1: \begin{abstract} 

2: Nonnegative matrix factorization (NMF) has been successfully

3: applied in several data mining tasks. Recently, there is an increasing

4: interest in the acceleration of NMF, due to its high cost on large matrices.

5: On the other hand, the privacy issue of NMF over federated data is worthy of

6: attention, since NMF is prevalently applied in image and text analysis which

7: may involve leveraging privacy data (e.g, medical image and record) across

8: several parties (e.g., hospitals). In this paper, we study the

9: \emph{acceleration} and \emph{security} problems of distributed NMF. Firstly,

10: we propose a {\em distributed sketched alternating nonnegative least squares}

11: (DSANLS) framework for NMF, which utilizes a matrix sketching technique to

12: reduce the size of nonnegative least squares subproblems with a convergence

13: guarantee. For the second problem, we show that DSANLS with modification can

14: be adapted to the security setting, but only for \emph{one or limited

15: iterations}. Consequently, we propose four efficient distributed NMF methods

16: in both synchronous and asynchronous settings with a security guarantee. We

17: conduct extensive experiments on several real datasets to show the superiority

18: of our proposed methods. The implementation of our methods is available at

19: \url{https://github.com/qianyuqiu79/DSANLS}.

20: \end{abstract}