Stein Variational Inference for Uncertainty Quantification: Breaking the Computational, Data, and Human Bottlenecks

Abstract

Uncertain quantification (UQ) is a central component of countless computational and real-world applications. However, with the unprecedented surge of scale and complexity of modern applications, UQ in modern systems is largely hindered by three major bottlenecks:1) the computational bottleneck for developing fast and scalable algorithms for reasoning with computationally intensive systems; 2) the data bottleneck due to the high cost of collecting informative measurement in critical areas; and 3) the human bottleneck due to the challenge of conveying the results of increasingly sophisticated algorithms to human decision makers in a reliable and transparent way. This project will tackle the key challenges in the computational, data and human bottlenecks by developing and enriching a fundamentally new paradigm for uncertainty quantification and approximate inference called Stein variational inference. This new framework will yield a spectrum of novel algorithms with both practical efficiency and attractive theoretical properties, offering a powerful approach for tackling the key challenges in UQ, including fast computation, optimal data acquisition, and human interpretability.

Document Details

Document Type
DoD Grant Award
Publication Date
Aug 05, 2021
Source ID
N000142112665

Entities

People

  • Qiang Liu

Organizations

  • Office of Naval Research
  • United States Navy
  • University of Texas at Austin

Tags

Fields of Study

  • Computer science

Readers

  • Adaptive Control and Estimation with Uncertainty in Dynamic Systems.
  • Neural Network Machine Learning.
  • Systems Analysis and Design

Technology Areas

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