Exploiting Task-Based Parallelism in Bayesian Uncertainty Quantification
We introduce a task-parallel framework for non-intrusive Bayesian Uncertainty Quantification and Propagation of complex and computationally demanding physical models on massively parallel computing architectures. The framework incorporates Laplace asymptotic approximations and stochastic algorithms along with distributed numerical differentiation. Sampling is based on the Transitional Markov Chain Monte Carlo algorithm and its variants while the optimization tasks associated with the asymptotic approximations are treated via the Covariance Matrix Adaptation Evolution Strategy. Exploitation of task-based parallelism is based on a platform-agnostic adaptive load balancing library that orchestrates scheduling of multiple physical model evaluations on computing platforms that range from multicore systems to hybrid GPU clusters. Experimental results using representative applications demonstrate the flexibility and excellent scalability of the proposed framework.
KeywordsTask-based parallelism Bayesian uncertainty quantification
- 5.Wu, S., Beck, J.L., Heaton, T.H.: Earthquake probability-based automated decision-making framework for earthquake early warning applications. Comp. Aid. Civ. Infr. Eng. 28, 737–752 (2013)Google Scholar
- 6.Adams, B., Bohnhoff, W., Dalbey, K., Eddy, J., Eldred, M., Gay, D., Haskell, K., Hough, P., Swiler, L.: DAKOTA, a multilevel parallel object-oriented framework for design optimization, parameter estimation, uncertainty quantification, and sensitivity analysis. Sandia Technical report (2013)Google Scholar
- 7.Lawrence Livermore National Laboratory. The PSUADE UQ project. http://computation.llnl.gov/casc/uncertainty_quantification/
- 9.Hadjidoukas, P.E., Lappas, E., Dimakopoulos, V.V.: A runtime library for platform-independent task parallelism. In: 20th International Conference on Parallel, Distributed and Network-Based Processing, pp. 229–236 (2012)Google Scholar