A State-Space Approach to Dynamic Nonnegative Matrix Factorization

被引:18
|
作者
Mohammadiha, Nasser [1 ]
Smaragdis, Paris [2 ,3 ]
Panahandeh, Ghazaleh [4 ]
Doclo, Simon [1 ]
机构
[1] Carl von Ossietzky Univ Oldenburg, Dept Med Phys & Acoust, D-26111 Oldenburg, Germany
[2] Univ Illinois, Dept Comp Sci, Urbana, IL 61801 USA
[3] Univ Illinois, Dept Elect & Comp Engn, Urbana, IL 61801 USA
[4] KTH Royal Inst Technol, Sch Elect Engn, Signal Proc Grp, Stockholm, Sweden
关键词
Constrained Kalman filtering; nonnegative dynamical system (NDS); nonnegative matrix factorization (NMF); prediction; probabilistic latent component analysis (PLCA); BASES;
D O I
10.1109/TSP.2014.2385655
中图分类号
TM [电工技术]; TN [电子技术、通信技术];
学科分类号
0808 ; 0809 ;
摘要
Nonnegative matrix factorization (NMF) has been actively investigated and used in a wide range of problems in the past decade. A significant amount of attention has been given to develop NMF algorithms that are suitable to model time series with strong temporal dependencies. In this paper, we propose a novel state-space approach to perform dynamic NMF (D-NMF). In the proposed probabilistic framework, the NMF coefficients act as the state variables and their dynamics are modeled using a multi-lag nonnegative vector autoregressive (N-VAR) model within the process equation. We use expectation maximization and propose a maximum-likelihood estimation framework to estimate the basis matrix and the N-VAR model parameters. Interestingly, the N-VAR model parameters are obtained by simply applying NMF. Moreover, we derive a maximum a posteriori estimate of the state variables (i.e., the NMF coefficients) that is based on a prediction step and an update step, similarly to the Kalman filter. We illustrate the benefits of the proposed approach using different numerical simulations where D-NMF significantly outperforms its static counterpart. Experimental results for three different applications show that the proposed approach outperforms two state-of-the-art NMF approaches that exploit temporal dependencies, namely a nonnegative hidden Markov model and a frame stacking approach, while it requires less memory and computational power.
引用
下载
收藏
页码:949 / 959
页数:11
相关论文
共 50 条
  • [1] DYNAMIC MATRIX FACTORIZATION: A STATE SPACE APPROACH
    Sun, John Z.
    Varshney, Kush R.
    Subbian, Karthik
    2012 IEEE INTERNATIONAL CONFERENCE ON ACOUSTICS, SPEECH AND SIGNAL PROCESSING (ICASSP), 2012, : 1897 - 1900
  • [2] A state-space approach to indefinite spectral factorization
    Clements, DJ
    SIAM JOURNAL ON MATRIX ANALYSIS AND APPLICATIONS, 2000, 21 (03) : 743 - 767
  • [3] An Efficient Nonnegative Matrix Factorization Approach in Flexible Kernel Space
    Zhang, Daoqiang
    Liu, Wanquan
    21ST INTERNATIONAL JOINT CONFERENCE ON ARTIFICIAL INTELLIGENCE (IJCAI-09), PROCEEDINGS, 2009, : 1345 - 1350
  • [4] A multilevel approach for nonnegative matrix factorization
    Gillis, Nicolas
    Glineur, Francois
    JOURNAL OF COMPUTATIONAL AND APPLIED MATHEMATICS, 2012, 236 (07) : 1708 - 1723
  • [5] A PROJECTIVE APPROACH TO NONNEGATIVE MATRIX FACTORIZATION
    Groetzner, Patrick
    ELECTRONIC JOURNAL OF LINEAR ALGEBRA, 2021, 37 : 583 - 597
  • [6] Nonnegative matrix factorization in polynomial feature space
    Buciu, Ioan
    Nikolaidis, Nikos
    Pitas, Ioannis
    IEEE TRANSACTIONS ON NEURAL NETWORKS, 2008, 19 (06): : 1090 - 1100
  • [7] STATE-SPACE FORMULAS FOR THE FACTORIZATION OF ALL-PASS MATRIX FUNCTIONS
    GREEN, M
    ANDERSON, BDO
    INTERNATIONAL JOURNAL OF CONTROL, 1987, 45 (05) : 1575 - 1602
  • [8] A STATE-SPACE APPROACH TO DISCRETE-TIME SPECTRAL FACTORIZATION
    FAIRMAN, FW
    DANYLCHUK, GJ
    LOUIE, J
    ZAROWSKI, CJ
    IEEE TRANSACTIONS ON CIRCUITS AND SYSTEMS II-EXPRESS BRIEFS, 1992, 39 (03) : 161 - 170
  • [9] A state-space algorithm for the spectral factorization
    Kraffer, F
    Kwakernaak, H
    PROCEEDINGS OF THE 36TH IEEE CONFERENCE ON DECISION AND CONTROL, VOLS 1-5, 1997, : 4822 - 4823
  • [10] Nonnegative Matrix Factorization Approach for Image Reconstruction
    Wang, Yueyang
    Shafai, Bahram
    2021 INTERNATIONAL CONFERENCE ON COMPUTATIONAL SCIENCE AND COMPUTATIONAL INTELLIGENCE (CSCI 2021), 2021, : 1639 - 1642