The paper studies on the modeling and prediction method for multi-attribute events, which exist widely in practical application. In view of the features of multi-attribute events, the paper proposes an event processing framework including the construction of training sample space based on multi-attribute event grid (MAEG), bi-direction SVR (Support Vector Regression) modeling method for prediction based on longitudinal modeling and transverse modeling, and model updating strategy based on ?-insensitive error. The performance of the proposed methods of the framework is demonstrated by extensive experiments with real datasets.

In practical application, GPS trajectories are often sparse due to the sampling points lost or a new user appearing, which makes the accuracy of location prediction low based on a single user data. To solve this problem, this paper proposes a novel Markov location prediction approach based on user mobile behavior similarity clustering. First, this paper proposes a region partitioning method based on Voronoi diagram, transforms the GPS trajectories into region trajectories and predicts the locations over region trajectories. Second, this paper proposes a new approach to measure the similarity of users' mobile behavior by considering users' transferring features and regional features. Third?based on the mobile behavior similarity, this paper divides users into various groups and employs the first-order Markov model on the groups for location prediction. The experiments over real GPS trajectory dataset indicate that the proposed method is effective for location prediction.

The virtual machine (VM) live migration could achieve VM redistribution among distributed system hosts without reducing normal working performance. Post-copy is one of the wildly used VM live migration algorithm and has lots of advantages, such as less total migration time, less downtime, lower network overhead and so on. Its disadvantage is that the VM will be suspended frequently due to the page faults caused by the incomplete memory while VM is restored to run on destination host, which may lead to an extremely negative affect on VM work efficiency. To solve this problem, this paper proposes the pre-record algorithm. Pre-record extends the VM execution on source host, records the accessed memory pages during this period to obtain pre-recorded page set (PPS), and preferentially completes the migration of PPS to avoid page faults as much as possible. It also proposes PDoPMP algorithm to analysis the trend of the trajectory of memory address in PPS, in order to further expand the predict range of memory pages. The experimental results show that the pre-record has better efficiency than traditional post-copy, especially after combining with PDoPMP. It could obviously reduce the page faults number and then the total VM migration time without prolonging the downtime, and could improve VM migration efficiency under different workload and network conditions.

Given a set of spatial objects, facilities can influence the objects located within their influence regions that are represented by circular disks with the same radius $(r)$. Our task is to select the minimum number of locations such that establishing a temporary facility at each selected location would ensure that all the objects are influenced. Aiming to solve this location selection problem, we propose a novel kind of location selection query, called group location selection (GLS) queries. In many real-world applications, every object is usually located within an uncertainty region instead of at an exact point. Due to the uncertainty of the data, GLS processing needs to ensure that the probability of each uncertain object being influenced by one facility is not less than a given threshold $(\tau)$. An analysis of the time cost reveals that it is infeasible to exactly answer GLS queries over uncertain objects in polynomial time. Hence, this paper proposes an approximate query framework for answering queries efficiently while guaranteeing that the results of GLS queries are correct with a bounded probability. The performance of the proposed methods of the framework is demonstrated by theoretical analysis and extensive experiments with both real and synthetic data sets.

Complex event detection in stream is an important problem in event stream processing field. In this paper, we propose a new complex event detection algorithm in probabilistic stream, Instance Pruning and Filter-Detection Algorithm (IPF-DA). This algorithm is based on a kind of data structure called Chain Instance Queues (CIQ), to detect complex events satisfying query requirements with single-scanning probabilistic stream. In the process of complex event detection, IPF-DA prunes unnecessary event instances with query requirements and achieves filter for complex events with the given threshold. And it further improves the efficiency by setting proper tolerance, while insuring high recall. In addition, we construct Bayesian network to express and infer the probability distribution of uncertain events. Conditional Probability Indexing-Tree (CPI-Tree) is defined to store conditional probabilities of Bayesian network, saving query time compared with traditional Conditional Probability Table (CPT). Experimental results show that a series of strategies proposed by this paper are effective for complex event detection in probabilistic stream.

The key of distributed computing is to fully utilize computing resources. However, the efficiency of distributed computing can always be affected by volatility of available nodes and uncertainty of network environment. In order to enhance resource allocation efficiency in distributed computing, this paper presents an inframarginal analysis based resource allocation method, which distributes computing task to the node with more comparative advantage carrying on the task. Simulation results proved that it is effective on resource allocation in the distributed computing environment.

In this paper, authors propose a new learning based method for medical image retrieval which is based on fusing different features by linearly combining different similarities. Considering the abundant classes of medical images, this paper avoid to train a classifier for each class by using large amount training data. Instead, by using optimization method to combine different features’ similarity, new method can get good performance while has no much training computation. Experimental results show that the algorithm has potential practical values for clinical routine application.