Skip to Login Skip to Content Skip to Secondary Navigation


Nazmus Sakib
Sep 30, 2016

Dr. Enrique del Castillo from Penn State visits USF

Sep 30, 2016 by Nazmus Sakib

Del and Beth Kimbler Lecture Series:
Dr. Enrique del Castillo from Penn State

Topic: High dimensional statistical inference in non-parametric models with applications

Statistical methods for large data-sets are the norm today in applications in science and engineering even for conducting the most fundamental types of inferences. For instance, a considerable body of literature exists for testing very large number of hypotheses in parallel given this is a problem that occurs frequently in the sciences. Also common in applications, but studied relatively much less, are confidence region methods in high dimensions. These regions are necessary when fitting non-parametric response models such as Splines, which actually are functions of a large number of parameters. In this talk, present three diverse applications of high dimensional confidence regions for non parametric functions in engineering and science are presented, namely 1) applications in pharmaceutical drug development, 2) applications in internet-based video streaming optimization, and 3) applications in nutrition experiments in evolutionary biology. All of these problems are instances of a common theme: finding a confidence region for the optimum point of a flexible non-parametric function (a Thin Plate Spline), usually fitted in a mixture-amount experimental space, while dealing with a high dimensional parameter space. New methodology based on the notion of data-depth for finding the desired regions with guaranteed coverage properties is presented. Towards the end of the talk a brief discussion of on-going work on the design and analysis of experiments where the controllable factors and responses are located on nodes over a network or graph on aforementioned application number 2 was held by introducing the concept of an experimental design on a graph.