Structured Multi-Way Learning
Matrix, tensor, network, and spatial-temporal data; tensor-structured neural networks
This stream develops statistical foundations for learning from high-dimensional multi-way data — matrices, tensors, and networks — where the signal is carried by low-rank and parsimonious structure across modes. The foundational layer establishes factor models, low-rank regression, and estimation for matrix- and tensor-valued data under Kronecker/Tucker structure, with emphasis on time-dependent and spatially structured settings. My independent program extends this framework in three directions: beyond Tucker to CP, tensor-train, and mode-wise additive structures; from unsupervised dimension reduction to supervised tensor learning (regression, discriminant analysis, Neyman-Pearson classification); and from classical estimators to tensor-augmented neural networks and transformers.
Representative work: Semiparametric Tensor Factor Analysis (JRSS-B, 2024), Factor Augmented Matrix Regression (JASA, 2025), Distributed Tensor PCA with Data Heterogeneity (JASA, 2025), Dual-Channel Tensor Neural Networks. See publications for the full list.