Communications of the ACM
The Reality of Problem Solving
Researchers say a new adaptive importance sampling algorithm can help generate more accurate predictions when uncertainty is involved.
Credit: Pacific Northwest National Laboratory
A new adaptive importance sampling algorithm can help generate more accurate predictions than a data assimilation algorithm when uncertainty is involved.
Weixuan Li from the Pacific Northwest National Laboratory and Purdue University professor Guang Lin have proposed the algorithm.
Data assimilation algorithms are known for incurring significant computational resource costs because uncertainty in model parameters can lead to a large number of repetitive model evaluations. However, Li and Lin say test cases for the proposed algorithm show it alleviates this burden.
The researchers demonstrated the algorithm can effectively capture the complex posterior parametric uncertainties for the specific problems being examined while also enhancing computational efficiency.
Li and Lin adapted a Gaussian mixture model and also implemented a mixture of polynomial chaos expansions built as a surrogate model. They say the techniques give the algorithm the flexibility to handle complex multimodal distributions and strongly nonlinear models.
The adaptive importance sampling algorithm already works well for problems involving a small number of uncertain parameters due to limited data or incomplete knowledge.
From Pacific Northwest National Laboratory
View Full Article
Abstracts Copyright © 2015 Information Inc., Bethesda, Maryland, USA
No entries found