Qing Liao

COD: A Cooperative Cell Outage Detection Architecture for Self-Organizing Femtocell Networks

The vision of Self-Organizing Networks (SON) has been drawing considerable attention as a major axis for the development of future networks. As an essential functionality in SON, cell outage detection is developed to autonomously detect macrocells or femtocells that are inoperative and unable to provide service. Previous cell outage detection approaches have mainly focused on macrocells while the outage issue in the emerging femtocell networks is less discussed. However, due to the two-tier macro-femto network architecture and the small coverage nature of femtocells, it is challenging to enable outage detection functionality in femtocell networks. Based on the observation that spatial correlations among users can be extracted to cope with these challenges, this paper proposes a Cooperative femtocell Outage Detection (COD) architecture which consists of a trigger stage and a detection stage. In the trigger stage, we design a trigger mechanism that leverages correlation information extracted through collaborative filtering to efficiently trigger the detection procedure without inter-cell communications. In the detection stage, to improve detection accuracy, we introduce a sequential cooperative detection rule to process spatially and temporally correlated user statistics. Numerical studies for a variety of femtocell deployments and configurations demonstrate that COD outperforms the existing scheme in both communication overhead and detection accuracy.

Local coordinate based graph-regularized NMF for image representation

Abstract Non-negative matrix factorization (NMF) has been a powerful data representation tool which has been widely applied in pattern recognition and computer vision due to its simplicity and effectiveness. However, existing NMF methods suffer from one or both of the following deficiencies: (1) they cannot theoretically guarantee the decomposition results to be sparse, and (2) they completely neglect geometric structure of data, especially when some examples are heavily corrupted. In this paper, we propose a local coordinate based graph regularized NMF method (LCGNMF) to simultaneously overcome both deficiencies. In particular, LCGNMF enforces the learned coefficients to be sparse by incorporating the local coordinate constraint over both factors meanwhile preserving the geometric structure of the data by incorporating graph regularization. To enhance the robustness of NMF, LCGNMF removes the effect of the outliers via the maximum correntropy criterion (MCC). LCGNMF is difficult because the MCC induced objective function is neither quadratic nor convex. We therefore developed a multiplicative update rule to solve LCGNMF and theoretically proved its convergence. Experiments of image clustering on several popular image datasets verify the effectiveness of LCGNMF compared to the representative methods in quantities.

Analyzing the influential people in Sina Weibo dataset

With the increasingly rapid growth of micro-blogging services, influence analysis is becoming a very important topic in this area. Sina Weibo, one of the largest mirco-blogging services in China, has provided a new operation comment-only, which allows users to give feedback on a post without forwarding. However, most of existing works focus on Twitter, which fail to consider this new operation. In paper, we propose a new influence measurement method called WeiboRank on Sina Weibo which applies comment-only operation to help researchers to find ignored influential users who have big influential in comment dimension. Furthermore we analyze why a particular user is influential based on tracing the source of influence to find out which aspects contribute to influence. Our experiments based on a subset of a whole Sina Weibo datatset, which includes three-month records of 22,514,394 users. We conduct extensive experiments and results show that we can accurately find the most influential individuals among entire social networks, while the running time of the algorithm increases linearly with any increase in data size, which is suitable for large scale networks.

Logdet Divergence Based Sparse Non-Negative Matrix Factorization for Stable Representation

Non-negative matrix factorization (NMF) decomposes any non-negative matrix into the product of two low dimensional non-negative matrices. Since NMF learns effective parts-based representation, it has been widely applied in computer vision and data mining. However, traditional NMF has the riskrisk learning rank-deficient basis learning rank-deficient basis on high-dimensional dataset with few examples especially when some examples are heavily corrupted by outliers. In this paper, we propose a Logdet divergence based sparse NMF method (LDS-NMF) to deal with the rank-deficiency problem. In particular, LDS-NMF reduces the risk of rank deficiency by minimizing the Logdet divergence between the product of basis matrix with its transpose and the identity matrix, meanwhile penalizing the density of the coefficients. Since the objective function of LDS-NMF is nonconvex, it is difficult to optimize. In this paper, we develop a multiplicative update rule to optimize LDS-NMF in the frame of block coordinate descent, and theoretically prove its convergence. Experimental results on popular datasets show that LDS-NMF can learn more stable representations than those learned by representative NMF methods.

Efficient rank-one residue approximation method for graph regularized non-negative matrix factorization

Nonnegative matrix factorization NMF aims to decompose a given data matrix X into the product of two lower-rank nonnegative factor matrices UV T . Graph regularized NMF GNMF is a recently proposed NMF method that preserves the geometric structure of X during such decomposition. Although GNMF has been widely used in computer vision and data mining, its multiplicative update rule MUR based solver suffers from both slow convergence and non-stationarity problems. In this paper, we propose a new efficient GNMF solver called rank-one residue approximation RRA. Different from MUR, which updates both factor matrices U and V as a whole in each iteration round, RRA updates each of their columns by approximating the residue matrix by their outer product. Since each column of both factor matrices is updated optimally in an analytic formulation, RRA is theoretical and empirically proven to converge rapidly to a stationary point. Moreover, since RRA needs neither extra computational cost nor parametric tuning, it enjoys a similar simplicity to MUR but performs much faster. Experimental results on real-world datasets show that RRA is much more efficient than MUR for GNMF. To confirm the stationarity of the solution obtained by RRA, we conduct clustering experiments on real-world image datasets by comparing with the representative solvers such as MUR and NeNMF for GNMF. The experimental results confirm the effectiveness of RRA.

Predicting Unknown Interactions Between Known Drugs and Targets via Matrix Completion

Drug-target interactions map patterns, associations and relationships between drugs and target proteins. Identifying interactions between drug and target is critical in drug discovery, but biochemically validating these interactions are both laborious and expensive. In this paper, we propose a novel interaction profiles based method to predict potential drug-target interactions by using matrix completion. Our method first arranges the drug-target interactions in a matrix, whose entries include interaction pairs, non-interaction pairs and undetermined pairs, and finds its approximation matrix which contains the predicted values at undetermined positions. Then our method learns an approximation matrix by minimizing the distance between the drug-target interaction matrix and its approximation subject that the values in the observed positions equal to the known interactions at the corresponding positions. As a consequence, our method can directly predict new potential interactions according to the high values at the undetermined positions. We evaluated our method by comparing against five counterpart methods on “gold standard” datasets. Our method outperforms the counterparts, and achieves high AUC and \(F_1\)-score on enzyme, ion channel, GPCR, nuclear receptor and integrated datasets, respectively. We showed the intelligibility of our method by validating some predicted interactions in both DrugBank and KEGG databases.

Robust Local Coordinate Non-negative Matrix Factorization via Maximum Correntropy Criteria

Non-negative matric factorization (NMF) decomposes a given data matrix X into the product of two lower dimensional non-negative matrices U and V. It has been widely applied in pattern recognition and computer vision because of its simplicity and effectiveness. However, existing NMF methods often fail to learn the sparse representation on high-dimensional dataset, especially when some examples are heavily corrupted. In this paper, we propose a robust local coordinate NMF method (RLCNMF) by using the maximum correntropy criteria to overcome such deficiency. Particularly, RLCNMF induces sparse coefficients by imposing the local coordinate constraint over both factors. To solve RLCNMF, we developed a multiplicative update rules and theoretically proved its convergence. Experimental results on popular image datasets verify the effectiveness of RLCNMF comparing with the representative methods.

Local Coordinate Projective Non-negative Matrix Factorization

Non-negative matrix factorization (NMF) decomposes a group of non-negative examples into both lower-rank factors including the basis and coefficients. It still suffers from the following deficiencies: 1) it does not always ensure the decomposed factors to be sparse theoretically, and 2) the learned basis often stays away from original examples, and thus lacks enough representative capacity. This paper proposes a local coordinate projective NMF (LCPNMF) to overcome the above deficiencies. Particularly, LCPNMF induces sparse coefficients by relaxing the original PNMF model meanwhile encouraging the basis to be close to original examples with the local coordinate constraint. Benefitting from both strategies, LCPNMF can significantly boost the representation ability of the PNMF. Then, we developed the multiplicative update rule to optimize LCPNMF and theoretically proved its convergence. Experimental results on three popular frontal face image datasets verify the effectiveness of LCPNMF comparing to the representative methods.