Special Lectures-浙江大学数据科学研究中心

English

搜索

当前位置：首页

学术交流

第十一届泛华统计协会大会

Special Lectures

Peter Hall Lecture

Jianqing Fan, Princeton University

Title: Communication-Efficient Accurate Statistical Estimation

Abstract:

When the data are stored in a distributed manner, direct application of traditional statistical inference procedures is often prohibitive due to communication cost and privacy concerns. This paper develops and investigates two Communication-Efficient Accurate Statistical Estimators (CEASE), implemented through iterative algorithms for distributed optimization. In each iteration, node machines carry out computation in parallel and communicates with the central processor, which then broadcasts aggregated gradient vector to node machines for new updates. The algorithms adapt to the similarity among loss functions on node machines, and converge rapidly when each node machine has large enough sample size. Moreover, they do not require good initialization and enjoy linear converge guarantees under general conditions. The contraction rate of optimization errors is derived explicitly, with dependence on the local sample size unveiled. In addition, the improved statistical accuracy per iteration is derived. By regarding the proposed method as a multi-step statistical estimator, we show that statistical efficiency can be achieved in finite steps in typical statistical applications. In addition, we give the conditions under which one-step CEASE estimator is statistically efficient. Extensive numerical experiments on both synthetic and real data validate the theoretical results and demonstrate the superior performance of our algorithms.

(Joint work with Yongyi Guo and Kaizheng Wang)

Pao-Lu Hsu Award Lecture

Hongyu Zhao, Yale School of Public Health

Title: Fisher’s 1918 Quantitative Genetics Model In the Genomics Era

Abstract:

In 1918, R. A. Fisher reported a comprehensive study of a statistical model relating an individual’s quantitative traits to his/her genetic factors in his seminal paper entitled “The Correlation between Relatives on the Supposition of Mendelian Inheritance”. This model laid the foundation for the field of quantitative genetics. More than a century later, this model still proves effective in understanding the genetic basis of human complex traits when tens of thousands of chromosomal regions have been implicated for hundreds of traits through Genome-Wide Association Studies (GWAS) in the past 15 years. In this presentation, I will discuss how Fisher’s model has been used to quantify the genetic contributions to complex traits using GWAS results, its robustness to model misspecifications, and its extensions to identify relevant tissues/cell types for a specific trait and genetic correlations between different traits. I will also discuss statistical inference using either individual genotype and phenotype data, a typical set up for traditional statistical analysis, or summary statistics, which are more easily accessible for GWAS data. This is joint work with Can Yang, Jiming Jiang, Qiongshi Lu, Debashis Paul, Wei Jiang, Cecilia Dao, Yiliang Zhang, and others.

Keynote Lecture

Zhiliang Ying, Columbia University

Title: Statistical models and methods for educational and psychological measurement

Abstract:

Statistical models have played important and fundamental roles in educational and psychological measurement. The increased computing power and data collection capability provide new opportunities as well as challenges. The first part of this talk covers the classical item response theory models which have been widely used in standardized testing, as well as recent developments on related multidimensional latent factor/class models with focus on important issues such as local independence or lack of it, identifiability among others. The second part covers modeling and analysis of process data arise from modern computer-based tests with items for assessing complex problem solving skills in technology-rich environments. Examples from educational assessment and psychological evaluation will be used throughout the presentation.