Seminars

Estimating subgroups with spatial information under repeated measures

Thursday, September 21, 2017 - 2:10pm to 3:00pm
Event Type: 

Abstract: Spatial clustering or spatial boundaries detection is an important problem in disease mapping, spatial epidemiology, population genetics. Clustering problem can be solved through a convex optimization problem. We extend the regression problem to spatial areal data with repeated measures. The objective function has two parts, the first part is weighted least squares which use sample size inverse as the weight. The second part is pairwise penalties, which have pairwise weights associated with each pairwise penalty based on spatial information. Read more about Estimating subgroups with spatial information under repeated measures

Semiparametric Covariance estimation under repeated measures

Thursday, September 28, 2017 - 2:10pm to 3:00pm
Event Type: 

Abstract: Covariance estimation is very important issue in longitudinal study and spatial analysis. The estimation procedure should persist the positive semi-definite.  However, most existing methods obtain the positive semi-definite by truncation after the estimation procedure. In spatial analysis, the kriging is feasible only for low rank covariance matrix. Read more about Semiparametric Covariance estimation under repeated measures

Uncertainty Quantification for NASA Earth Science Applications

Friday, September 29, 2017 - 11:00am to 12:00pm
Event Type: 

This presentation will provide an overview of methods for uncertainty quantification in the use of a retrieval algorithm to infer a physical quantity of interest from the satellite’s observed intensity of radiation and in propagating input uncertainty through a high-resolution process model. The techniques are implemented for applications in carbon cycle science and terrestrial hydrology. Read more about Uncertainty Quantification for NASA Earth Science Applications

Panel attrition in National Survey of Tax and Benefit

Thursday, October 5, 2017 - 2:10pm to 3:00pm
Event Type: 

Abstract: This study is concerned with the determination of panel attrition from the National Survey of Tax and Benefit. Panel attrition is one of key concerns for data quality and it may bring serious problems such as panel attrition bias. Even an initial panel with a good representativeness of the population of interest can become unrepresentative as small attrition rate accumulates. We present how the panel attrition can be detected and introduce the weigh adjustment method that enables the panel to retain the representativeness. Read more about Panel attrition in National Survey of Tax and Benefit

Efficient gap-fill algorithm for satellite images

Thursday, October 12, 2017 - 2:10pm to 3:00pm
Event Type: 

Many scientific applications and signal processing algorithms require complete satellite data, while often times the quality of the remote sensed satellite data is not good due to variety of reasons, such as cloud cover, shadows or sensor failure. The satellite data is a typical spatio-temporal data, which has repeated measures over time and spatial correlations among nearby locations. Read more about Efficient gap-fill algorithm for satellite images

An Introduction to Generative Adversarial Nets and Application to NRI image data

Thursday, October 19, 2017 - 2:10pm to 3:00pm
Event Type: 

In National Resource Inventory (NRI), yearly satellite image data are provided for each

states. The goal of our project is to detect whether there are road area changes according to the

satellite images. We study the Generative Adversarial Network (GANs), a recently popular

model in deep learning. In our presentation, we will introduce the framework of GANs and one

of its derivatives, Conditional Generative Adversarial Network (CGANs). We’ll also show some

preliminary results of GANs on our image data. Read more about An Introduction to Generative Adversarial Nets and Application to NRI image data

Predicting rainfall-erosion losses and visualizing covariates from cropland in South Dakota

Thursday, October 26, 2017 - 2:10pm to 3:00pm
Event Type: 

Abstract: In this talk, I’ll talk about how we overlaid soil map shapefile and CDL raster in R to obtain a list of cropland map units for South Dakota, which is our target population frame for small area prediction. The predicted population mean of rainfall-erosion losses from cropland at county level for South Dakota will be presented. A Shiny app visualizing the procedure and the covariates will be demonstrated. Read more about Predicting rainfall-erosion losses and visualizing covariates from cropland in South Dakota

Efforts to Quantify the Causal Effect of Fine Particulate Matter on Mortality

Thursday, November 2, 2017 - 2:10pm to 3:00pm
Event Type: 

Abstract: Fine particulate matter (often referred to as PM2.5) is one pollutant incur a major environmental health problem affecting many countries. Sequential studies have been done indicating that higher PM2.5 exposure levels are associated with increase in mortality. However, it remains a challenging task to quantify the causal effect of PM2.5 on mortality. One difficulty is that PM2.5 exposure level changes over time and it is confounded by meteorological variables. Read more about Efforts to Quantify the Causal Effect of Fine Particulate Matter on Mortality

Bootstrap variance estimation for one-per-stratum spatial sampling design

Thursday, November 9, 2017 - 2:10pm to 3:00pm
Event Type: 

Abstract: In areal sampling, one-per-stratum design is a common approach which can achieve spatial balance and improve the precision of the resulting estimators. The downside of such design is that it is more challenging to have a good design-unbiased variance estimation. In this paper, we propose a general class of stratified sampling design which produces spatially balanced sample. Read more about Bootstrap variance estimation for one-per-stratum spatial sampling design

A Review of Structural Breaks

Thursday, November 30, 2017 - 2:10pm to 3:00pm
Event Type: 

Abstract: The analysis of structural breaks, or change-points, origins in 1950s, has become an integral part of variety fields, like statistics, econometrics, finance, engineering and climatology. The amount of work on this topic is especially voluminous in both statistics and econometrics, and there are several perspectives to view and investigate structural beaks. In this talk, I will introduce three branches: 1. classical approach to identification breakpoints via testing, estimation the break dates and parameters in each regime; 2. Read more about A Review of Structural Breaks

Bootstrap inference for the finite population total under complex sampling designs

Tuesday, January 16, 2018 - 3:30pm to 4:30pm
Event Type: 

Speaker: Zhonglei Wang / Abstract: Bootstrap is a useful tool for making statistical inference, but it may provide erroneous results in survey sampling if the sampling design is ignored. Most studies about bootstrap-based inference are developed under simple random sampling and stratified random sampling. In this work, we propose a new bootstrap method applicable to some complex sampling designs, including Poisson sampling and probability proportional to size sampling. Read more about Bootstrap inference for the finite population total under complex sampling designs

Data Integration for Big Data Analysis

Tuesday, January 23, 2018 - 3:30pm to 4:30pm
Event Type: 

Speaker: Jae-kwang Kim / Abstract: In analyzing big data for finite population inference, it is critical to adjust for the selection bias in the big data. Using an independent probability sampling, the selection bias of big data can be removed safely. Such technique can be called data integration for big data analysis. In this talk, several methods of data integration are presented in the context of big data analysis. Results from simulation studies are also presented. Read more about Data Integration for Big Data Analysis

Inverse Conditional Probability Weighting with Clustered Data in Causal Inference

Tuesday, January 30, 2018 - 3:30pm to 4:30pm
Event Type: 

Speaker: Zhulin He / Abstract: Estimating the average treatment causal effect in clustered data often involves dealing with unmeasured cluster-specific confounding variable. The unmeasured cluster-specific confounding variable may be correlated with the measured covariates and the outcome. When such correlations are ignored, the causal effect estimation can be biased. Read more about Inverse Conditional Probability Weighting with Clustered Data in Causal Inference

Semiparametric Estimation and Testing for Nonignorable Nonresponse

Tuesday, February 6, 2018 - 3:30pm to 4:30pm
Event Type: 

Speaker: Hejian Sang / Abstract: Statistical inference with nonresponse is quite challenging, especially when the response mechanism is not missing at random. The existing methods often require correct model specifications for both the outcome regression model and the response model. However, due to nonresponse, both model assumptions cannot be verified from the data and model misspecification can seriously lead to biased inferences. To overcome this limitation, we develop a robust semiparametric method based on the profile likelihood. Read more about Semiparametric Estimation and Testing for Nonignorable Nonresponse

Asynchronous stochastic gradient descent with unbounded delay on nonconvex problem

Tuesday, February 13, 2018 - 3:30pm to 4:30pm
Event Type: 

Speaker: Xin Zhang / Abstract:  Parallel distributed implementations of stochastic gradient, especially asynchronous scheme, have been widely applied in optimizing nonconvex problems. So far, several works have been done to analyze the asymptotic convergence rate of asynchronous gradient descent method. But these works have been restricted by bounded delay. In our work, we focus on the asynchronous stochastic gradient descent methods with unbounded delay on nonconvex optimization problem. Read more about Asynchronous stochastic gradient descent with unbounded delay on nonconvex problem

Mass Imputation for Two-phase Sampling

Tuesday, February 20, 2018 - 3:30pm to 4:30pm
Event Type: 

Speaker: Seho Park / Abstract: Two-phase sampling is a convenient and economical sampling design that sample selection is conducted in two phases. A large sample is collected from population measuring auxiliary variables and then a smaller sample is collected measuring study variables by incorporating the auxiliary information. The structure of two-phase sample can be seen as a missing data problem: some are observed and the others are missing. Read more about Mass Imputation for Two-phase Sampling