模型对股票指数的预测精度

更高，说明加权情感得分比其他市场情绪指标包含

更多准确情感信息，对股票指数的预测能力更强。

本文丰富了自然语言处理和深度学习技术在股票

投资领域的研究，也为量化交易指标的设计提供了

新思路。

Joint work with Zhifang Chen.

Poster 023: High-Frequency Trading Strategy Optimization with Reinforcement Learning

基于强化学习的高频交易策略优化及应用

Yuanying Zhuang

Jimei University

摘要：随着人工智能成为人们关注的焦点以及 AI

技术的突破，机器学习在各个领域的应用也成为了

研究人员们的焦点，目前国内已经有不少学者考虑

将机器学习和高频交易相结合，通过机器学习来训

练出更加高效的高频交易策略。本文研究目的便在

于探究强化学习用在高频交易的效果，在本文中，

将采用 DQN 算法在仿真的期货环境中，学习训练，

以此来形成一个有效的高频交易策略。通过实验结

果我们可以看出，强化学习训练后的高频交易策略

是有可能取得较为良好的回报率的。

Joint work with Chunxu Jin.

Poster 024: Linking Economic Policy Uncertainty,

Investor Sentiment, and Financial Risk

经济政策不确定性、投资者情绪和系统性金融风险

Xuehua Zhou

Anhui University of Finance and Economics

摘要：本文在运用 TVP-FAVAR 模型构建系统性金

融风险度量指标的基础上，利用 TVP-SVVAR 模型

对经济政策不确定性（ EPU）冲击下系统性金融

风险的动态演变关系展开研究，并从金融市场和经

济基本面两个层面检验 EPU 冲击下系统性金融风

险的跨市场传染机制。此外，考虑到中国资本市场

投资者存在非理性行为，本文还评估了不同维度的

投资者情绪在 EPU 与系统性金融风险关系中的作

用。研究发现：EPU 对系统性金融风险具有正向冲

击，中短期效应较长期效应要稳定且更显著，而长

期效应持续上升、反映强烈，表明了经济政策长期

的不确定性；不同维度的投资者情绪对系统性金融

风险也具有正向冲击，其中微观投资者情绪冲击效

应最强，中观投资者情绪次之，宏观投资者情绪冲

击效应最小；各金融子市场和经济部门受 EPU 直

接冲击的时间和程度存在差异，相互之间的风险传

染效应使得系统性金融风险水平进一步攀升；EPU

对经济金融市场投资者情绪的显著冲击更使得经

济金融市场受到直接和间接的双重影响，从而加剧

了市场及投资者情绪的不稳定性。本研究对于应对

EPU 和投资者情绪冲击、防范化解系统性金融风

险具有重要意义。

Joint work with Guobin Fang.

Poster 025: Assessing and Tracking Chinese-Style

Modernization

中国式现代化发展水平及演变测度

Suting Zhang

Anhui University of Finance and Economics

231

摘要：为科学建构中国式现代化指标体系和深入剖

析其时空演变特征，基于 2011-2021 年 30 个省份面

板数据，从经济建设现代化、社会建设现代化、文

化建设现代化及生态建设现代化四个基本维度建

立合理有效的中国式现代化指标体系，运用熵权法

对其进行测度，并采取描述性统计、泰尔指数、莫

兰指数等方法进行实证分析。研究发现：中国式现

代化水平整体呈波动上升态势，但总体水平仍然较

低，且四大区域之间存在显著的异质性，东部地区

远高于其他地区，省际存在较大差异，多数省份现

代化水平较低；总体差距逐渐减小，差距来源于区

域间和区域内，区域间差异是其主要根源，且东部

区域的贡献最大；空间分布上呈非均衡性，经济建

设现代化、社会建设现代化和文化建设现代化总体

呈现东高西低的空间格局，生态建设现代化呈现随

机性分布态势；各省间中国式现代化水平存在显著

的空间正相关性，同时也表现出较为明显的空间集

聚格局。研究丰富和完善了中国式现代化指标体系

的建构与测度分析，为加快推动现代化发展政策制

定提供理论支撑。

Poster 026: Efficient and Flexible LLMs- Generated Content Detection

Wan Tian

Beihang University

Abstract: We propose an efficient content detection

method generated by large language models (LLMs),

referred to as MDCSD, that fully harnesses the universality of LLMs across multiple knowledge domains.

Its essence lies in utilizing the Mahalanobis distancebased confidence score (MDCS) of the text as the

ultimate discriminative feature. The prerequisite for

computing the MDCS is the efficient joint estimation

of the precision matrices corresponding to multiple

knowledge domains. We provide regulariza- tion

techniques for constructing the joint estimator and

offer corresponding computationally feasible optimization algorithms. The sparsity and consisten- cy of

the joint estimator are demonstrated. Based on offset

Rademacher complexity, we further derive the optimal

excess risk bound for the detection problem. Superiority of MDCSD is validated on the recently released

Human ChatGPT Comparison Corpus (HC3) dataset.

Joint work with Yijie Peng and Zhongfeng Qin.

Poster 027: Latent Space Model with Interactions

between Nodal Features and Latent Variables

Ruixuan Qin

Xiamen University

Abstract: Latent space models with assistance from

nodal features have garnered significant attention in

recent years. However, existing research still exhibits

limitations. In this paper, we propose a latent space

model that incorporates the significance of nodal features, as well as the interactions among nodal features

and between nodal features and latent variables (LSI).

We establish sufficient conditions for the identifiability of the model, and devise a projected gradient descent algorithm along with a tailored initialization

method to estimate its parameters. Rigorous theoretical properties of the estimates are then established.

Furthermore, to take account of the sparsity of interactions, we introduce a sparse latent space model

considering interactions (SLSI). Significant features

and interactions are selected through penalization. The

corresponding algorithm for the sparse model is also

developed and its theoretical properties are established.

The utility of the proposed methods is demonstrated

through numerical studies involving extensive simulations and the analysis of three real datasets.

Joint work with Kuangnan Fang and Xinyan Fan.

232

233