Secure and Verifiable Outsourcing of Nonnegative Matrix
Factorization (NMF)
Jia Duan
University of Macau
Macau, China
xuelandj@gmail.com
Jiantao Zhou
University of Macau
Macau, China
jtzhou@umac.mo
Yuanman Li
University of Macau
Macau, China
yuanmanx.li@gmail.com
ABSTRACT
Cloud computing platforms are becoming increasingly preva-
lent and readily available nowadays, providing us alterna-
tive and economic services for resource-constrained clients
to perform large-scale computation. In this work, we ad-
dress the problem of secure outsourcing of large-scale non-
negative matrix factorization (NMF) to a cloud in a way
that the client can verify the correctness of results with
small overhead. The input matrix protection is achieved
by a lightweight, permutation-based encryption mechanism.
By exploiting the iterative nature of NMF computation, we
propose a single-round verification strategy, which can be
proved to be effective. Both theoretical and experimental
results are given to demonstrate the superior performance
of our scheme.
CCS Concepts
•Security and privacy → Privacy-preserving proto-
cols; •Computer systems organization → Cloud com-
puting;
Keywords
Cloud computing; NMF; secure outsourcing; verification
1. INTRODUCTION
Multimedia, as the most important and valuable source
for insights and information, is increasingly becoming the
“biggest big data”. Analyzing multimedia big data and turn-
ing them into significant insights for clients, however, have
been proven to be quite challenging. Nonnegative matrix
factorization (NMF), as a fundamental data analysis tech-
nique, has been widely utilized in many applications for
analyzing multimedia data, e.g., source separation [5] and
feature extraction [9, 18]. In the new era of big data, the
usage of NMF in practical settings has undergone a few new
challenges. The matrices to be analyzed are becoming aston-
Permission to make digital or hard copies of all or part of this work for personal or
classroom use is granted without fee provided that copies are not made or distributed
for profit or commercial advantage and that copies bear this notice and the full cita-
tion on the first page. Copyrights for components of this work owned by others than
ACM must be honored. Abstracting with credit is permitted. To copy otherwise, or re-
publish, to post on servers or to redistribute to lists, requires prior specific permission
and/or a fee. Request permissions from permissions@acm.org.
IH&MMSec 2016, June 20–23, 2016, Vigo, Spain.
c
2016 ACM. ISBN 978-1-4503-4290-2/16/06. . . $15.00
DOI: http://dx.doi.org/10.1145/2909827.2930794
ishingly big, and the involved computation is of very large-
scale, making it prohibitive to resource-constrained clients.
Fortunately, cloud computing paradigm provides a viable
solution for small to medium size business to perform large-
scale data computation [19]. In the outsourcing computation
framework, the clients with insufficient computing facilities
can outsource the heavy computation workload to the cloud
server, and enjoy the unlimited computing resources. Secu-
rity and privacy considerations, however, stand on the way
of fully utilizing the benefits of such cloud services and ar-
chitectures [3, 16]. First of all, the computation tasks often
contain some sensitive information that should not be ex-
posed to the cloud server, which usually is untrusted. Tra-
ditional encryption algorithms can only partially solve the
protection problem, as it is difficult to perform meaningful
(complex) computations over the encrypted domain. Fur-
thermore, clients should have a mechanism to verify the cor-
rectness of the results, because the cloud may return invalid
results for financial incentives. Certainly, the overhead of
performing the verification should be minimized. Along this
line, Wang et al. [17] proposed to outsource the large-scale
systems of linear equations to a public cloud. Lei et al. [11]
addressed the problem of designing a protocol for securely
outsourcing the matrix determinant computation to a cloud.
Recently, Blanton et al. suggested to outsource the large-
scale biometric computations to a cloud [2].
In this work, we address the problem of secure outsourcing
of large-scale nonnegative matrix factorization (NMF) to a
cloud in a way that the client can verify the correctness of
results with small overhead. The input matrix protection is
achieved by a lightweight encryption mechanism, involving
multiplication of permutation matrices. The task of verify-
ing the correctness of NMF is challenging, as factorization
results are not unique and the multiplication of the obtained
factors is only an approximation of the original matrix. To
overcome this difficulty, we design a single-round verifica-
tion strategy by exploiting the iterative nature of NMF com-
putation, and prove that the simple verification strategy is
very effective in verifying the results. Both theoretical and
experimental results are given to demonstrate the superior
performance of our scheme.
The rest of the paper is organized as follows. Section 2
gives an overview of NMF. In Section 3, we describe the
system model and the design goals. Section 4 presents the
proposed protocol for secure and verifiable outsourcing of
NMF. Section 5 is devoted to system performance analysis
and Section 6 offers the experimental results. Finally, we
conclude in Section 7.