Skip to Main Content
 

Global Search Box

 
 
 
 

Files

File List


Dissertation_formatted - final.pdf (2.11 MB)

ETD Abstract Container

Abstract Header

A Robust Adaptive Autonomous Approach to Optimal Experimental Design

GU, Hairong
http://rave.ohiolink.edu/etdc/view?acc_num=osu1471590586

Abstract Details

2016, Doctor of Philosophy, Ohio State University, Psychology.
Experimentation is the fundamental tool of scientific inquiries to understand the laws governing the nature and human behaviors. Many complex real-world experimental scenarios, particularly in quest of prediction accuracy, often encounter difficulties to conduct experiments using an existing experimental procedure for the following two reasons. First, the existing experimental procedures require a parametric model to serve as the proxy of the latent data structure or data-generating mechanism at the beginning of an experiment. However, for those experimental scenarios of concern, a sound model is often unavailable before an experiment. Second, those experimental scenarios usually contain a large number of design variables, which potentially leads to a lengthy and costly data collection cycle. Incompetently, the existing experimental procedures are unable to optimize large-scale experiments so as to minimize the experimental length and cost. Facing the two challenges in those experimental scenarios, the aim of the present study is to develop a new experimental procedure that allows an experiment to be conducted without the assumption of a parametric model while still achieving satisfactory prediction, and performs optimization of experimental designs to improve the efficiency of an experiment. The new experimental procedure developed in the present study is named robust adaptive autonomous system (RAAS). RAAS is a procedure for sequential experiments composed of multiple experimental trials, which performs function estimation, variable selection, reverse prediction and design optimization on each trial. Directly addressing the challenges in those experimental scenarios of concern, function estimation and variable selection are performed by data-driven modeling methods to generate a predictive model from data collected during the course of an experiment, thus exempting the requirement of a parametric model at the beginning of an experiment; design optimization is performed to select experimental designs on the fly of an experiment based on their usefulness so that fewest designs are needed to reach useful inferential conclusions. Technically, function estimation is realized by Bayesian P-splines, variable selection is realized by Bayesian spike-and-slab prior, reverse prediction is realized by grid-search and design optimization is realized by the concepts of active learning. The present study demonstrated that RAAS achieves statistical robustness by making accurate predictions without the assumption of a parametric model serving as the proxy of latent data structure while the existing procedures can draw poor statistical inferences if a misspecified model is assumed; RAAS also achieves inferential efficiency by taking fewer designs to acquire useful statistical inferences than non-optimal procedures. Thus, RAAS is expected to be a principled solution to real-world experimental scenarios pursuing robust prediction and efficient experimentation.
Jay Myung (Advisor)
Mark Pitt (Committee Member)
Paul de Boeck (Committee Member)
Trisha Van Zandt (Committee Member)
178 p.
Experimental Psychology; Experiments; Quantitative Psychology
experimentation; adaptive design optimization; nonparametric; variable selection; reverse prediction; robustness; efficiency

Recommended Citations

Citations

  • GU, H. (2016). A Robust Adaptive Autonomous Approach to Optimal Experimental Design [Doctoral dissertation, Ohio State University]. OhioLINK Electronic Theses and Dissertations Center. http://rave.ohiolink.edu/etdc/view?acc_num=osu1471590586

    APA Style (7th edition)

  • GU, Hairong. A Robust Adaptive Autonomous Approach to Optimal Experimental Design. 2016. Ohio State University, Doctoral dissertation. OhioLINK Electronic Theses and Dissertations Center, http://rave.ohiolink.edu/etdc/view?acc_num=osu1471590586.

    MLA Style (8th edition)

  • GU, Hairong. "A Robust Adaptive Autonomous Approach to Optimal Experimental Design." Doctoral dissertation, Ohio State University, 2016. http://rave.ohiolink.edu/etdc/view?acc_num=osu1471590586

    Chicago Manual of Style (17th edition)

osu1471590586
552
© 2016, all rights reserved.
This open access ETD is published by The Ohio State University and OhioLINK.