Efficient importance sampling for stochastic simulators using adaptive mixture density networks

Han Peng

doi:10.1016/j.strusafe.2026.102709

Efficient importance sampling for stochastic simulators using adaptive mixture density networks

2026-03-16

0

PRE

AI

H

Han Peng

J

Jize Zhang *

DOI：10.1016/j.strusafe.2026.102709

Original

Original request for help

Save

Abstract

En 中文

• Novel MDN-based importance sampling for stochastic simulator reliability analysis. • CRPS loss training optimizes conditional failure probability estimation accuracy. • Ensemble strategy with pruning reduces surrogate prediction variability. • Substantial fewer samples than IS baselines for equivalent estimation accuracy. • Effective for 12-dim structural reliability under stochastic seismic excitation.

Keywords:

Importance Sampling

Mixture Density Networks

Stochastic Simulators

Reliability Analysis

Conditional Failure Probability

Efficient importance sampling for stochastic simulators using adaptive mixture density networks

Abstract

Journal

Organization

Cited Papers

Citing Papers