2023 Seminars
Seminar Announcement
Qing Guo, Ph.D.
Virginia Tech, Department of Statistics
"Advancing Data Acquisition: A Novel Contrastive Mutual Information Estimator for
Bayesian Optimal Experimental Design"
Dec. 1, 2023 at 1:30 p.m. to 2:30 p.m.
E0134 HRKE
The efficiency of data collection is crucial in many areas, including agriculture, engineering, and intelligent conversational systems. In this talk, Guo will present her recent work on optimizing the data collection strategy by developing advanced machine-learning techniques. The proposed approach centers around leveraging the power of deep neural networks and maximizing the information gained from data within the framework of Bayesian optimal experimental design (BOED). To measure the information gain, she will introduce an innovative contrastive mutual information (MI) estimator to serve as an information-rich criterion under the BOED framework. This new MI estimator addresses the drawbacks of existing estimators by eliminating the need for explicit probabilistic descriptions of the model or likelihood functions. The performance of the proposed method is evaluated by both numerical examples and real applications.
Seminar Announcement
Andrew G. Chapple, Ph.D.
Biostatistics Core Director
Stanley S. Scott Cancer Center & LSUHSC School of Medicine
Nov. 3, 2023, 1:00 - 1:50 p.m.
E0134 HRKE
A multi-armed trial based on ordinal outcomes is proposed that leverages a flexible non-proportional odds cumulative logit model and numerical utility scores for each outcome to determine treatment optimality. This trial design uses a Bayesian clustering prior on the treatment effects that encourages the pairwise null hypothesis of no differences between treatments. A group sequential design is proposed to determine which treatments are clinically different with an adaptive decision boundary that becomes more aggressive as the sample size or clinical significance grows, or the number of active treatments decreases. A simulation study is conducted for three and five treatment arms, which shows that the design has superior operating characteristics (family wise error rate, generalized power, average sample size) compared to utility designs that do not allow clustering, a frequentist proportional odds model, or a permutation test based on empirical mean utilities.
Seminar Announcement
JMP Pro Statistical Software for Teaching and Research
Sept. 15, 2023
JMP Pro statistics software combines comprehensive statistical capabilities with an interactive, no-code interface. Its ease-of-use makes it a strong teaching tool, and its powerful analysis capabilities have led to adoption across academia and industry. All LSU faculty and students have access to JMP Pro (download on tigerware.lsu.edu).
On Sept. 15, the JMP Academic team will deliver two seminars on JMP Pro for basic-to-advanced data visualization, statistical modeling, and machine learning.