Direct Robust Matrix Factorizatoin for Anomaly Detection

Xiong, Liang; Chen, Xi; Schneider, Jeff

doi:10.1109/icdm.2011.52

Cited by 90 publications

(61 citation statements)

References 17 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In order to preserve the geometry of labelsets in their mappings, a manifold regularization term constructed from the graph Laplacian of labelsets is imposed to the decomposition. The decomposition is fast due to an application of the bilateral random projection (BRP) based low-rank approximation [22] [17]. Its convergence to a stationary point is guaranteed.…”

Section: Main Contributionsmentioning

confidence: 99%

Labelset anchored subspace ensemble (LASE) for multi-label annotation

Zhou

Tao

2012

Proceedings of the 2nd ACM International Conference on Multimedia Retrieval

View full text Add to dashboard Cite

In multimedia retrieval, multi-label annotation for image, text and video is challenging and attracts rapidly growing interests in past decades. The main crux of multi-label annotation lies on 1) how to reduce the model complexity when the label space expands exponentially with the increase of the number of labels; and 2) how to leverage the label correlations which have broadly believed useful for boosting annotation performance. In this paper, we propose "labelsets anchored subspace ensemble (LASE)" to solve both problems in an efficient scheme, whose training is a regularized matrix decomposition and prediction is an inference of group sparse representations. In order to shrink the label space, we firstly introduce "label distilling" extracting the frequent labelsets to replace the original labels. In the training stage, the data matrix is decomposed as the sum of several low-rank matrices and a sparse residual via a randomized optimization, where each low-rank part defines a feature subspace mapped by a labelset. A manifold regularization is applied to map the labelset geometry to the geometry of the obtained subspaces. In the prediction stage, the group sparse representation of a new sample on the subspace ensemble is estimated by group lasso. The selected subspaces indicate the labelsets that the sample should be annotated with. Experiments on several benchmark datasets of texts, images, web data and videos validate the appealing performance of LASE in multi-label annotation.

show abstract

Section: Main Contributionsmentioning

confidence: 99%

Labelset anchored subspace ensemble (LASE) for multi-label annotation

Zhou

Tao

2012

Proceedings of the 2nd ACM International Conference on Multimedia Retrieval

View full text Add to dashboard Cite

show abstract

“…In these works, the initial formulation of optimizing a matrix rank and l 0 norm is relaxed to optimizing a nuclear norm and l 1 norm. Very recent work [19] [20] shows that this relaxation may not be neccessary and direct factorization is possible.…”

Section: Introductionmentioning

confidence: 99%

“…Our derivation initially follows that of RASL ( [12]); however, our algorithm is different from RASL in several aspects. Our algorithm does not rely on the relaxation of matrix rank to nuclear norm and l 0 norm to l 1 norm; instead we apply direct factorization as in [19] [20]. Our formulation considers entry-wise noise, not considered in RASL.…”

Section: Introductionmentioning

confidence: 99%

Direct Matrix Factorization and Alignment Refinement: Application to Defect Detection

Qin

Beek²,

Chen

2014

2014 Canadian Conference on Computer and Robot Vision

View full text Add to dashboard Cite

Abstract-Defect detection approaches based on template differencing require precise alignment of the input and template image; however, such alignment is easily affected by the presence of defects. Often, non-trivial pre/post-processing steps and/or manual parameter tuning are needed to remove false alarms, complicating the system and hampering automation. In this work, we explicitly address alignment and defect extraction jointly, and provide a general iterative algorithm to improve both their performance to pixel-wise accuracy. We achieve this by utilizing and extending the robust rank minimization and alignment method of [12]. We propose an effective and efficient optimization algorithm to decompose a template-guided image matrix into a low-rank part relating to alignment-refined defect-free images and an explicit error component containing the defects of interest. Our algorithm is fully automatic, training-free, only needs trivial pre/postprocessing procedures, and has few parameters. The rank minimization formulation only requires a linearly correlated template image, and a template-guided approach relieves the common assumption of small defects, making our system very general. We demonstrate the performance of our novel approach qualitatively and quantitatively on a real-world dataset with defects of varying appearance.

show abstract

“…While the single-view based methods [157,158,143,144] are only designed to detect attribute-outliers. By representing the class-/attribute-outliers in the latent space and original feature space respectively, our proposed method well characterizes both types of outliers simultaneously.…”

Section: Definition Of Multi-view Outliermentioning

confidence: 99%

“…• DRMF, Direct Robust Matrix Factorization method is proposed by Xiong et al [158], which belongs to single-view outlier detection method. It has been shown the superiority to several other single-view baselines, e.g.…”

Section: Real-world Data and Settingsmentioning

confidence: 99%

Robust unsupervised subspace learning for visual representation

Zhao¹

View full text Add to dashboard Cite

Direct Robust Matrix Factorizatoin for Anomaly Detection

Cited by 90 publications

References 17 publications

Labelset anchored subspace ensemble (LASE) for multi-label annotation

Labelset anchored subspace ensemble (LASE) for multi-label annotation

Direct Matrix Factorization and Alignment Refinement: Application to Defect Detection

Robust unsupervised subspace learning for visual representation

Contact Info

Product

Resources

About