TL;DR: The paper presents mixPFC, a model combining subspace clustering with sufficient dimension reduction for heterogeneous data, achieving clustering, subspace estimation, and variable selection with theoretical guarantees.
Abstract: Scientific and engineering applications are often heterogeneous, making it beneficial to account for latent clusters or sub-populations when learning low-dimensional subspaces in supervised learning, and vice versa. In this paper, we combine the concept of subspace clustering with model-based sufficient dimension reduction and thus generalize the sufficient dimension reduction framework from homogeneous regression setting to heterogeneous data applications. In particular, we propose the mixture of principal fitted components (mixPFC) model, a novel framework that simultaneously achieves clustering, subspace estimation, and variable selection, providing a unified solution for high-dimensional heterogeneous data analysis. We develop a group Lasso penalized expectation-maximization (EM) algorithm and obtain its non-asymptotic convergence rate. Through extensive simulation studies, mixPFC demonstrates superior performance compared to existing methods across various settings. Applications to real world datasets further highlight its effectiveness and practical advantages.
Lay Summary: Many scientific and engineering problems involve complex, high-dimensional data that are heterogeneous and do not follow a single pattern. This paper introduces a new method called mixture of principal fitted components (mixPFC) that can uncover hidden groups in the data while also reducing the dimensionality and complexity of the data, and identifying the most important variables. The approach offers a unified solution for analyzing large, heterogeneous datasets more effectively.
Link To Code: https://github.com/leiyan-ly/mixPFC
Primary Area: General Machine Learning->Clustering
Keywords: High-dimensional inference, EM algorithm, Mixture model
Submission Number: 14682
Loading