Spectral density estimation of stochastic processes under missing data and uncertainty quantification with Bayesian deep learning

Yu Chen, Edoardo Patelli, Benjamin Edwards, Michael Beer

Research output: Chapter in Book/Report/Conference proceedingConference contribution book

39 Downloads (Pure)

Abstract

Stochastic processes are widely adopted in many domains to deal with problems which are stochastic in nature and involve strong nonlinearity, nonstationarity and uncertain system parameters. However, the uncertainties of spectral representation of the underlying stochastic processes have not been adequately acknowledged due to the data problems in practice, for instance, missing data. Therefore, this paper proposes a novel method for uncertainty quantification of spectral representation in the presence of missing data using Bayesian deep learning models. A range of missing levels are tested. An example in stochastic dynamics is employed for illustration.
Original languageEnglish
Title of host publicationUNCECOMP 2023: 5th International Conference on Uncertainty Quantification in Computational Science and Engineering
Subtitle of host publicationProceedings
EditorsM. Papadrakakis, V. Papadopoulos, G. Stefanou
Place of PublicationAthens
Pages745-754
Number of pages10
Publication statusPublished - 24 Oct 2023
Event5th International Conference on Uncertainty Quantification in Computational Science and Engineering - Athens, Greece
Duration: 12 Jun 202314 Jun 2023
https://2023.uncecomp.org/

Conference

Conference5th International Conference on Uncertainty Quantification in Computational Science and Engineering
Abbreviated titleUNCECOMP 2023
Country/TerritoryGreece
CityAthens
Period12/06/2314/06/23
Internet address

Keywords

  • missing data
  • uncertainty quantification
  • LSTM
  • variational inference (VI)

Fingerprint

Dive into the research topics of 'Spectral density estimation of stochastic processes under missing data and uncertainty quantification with Bayesian deep learning'. Together they form a unique fingerprint.

Cite this