1: \begin{abstract}

2: We present a high-dimensional analysis of three popular

3: algorithms, namely, Oja's method, GROUSE and PETRELS, for subspace estimation from streaming and highly incomplete observations.  We show that, with proper time scaling, the time-varying

4:  principal angles between the true subspace and its estimates given by the algorithms

5:  converge weakly to deterministic processes when the ambient dimension $n$ tends to infinity. Moreover, the limiting processes can be exactly characterized as the unique solutions of certain ordinary differential equations (ODEs). A finite sample bound is also given, showing that the rate of convergence towards such limits is $\mathcal{O}(1/\sqrt{n})$. In addition to providing asymptotically exact predictions of the dynamic performance of the algorithms, our high-dimensional analysis yields several insights, including an asymptotic equivalence between Oja's method and GROUSE, and a precise scaling relationship linking the amount of missing data to the signal-to-noise ratio. By analyzing the solutions of the limiting ODEs, we also establish phase transition phenomena associated with the steady-state performance of these techniques.

6:  

7:  

8: %Our analysis implies a non-local convergence guarantee, which may lead to the design of new efficient algorithms and guide how

9: %to adaptively tune the algorithmic parameters. 

10: %YE: What's our 'non-local convergence guarantee' is?

11: 

12: % YE: would be good if possible to give a flavor of the result.

13: % CW: We are actively working on design a new adaptive algorithm, which can achieve the fastest decay rate under the assumption of generative model when the dimension of the subspace is 1.

14: \end{abstract}

15: