## Fall 2014 MnDRIVE Seminar Series
## AbstractThe goal of uncertainty quantification (UQ) is to formulate metrics of confidence for computer simulations comparable to confidence metrics for experiments. In practice, computing these UQ metrics involves parameter studies – e.g., numerical integration, numerical optimization, calibration, or response surface construction. But performing these parameter studies become increasingly difficult as the number of input parameters increases, especially when the simulation is expensive. The benefits of dimension reduction cannot be overstated. If one is able to approximate a simulation's prediction with 10 inputs by a comparable interface with 2 inputs, then several otherwise intractable techniques become feasible. I will discuss our research efforts and progress on active subspaces for dimension reduction. The idea is to discover and exploit important linear combinations of the input parameters to reduce the effort for thorough parameter studies in complex simulations. ## BiosketchPaul Constantine is the Ben L. Fryrear Assistant Professor of Applied Math and Statistics at Colorado School of Mines. He received his PhD in 2009 from Stanford's Institute for Computational and Mathematical Engineering and spent two years at Sandia National Laboratories, Albuquerque, as the von Neuman Fellow. His interests include uncertainty quantification and dimension reduction for large-scale computational simulations. |