1: \begin{abstract}

2: Extracting latent low-dimensional structure from high-dimensional data is of paramount 

3: importance in timely inference tasks encountered with `Big Data' analytics. However, 

4: increasingly noisy, heterogeneous, and incomplete datasets as well as the need for 

5: {\em real-time} processing of streaming data pose major challenges to this end. In this 

6: context, the present paper permeates benefits from rank minimization to scalable imputation of missing data,

7: via tracking low-dimensional subspaces and unraveling latent (possibly multi-way) structure 

8: from \emph{incomplete streaming} data. For low-rank matrix data, a subspace estimator is proposed based on an 

9: exponentially-weighted least-squares criterion regularized with the nuclear norm. 

10: After recasting the non-separable nuclear norm into a form amenable to online optimization, 

11: real-time algorithms with complementary strengths are developed and their convergence is established under simplifying 

12: technical assumptions. In a stationary setting, the asymptotic estimates 

13: obtained offer the well-documented performance guarantees of the 

14: {\em batch} nuclear-norm regularized estimator. Under the same

15: unifying framework, a novel online (adaptive) algorithm is developed to obtain

16: multi-way decompositions of \emph{low-rank tensors} with

17: missing entries, and perform imputation as a byproduct. Simulated 

18: tests with both synthetic as well as real Internet and cardiac

19: magnetic resonance imagery (MRI) data confirm the efficacy of the proposed

20: algorithms, and their superior performance relative to state-of-the-art alternatives.

21: \end{abstract}

22: