We will have a meeting on Friday (January 30) at 3pm in SAS 5270 with speaker Troy Larsen.
Title: Introducing the Total HSIC Index for Global Sensitivity Analysis
Abstract: Global sensitivity analysis (GSA) is a fundamental component of uncertainty quantification used to identify which input parameters most influence a model’s output. Traditional methods, such as variance-based Sobol’ indices, are limited because they only capture influence that manifests through variance and assume statistical independence of inputs. This can lead to inaccurate sensitivity scores, particularly when dealing with complex, non-linear dependencies. We propose an alternative approach using the Hilbert-Schmidt independence criterion (HSIC). Unlike variance-based methods, HSIC is a moment-independent statistic that captures statistical dependence by embedding probability measures into reproducing kernel Hilbert spaces (RKHSs). Our findings demonstrate that this approach offers a compelling and flexible alternative to traditional GSA methods, capable of detecting a wider range of parameter influence.