Model-Based Optimal Experimental Design for Complex Physical Systems

Abstract

Experimental data play an essential role in developing and refining models of physical systems. Yet experimental observations can be difficult, time-consuming, and expensive to acquire. In this context, maximizing the value of experimental observations-e.g., choosing when and where to measure, and what experimental conditions to employ-is a critical task. This project addressed open challenges in optimal experimental design (OED) for complex physical systems, taking a Bayesian decision theoretic approach. Our focus has been on general nonlinear systems and information theoretic design objectives, for which existing theory and computational tools have been inadequate. Our goal has been to develop new mathematical formulations, estimation approaches, and approximation strategies to make rigorous OED feasible for systems accessible only through computational simulation. Key results include: (1) innovations in batch optimal experimental design, in particular a new multiple importance sampling scheme that improves the efficiency of expected information gain estimation by several orders of magnitude; and (2) new dynamic programming formulations and methods for sequential optimal experimental design, supplanting previous suboptimal approaches.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Dec 03, 2015
Accession Number
ADA627240

Entities

People

  • Chi Feng
  • Xun Huan
  • Youssef Marzouk

Organizations

  • Massachusetts Institute of Technology

Tags

Communities of Interest

  • Air Platforms
  • Human Systems
  • Space

DTIC Thesaurus Topics

  • Air Force
  • Air Force Research Laboratories
  • Applied Mathematics
  • Bayesian Inference
  • Computational Mechanics
  • Computational Science
  • Computer Programming
  • Data Science
  • Dynamic Programming
  • Estimators
  • Experimental Design
  • Information Science
  • Mathematics
  • Nonlinear Systems
  • Random Variables
  • Simulations
  • Statistics

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Systems Analysis and Design

Technology Areas

  • AI & ML
  • AI & ML - Bayesian Inference
  • AI & ML - Machine Learning Algorithms