Recently, a paper by Ying Chao, a PhD student in the School of Statistics, Faculty of Economics and Management, was published online in Biometrika, a top journal for statistics. The paper is titled Frechet Sufficient Dimension Reduction for Random Objects. The first author is Ying Chao, and the corresponding author is Prof. Yu Zhou.
The paper proposes a novel adequate dimensionality reduction method, Weighted Inverse Regression Ensemble (WIRE). This method solves for the first time the dimensionality reduction problem in the case where the dependent variable is a general metric space. Based on the regenerative kernel Hilbert method, the author further defines a new class of operators and constructs a nonlinear full dimensionality reduction method in the metric space. Furthermore, the author establishes a statistical convergence theory of linear and nonlinear WIRE dimensionality reduction methods.
Images of facial expressions and dimensionality reduction of the nonlinear WIRE method
The WIRE method proposed by the author can be widely applied to data structures with complex dependent variables, such as population mortality data and brain MRI images and can also handle classification problems, such as analyzing multilabel facial expression data and classifying yeast genes.
Statistical talents training is always committed to the improvement of academic ability. In recent years, graduate students Liu Yang, Tan Kai and Ying Chao, as the first authors, have published papers in the top journals of statistics, including Journal of the Royal Statistical Society Series B, Annals of Statistics, and Biometrika.
Copy editor: Henry Allen
Editor: Li Yinan
