Iterative SVD (Iterative Singular Value Decomposition) is a robust algorithm for imputing missing values in data matrices. It employs an EM-like two-step process: estimating missing entries (E-step) and updating the low-rank approximation via SVD (M-step). These steps iteratively refine the matrix, preserving its underlying multivariate structure. Ideal for multivariate datasets with missing data.
Roweis, S. (1997). EM algorithms for PCA and SPCA. Advances in Neural Information Processing Systems, 10.
