Low-Rank Matrix Recovery based on Smooth
Function Approximation
Hengyou Wang
1,2,3
, Ruizhen Zhao
1,2
, Yigang Cen
1,2,*
, Fengzhen Zhang
1,2
1
Institute of Information Science, Beijing Jiaotong University, Beijing, 100044, China;
2
Key Laboratory of Advanced Information Science and Network Technology of Beijing, Beijing, 100044, China;
3
School of Science, Beijing University of Civil Engineering and Architecture, Beijing, 100044, China;
Abstract—Recently, smooth rank function was proposed for
matrix completion problem. The main idea of this method was
based on a continuous and differentiable approximation of the
rank function. In this paper, a new approach for solving low-
rank matrix recovery problem based on smooth function is
proposed. It not only uses a smooth function to approximate the
rank function, but also approximates the l
0
-norm with a
continuous and differentiable function. In addition, gradient
decreasing approach can be used to solve the minimization
problem. Finally, experimental results show that this new
algorithm provides accurate results in almost all of our testing
scenarios with a reasonable running time. Especially, it has
higher approximation performance than other methods.
Keywords—low-rank matrix recovery, rank minimization,
nuclear norm, smooth function, compressed sensing
I. INTRODUCTION
Learning the intrinsic data structures via matrix analysis
[1,2] has received wide attentions in many fields, e.g., neural
networks[3], learning systems [4], control theory [5],
computer vision [6], and pattern recognition [7]. There are
quite a number of efficient mathematical tools for rank
analysis, e.g., principal component analysis (PCA) and
singular value decomposition (SVD). However, these typical
approaches could only handle some simple problems. The
main difficulty of the problem is due to the fact that the rank
function term and l
0
-norm are both discontinuous and non-
differentiable. Indeed, the optimization problem is NP-hard
and all available optimizers have doubly exponential
complexity [8].
With the development of compressive sensing (CS), a new
concept on nuclear norm optimization was emerged for rank
minimization, i.e., the l
0
-norm is replaced by l
1
-norm. This
modification was known to be the tightest convex relaxation
of the rank minimization problem [9]. Similar to the
techniques used in compressive sensing, it is shown that under
mild conditions and with overwhelming probability, the
nuclear norm minimization technique achieves the same
solution as the original rank minimization approach [10]. This
optimization problem can be implemented by convex optimal
methods, e.g., Robust Principal Component Analysis (RPCA)
[11], Singular Value Thresholding (SVT) [12], and
Augmented Lagrange Multiplier (ALM) [13]. However, when
the desired matrix becomes complicated, e.g., it has high
intrinsic rank structure or the corrupted errors become dense,
the convex relax approach may not achieve promising
performance.
Recently, Mohimani [14] presented a fast algorithm for
overcomplete sparse decomposition based on smooth l
0
-norm.
Ghasemi and Male-Mohammadi [15], [16] proposed a new
algorithm for matrix completion based on smooth rank
function. In these works, the advantages of smooth function
approximation were used to better enhance the sparseness of
signals. Geometrically, smooth terms generally lie much
closer to essential rank function and l
0
-norm than the nuclear
norm and l
1
-norm.
In this paper, we focus on the more general problem of
low-rank matrix recovery. A novel approach based on a
continuous and differentiable approximation of the
discontinuous rank function and l
0
-norm is proposed. Our
work is inspired by the work of Ghasemi et al. [15], which
uses smooth rank function to approximate the rank function,
and the work of Mohimani et al. [14], which uses smooth l
0
-
norm to obtain sparse solutions of underdetermined system of
linear equations. Compare with previous approaches, our
proposed approach has better property for Gaussian added
noise.
Remain of this paper is organized as follows. In section 2,
problem formulation of low-rank matrix recovery is
introduced and in section 3, the algorithm for low-rank matrix
recovery is proposed. Section 4 presents some simulation
results. Finally, section 5 concludes the paper.
II. T
HE MODE OF LOW-RANK MATRIX RECONVERY BASED
ON
SMOOTH FUNCTION
In this section, some related works and mathematical
models of low-rank matrix recovery will be described. Firstly,
the conventional mathematical model of low-rank matrix
recovery is reviewed in follows. And then the model of low-
rank matrix recovery based on smooth function was presented.
Let
D
be the observed corrupted matrix. The target of
low-rank matrix recovery is to find a low-rank matrix
X
to
approximate the original noise free matrix. This problem can
be formulated as follows:
0
() M
min rank X E
s.t. E D X
(1)
Where
X
and
E
are low-rank matrix and sparse corruption,
respectively. In (1),
()rank X
is adopt to describe the low-
rank structure of the matrix X, and the sparse errors are
729978-1-5090-1345-6/16/$31.00 ©2016 IEEE ICSP2016