1: \begin{abstract}

2: We consider the following data perturbation model, where the 

3: covariates incur multiplicative errors.

4: For two $n \times m$ random matrices $U, X$, we denote by $U \circ X$

5: the Hadamard or Schur product, which is defined as $(U \circ X)_{ij} = (U_{ij}) \cdot (X_{ij})$.

6: In this paper, we study the subgaussian matrix variate model,

7: where we observe the matrix variate data $X$ through a random mask $U$:

8: \bens

9: \label{eq::Xgenabs}

10: \X = U \circ X \; \; \; \text{ where} \; \; \;X = B^{1/2} \Z A^{1/2},

11: \eens

12: where $\Z$ is a random matrix with independent subgaussian entries, and

13: $U$ is a mask matrix with either zero or positive entries, where $\E

14: U_{ij} \in [0, 1]$ and all entries are mutually independent. Subsampling in rows, or columns, or random sampling 

15: of entries of $X$ are special cases of this model. Under the assumption of independence between $U$ and $X$,

16: we introduce componentwise unbiased estimators for estimating covariance $A$ and 

17: $B$, and prove the concentration of measure bounds in the sense of guaranteeing the restricted eigenvalue($\RE$) 

18: conditions to hold on the unbiased estimator for $B$, when columns of data

19: matrix $X$ are sampled with different rates.

20: %Equipped with such theory,

21: We further develop multiple regression methods for estimating the

22: inverse of $B$ and show statistical rate of convergence.

23: %under model~\eqref{eq::Xgenabs}

24: Our results provide insight for sparse recovery for relationships

25: among entities (samples, locations, items) when features (variables,

26: time points, user ratings) are present in the observed data matrix

27: $\X$ with heterogeneous rates. Our proof techniques can certainly be extended

28: to other scenarios. We provide simulation evidence illuminating the theoretical predictions.

29: \end{abstract}

30: