第72回データ同化セミナー
日時:2015年4月17日(金) 13:30-14:30
場所:セミナー室4 (D312B)
講演者:Stephen Wu(ETH Zürich Computational Science & Engineering Laboratory)
題目:Living with Uncertainty
要旨:
Uncertainty quantification improves modeling and predictions under
measurement noise, inadequate models and computational errors in various
science and engineering problems. In this presentation, I show examples
from molecular dynamics, pharmacokinetics, seismology and earthquake
engineering. I will discuss model calibration and model selection based
on a Bayesian framework. I will focus on four important issues when
using the Bayesian framework: (1) uncertainty quantification with
heterogeneous data or data with a hierarchical structure, (2) solutions
to the high computational demand in evidence calculations, (3)
uncertainty quantification for real-time predictions, and (4) estimation
of rare-event probability. I will demonstrate the applications of
various Monte Carlo methods, such as, Particle Filter, Transitional
Markov Chain Monte Carlo (similar to Annealed Importance Sampling) and
Subset Simulation (aka Splitting Method).