Time filter

Source Type

Wu J.,Beijing Jiaotong University | Wu J.,Beijing Key Laboratory of Traffic Data Analysis and Mining | Shen H.,Sun Yat Sen University | Shen H.,University of Adelaide | And 5 more authors.
Journal of Internet Technology | Year: 2014

Manifold Ranking (MR) is one popular and successful technique for relevance feedback in content-based image retrieval (CBIR). However, existing MR methods have two main drawbacks. First, the affinity matrix used by MR is computed purely based on the visual features of images, which fails to accurately capture the semantic structure of image database. Second, the existing MR methods often suffer from the "cold start" problem where the feedback example set is quite small. In this paper, we propose a novel scheme that double exploits the query log in MR to address the drawbacks. In details, the correlation between each pair of database images is first estimated based on a query log, which serves to adjust the affinity matrix towards semantic structure. Then, the relevance score of each database image to the user's query is further inferred from the query log, which could be used to produce more pseudo-labeled examples to handle the "cold start" problem. An empirical study shows that the proposed scheme is more effective than the state-of-the-art approaches.

Wu J.,Beijing Jiaotong University | Wu J.,Beijing Key Laboratory of Traffic Data Analysis and Mining | Xiao Z.-B.,Dalian Maritime University | Wang H.-S.,University of Technology, Sydney | And 2 more authors.
Computers and Electrical Engineering | Year: 2014

One of the challenges in image search is to learn with few labeled examples. Existing solutions mainly focus on leveraging either unlabeled data or query logs to address this issue, but little is known in taking both into account. This work presents a novel learning scheme that exploits both unlabeled data and query logs through a unified Manifold Ranking (MR) framework. In particular, we propose a local scaling technique to facilitate MR by self-tuning the scale parameter, and a soft label propagation strategy to enhance the robustness of MR against erroneous query logs. Further, within the proposed MR framework, a hybrid active learning method is developed, which is effective and efficient to select the informative and representative unlabeled examples, so as to maximally reduce users' labeling effort. An empirical study shows that the proposed scheme is significantly more effective than the state-of-the-art approaches. © 2013 Elsevier Ltd. All rights reserved.

Dong L.,Beijing Jiaotong University | Dong L.,Beijing Key Laboratory of Traffic Data Analysis and Mining | Wu H.,Beijing Jiaotong University
International Journal of Circuits, Systems and Signal Processing | Year: 2015

This paper focuses on the optimization of the brand wireless handover approach of the high speed train cyber physical system. It can improve the performance of handover. The simulation results show the handover approach can reduce the frequency of handover greatly. It can reduce the handover dropping probability and outage probability to support the safety of the high speed railway communication. © 2015, North Atlantic University Union. All rights reserved.

Jin Y.,Beijing Jiaotong University | Jin Y.,Beijing Key Laboratory of Traffic Data Analysis and Mining | Li J.,Beijing Jiaotong University | Li J.,Beijing Key Laboratory of Traffic Data Analysis and Mining | And 4 more authors.
Multidimensional Systems and Signal Processing | Year: 2016

Extreme learning machine (ELM) as a new emergent and efficient machine learning algorithm has shown its good performance in many real regression applications as well as large data classification. In this paper, we propose a new multi-task clustering ELM for cross-modal feature learning. Different to traditional face recognition methods, a coupled cross-modal feature learning based face descriptor is proposed to reduce the cross-modal differences, meanwhile, the multi-task learning is integrated with ELM for cross-modal classification. In this method, the discriminant feature learning is firstly proposed to learn the cross-modality feature representation. Then, common subspace learning based method is utilized to reduce the obtained cross-modality features. Finally, a multi-task clustering based ELM is proposed to improve the recognition accuracy by learning the shared information between tasks. Experiments conducted on two different VIS-NIR face recognition scenarios demonstrate the effectiveness of our proposed approach. © 2016 Springer Science+Business Media New York

Chai B.-F.,Beijing Key Laboratory of Traffic Data Analysis and Mining | Chai B.-F.,Shijiazhuang University of Economics | Yu J.,Beijing Key Laboratory of Traffic Data Analysis and Mining | Jia C.-Y.,Beijing Key Laboratory of Traffic Data Analysis and Mining | Wang J.-H.,Hebei Normal University
Ruan Jian Xue Bao/Journal of Software | Year: 2013

A stochastic block model can produce a wide variety of networks with different structures (named as general community, including traditional community, bipartite structure, hierarchical structure and etc); it also can detect general community in networks according to the rules of stochastic equivalence. However, the simple stochastic block model has some problems in modeling the generation of the networks and learning the models, showing poor results in fitting the practical networks. The GSB (general stochastic block) model is an extension of the stochastic block model, which is based on the idea of link community and is provided to detect general communities. But its complexity limits its applications in medium and large networks. In order to explore the latent structures of networks with different scales without prior knowledge about networks, a fast algorithm on the GSB model (FGSB) is designed to explore general communities in networks faster. FGSB dynamically learns the parameters related to the network structure in the process of iterations. It reduces the storage memory by reorganizing parameters to cut down unnecessary parameters, and saves the running time by pruning the related parameters of converging nodes and edges to decrease the computing time of each iteration. FGSB has the same ability of structure detection as the GSB model, but its complexities of time and storage are lower. Tests on synthetic benchmarks and real-world networks have demonstrated that FGSB not only can run faster than the algorithm of the GSB model in the similar accuracy, but also can detect general communities for large networks. ©Copyright 2013, Institute of Software, the Chinese Academy of Sciences.

Li Y.,Beijing Jiaotong University | Li Y.,Beijing Key Laboratory of Traffic Data Analysis and Mining | Jia C.,Beijing Jiaotong University | Jia C.,Beijing Key Laboratory of Traffic Data Analysis and Mining | And 4 more authors.
Shuju Caiji Yu Chuli/Journal of Data Acquisition and Processing | Year: 2015

Community structure is one of the most important topological characteristics in the complex network, being a hot research area in different fields. A novel community detection algorithm is proposed based on edges rank and modularity optimization. Local graph is sparsificated and edges are ranked according to the similarity. Therefore, a method called the fast rank-based community detection (FRCD) by maximizing modularity and fast mergement of edges is achieved. Meanwhile the method is also extended to dynamic and real-time community detection on the basis of initial community structure, and a fast and robust dynamic community detection algorithm called the incremental dynamic community detection (IDCD) is presented. Theoretical analysis exhibit that FRCD has linear complexity for network edges. Experimental results in real-world and artificial networks demonstrate the high accuracy and good performance of the algorithm on static community detection and tracking dynamic structure of networks. © 2015, Journal of Data Acquisition and Processing. All right reserved.

Yuan J.-D.,Beijing Jiaotong University | Yuan J.-D.,Beijing Key Laboratory of Traffic Data Analysis and Mining | Wang Z.-H.,Beijing Jiaotong University | Wang Z.-H.,Beijing Key Laboratory of Traffic Data Analysis and Mining | And 3 more authors.
Jisuanji Xuebao/Chinese Journal of Computers | Year: 2015

The shapelets of time series are subsequences of time series that are representative of class members. One of the most promising approaches to solve problems of time series classification is to separate the process of finding shapelets from classification algorithms by applying a shapelet transformation. The main advantages of this approach are that it could optimize the process of shapelets selection, and then various classification strategies might be adopted. However, there also exist some limitations in the process. First, only directly employing these shapelets, while ignoring logical relationship between them, the classification accuracy may be cut down. Second, the process is much more time consuming significantly, even though shapelets are computed offline. In this paper, the latter problem is addressed by using an intelligent caching based and reusable skill, which reduces the time complexity of finding shapelets by an order of magnitude. On this basis, a novel transformation that is based on conjunctive or disjunctive of shapelets is proposed. Experimental results have shown the efficiency of logical shapelets transformation on classic benchmark datasets used for these problems, which can improve classification accuracy, and whilst retaining their interpretability. ©, 2015, Science Press. All right reserved.

Loading Beijing Key Laboratory of Traffic Data Analysis and Mining collaborators
Loading Beijing Key Laboratory of Traffic Data Analysis and Mining collaborators