Experimental Design for Data Science and Engineering

Overview & Synopsis

Theory, experiments, computation, and data are considered as the four pillars of science and engineering. Experimental Design for Data Science and Engineering describes efficient statistical methods for making the experiments cheaper and computations faster for extracting valuable information from data and help identify discrepancies in the theory. The book also includes recent advances in experimental designs for dealing with large amounts of observational data. Traditionally the design and analysis of physical and computer experiments are treated differently, but this book attempts to create a unified framework using Gaussian process models. Although optimal designs are formulated using Gaussian process models, the focus is on obtaining practical experimental designs that are robust to model assumptions. A wide variety of topics are covered in the book -- from designs for interpolating or integrating simple functions to designs that are useful for optimizing and calibrating complex computer models. It draws techniques that are spread across the fields of statistics, applied mathematics, operations research, uncertainty quantification, and information theory, and build experimental design as a fundamental data analytic tool for engineering and scientific discoveries. Designs for both computer and physical experiments are discussed in a unified framework. Integrates several concepts from numerical analysis, Monte Carlo methods, sensitivity analysis, optimization, and machine learning with experimental design techniques in statistics. Methods are explained using many real experiments from physical sciences and engineering. Experimental design techniques for analysis and compression of big data are discussed. All the numerical illustrations in the book are reproducible using R and Python codes provided in the author’s GitHub site.