Complex computational science and engineering models contain several parameters representing physical inputs. Parameter studies, such as uncertainty quantification or design optimization, need to evaluate many models at different parameter values to endow predictions with credibility and confidence. However, the cost of parameter studies may grow exponentially with the number of inputs. One way to enable parameter studies in highly parameterized models is to identify low-dimensional structures in the map from input parameters to output predictions.
The DR17 workshop brings together researchers across mathematics, statistics, and engineering to explore a range of emerging techniques for parameter space dimension reduction. Topics of interest include:
- active subspaces
- basis adaptation
- inverse regression
- sufficient dimension reduction
- sloppy models
- sensitivity analysis
- ridge recovery and approximation
- deterministic and statistical parameter estimation
- applications with science and engineering simulations or data sets