TITLE: Learning with Factor Analysis Models

ABSTRACT: This dissertation develops the theoretical and algorithmic framework that extends the principles of factor analysis (FA) to modern signal processing (SP) and machine learning (ML) applications. Classical FA models provide a foundation for discovering low-dimensional latent structures in high-dimensional data. However, their reliance on idealized assumptions—such as full-precision measurements, linear modeling, and paired supervision—limits their applicability in realistic scenarios involving quantization, nonlinear transformations, and unpaired multimodal data. To address these challenges, this work advances learning with factor analysis models by developing provable methods that integrates deep learning tools under complex data settings. The dissertation focuses on two ...