Scalable Inference for Rare Events: Computational Methods for Estimating Probability of Tail Events

Abstract

This report presents key findings and results in the Scalable Inference for Rare Events (SIRE) project (FA8650-16-C-7646). Under the project, we discovered deep theoretical connections between Koopman operator theory and rare event simulation in stochastic differential equations. We then developed a generalized approach for constructing efficient importance sampling methods for linear stochastic differential equations using the Kolmogorov Backward (Ornstein-Uhlenbeck) operator. We show that this approach is a special case of the Koopman operator approach. Additionally, we constructed rotorcraft models that capture critical stall phenomena that was used for computation. We then demonstrate large deviations-based importance sampling and splitting methods on rotorcraft and electrical models.

Open PDF

Document Details

Document Type
Technical Report
Publication Date
Aug 15, 2019
Accession Number
AD1090887

Entities

People

  • Quan Long
  • Tuhin Sahai
  • Yibin B. Zhang
  • Youssef Marzouk

Organizations

  • United Technologies Research Center

Tags

Communities of Interest

  • Air Platforms
  • Energy and Power Technologies
  • Space

DTIC Thesaurus Topics

  • Air Force
  • Aircrafts
  • Computational Fluid Dynamics
  • Computational Science
  • Computations
  • Differential Equations
  • Equations
  • Fluid Dynamics
  • Fokker Planck Equations
  • Information Science
  • Kolmogorov Equations
  • Monte Carlo Method
  • Partial Differential Equations
  • Probability
  • Random Variables
  • Sampling
  • Stochastic Processes

Readers

  • Distributed Systems and Data Platform Development
  • Mathematical Modeling and Probability Theory.

Technology Areas

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