Covariance Matrix Estimation with Non Uniform and Data Dependent Missing Observations
arXiv:1910.00667 · doi:10.1109/TIT.2020.3039118
Abstract
In this paper we study covariance estimation with missing data. We consider missing data mechanisms that can be independent of the data, or have a time varying dependency. Additionally, observed variables may have arbitrary (non uniform) and dependent observation probabilities. For each mechanism, we construct an unbiased estimator and obtain bounds for the expected value of their estimation error in operator norm. Our bounds are equivalent, up to constant and logarithmic factors, to state of the art bounds for complete and uniform missing observations. Furthermore, for the more general non uniform and dependent cases, the proposed bounds are new or improve upon previous results. Our error estimates depend on quantities we call scaled effective rank, which generalize the effective rank to account for missing observations. All the estimators studied in this work have the same asymptotic convergence rate (up to logarithmic factors).
arXiv comment16 pages, 4 figures. Accepted at IEEE Transactions on Information Theory
References in corpus (7)
- Shrinkage Algorithms for MMSE Covariance Estimation
- High-dimensional regression with noisy and missing data: Provable guarantees with nonconvexity
- Active sequential hypothesis testing
- Missing values: sparse inverse covariance estimation and an extension to sparse regression
- On the sample covariance matrix estimator of reduced effective rank population matrices, with applications to fPCA
- Preconditioned Data Sparsification for Big Data with Applications to PCA and K-means
- Estimation of the sample covariance matrix from compressive measurements