Title: Generalized sufficient dimension reduction in the presence of categorical predictors Authors:  Ben Jones - Aerospace Sector (United Kingdom) [presenting]
Abstract: Measure-theoretic developments in sufficient dimension reduction have enabled its application with predictors and/or responses lying in separable metric spaces, while allowing nonlinear reductions. A significant limitation of these developments is that they do not allow for the presence of additional categorical predictors, which we want to use to constrain the dimension reduction. An extension which overcomes this limitation is presented. Conceptual definitions are first given to set up the problem technically, then the novel estimator "partial generalized sliced inverse regression" of the target of estimation is described. The results of this method are further illustrated by real-world data.