skip to main content
OSTI.GOV title logo U.S. Department of Energy
Office of Scientific and Technical Information

Title: Active Subspace Methods for Data-Intensive Inverse Problems

Abstract

The project has developed theory and computational tools to exploit active subspaces to reduce the dimension in statistical calibration problems. This dimension reduction enables MCMC methods to calibrate otherwise intractable models. The same theoretical and computational tools can also reduce the measurement dimension for calibration problems that use large stores of data.

Authors:
 [1]
  1. Massachusetts Inst. of Technology (MIT), Cambridge, MA (United States)
Publication Date:
Research Org.:
Massachusetts Inst. of Technology (MIT), Cambridge, MA (United States)
Sponsoring Org.:
USDOE Office of Science (SC), Advanced Scientific Computing Research (ASCR) (SC-21)
OSTI Identifier:
1353429
Report Number(s):
DE-SC0011089
DOE Contract Number:
SC0011089
Resource Type:
Technical Report
Country of Publication:
United States
Language:
English
Subject:
97 MATHEMATICS AND COMPUTING

Citation Formats

Wang, Qiqi. Active Subspace Methods for Data-Intensive Inverse Problems. United States: N. p., 2017. Web. doi:10.2172/1353429.
Wang, Qiqi. Active Subspace Methods for Data-Intensive Inverse Problems. United States. doi:10.2172/1353429.
Wang, Qiqi. Thu . "Active Subspace Methods for Data-Intensive Inverse Problems". United States. doi:10.2172/1353429. https://www.osti.gov/servlets/purl/1353429.
@article{osti_1353429,
title = {Active Subspace Methods for Data-Intensive Inverse Problems},
author = {Wang, Qiqi},
abstractNote = {The project has developed theory and computational tools to exploit active subspaces to reduce the dimension in statistical calibration problems. This dimension reduction enables MCMC methods to calibrate otherwise intractable models. The same theoretical and computational tools can also reduce the measurement dimension for calibration problems that use large stores of data.},
doi = {10.2172/1353429},
journal = {},
number = ,
volume = ,
place = {United States},
year = {Thu Apr 27 00:00:00 EDT 2017},
month = {Thu Apr 27 00:00:00 EDT 2017}
}

Technical Report:

Save / Share:
  • The goal of the Active System Area Networks (ASAN) project is to develop hardware and software technologies for the implementation of active system area networks (ASANs). The use of the term ''active'' refers to the ability of the network interfaces to perform application-specific as well as system level computations in addition to their traditional role of data transfer. This project adopts the view that the network infrastructure should be an active computational entity capable of supporting certain classes of computations that would otherwise be performed on the host CPUs. The result is a unique network-wide programming model where computations aremore » dynamically placed within the host CPUs or the NIs depending upon the quality of service demands and network/CPU resource availability. The projects seeks to demonstrate that such an approach is a better match for data intensive network-based applications and that the advent of low-cost powerful embedded processors and configurable hardware makes such an approach economically viable and desirable.« less
  • The research emphasis under this grant's funding is in the area of algebraic multigrid methods. The research has two main branches: 1) exploring interdisciplinary applications in which algebraic multigrid can make an impact and 2) extending the scope of algebraic multigrid methods with algorithmic improvements that are based in strong analysis.The work in interdisciplinary applications falls primarily in the field of biomedical imaging. Work under this grant demonstrated the effectiveness and robustness of multigrid for solving linear systems that result from highly heterogeneous finite element method models of the human head. The results in this work also give promise tomore » medical advances possible with software that may be developed. Research to extend the scope of algebraic multigrid has been focused in several areas. In collaboration with researchers at the University of Colorado, Lawrence Livermore National Laboratory, and Los Alamos National Laboratory, the PI developed an adaptive multigrid with subcycling via complementary grids. This method has very cheap computing costs per iterate and is showing promise as a preconditioner for conjugate gradient. Recent work with Los Alamos National Laboratory concentrates on developing algorithms that take advantage of the recent advances in adaptive multigrid research. The results of the various efforts in this research could ultimately have direct use and impact to researchers for a wide variety of applications, including, astrophysics, neuroscience, contaminant transport in porous media, bi-domain heart modeling, modeling of tumor growth, and flow in heterogeneous porous media. This work has already led to basic advances in computational mathematics and numerical linear algebra and will continue to do so into the future.« less
  • Inverse radiation transport focuses on identifying the configuration of an unknown radiation source given its observed radiation signatures. The inverse problem is traditionally solved by finding the set of transport model parameter values that minimizes a weighted sum of the squared differences by channel between the observed signature and the signature pre dicted by the hypothesized model parameters. The weights are inversely proportional to the sum of the variances of the measurement and model errors at a given channel. The traditional implicit (often inaccurate) assumption is that the errors (differences between the modeled and observed radiation signatures) are independent acrossmore » channels. Here, an alternative method that accounts for correlated errors between channels is described and illustrated using an inverse problem based on the combination of gam ma and neutron multiplicity counting measurements.« less
  • A mathematical approach is developed for empirically characterizing a given source region using waveforms from a collection of calibration events. A region is considered to be adequately characterized if the waveforms from any event in the source region can be represented as a linear combination of calibration event waveforms. The purpose of such characterizations is to build waveform recognizers'' for specific regions for precision location applications, and to provide a means of separating superimposed waveforms from multiple events in different source regions. The particular form of characterization used is insensitive to variations in the source time function and to anythingmore » but changes from the normal range of source mechanisms encountered in the source region. The standard waveform correlation coefficient used to estimate event clustering is generalized to estimate separation between single events and event clusters, and between two clusters of events. The generalized correlation coefficient is insensitive to variations in source time function and, to some extent, mechanism. The statistics of waveform correlation coefficients are developed, and show that conventional estimates made from single station data are often developed for network or array data removes the ambiguity. 23 refs., 4 figs.« less
  • This paper seeks to evaluate the accuracy of the regression technique used in 2006 by using the same least-squares regression analysis on data created from hourly energy simulation software.