Coursera Spring Sale
40% Off Coursera Plus Annual!
Grab it
Explore a mathematical lecture on advanced factor analysis techniques presented by Xin Bing from the University of Toronto at the Centre de recherches mathématiques. Delve into vintage factor analysis, a crucial statistical method that first creates low-dimensional data representations and then applies rotations to make these representations scientifically meaningful. Learn about the limitations of traditional Principal Component Analysis (PCA) followed by varimax rotation, particularly the theoretical challenges posed by non-convex optimization over orthogonal matrices. Discover the proposed deflation varimax procedure that sequentially solves each row of an orthogonal matrix, offering computational advantages and theoretical guarantees in broader contexts. Examine how this two-step procedure combining PCA with deflation varimax performs under general factor models, achieving minimax optimal rates for factor loading matrix estimation when signal-to-noise ratios are moderate to large. Understand the modifications needed for low SNR regimes and how structured additive noise can improve performance. Gain insights into finite sample theory that accommodates growing numbers of latent factors and ambient dimensions that can exceed sample sizes, supported by extensive simulation studies and real data analysis.
