Robust Covariance Estimation for Entry-wise Contaminated and Partially Observed Data via Expectation Maximization Algorithm

The paper presents a robust covariance estimation method for data that is both entry-wise contaminated and partially observed. It extends the Student‑t* model by using a stochastic approximation EM algorithm that operates on a reduced latent space, avoiding explicit sampling of latent Gaussian variables while maintaining a likelihood‑based framework. Experiments on synthetic and real sensor data demonstrate that this approach achieves lower Fisher‑Rao error than Gaussian, Student‑t, imputation‑based, and detect‑then‑estimate estimators, especially at moderate to high outlier rates under diffuse contamination.