Research on learned cardinality estimation has achieved significant progress in recent years. However, existing methods still face distinct challenges that hinder their practical deployment in production environments....
详细信息
Trajectory Prediction (TP) is an important research topic in computer vision and robotics fields. Recently, many stochastic TP models have been proposed to deal with this problem and have achieved better performance t...
详细信息
Outlier detection is an important data mining task, and developing effective methods to detect outliers is challenging in cases where there is insufficient labeled data. Manually labeling the data is labor-intensive a...
详细信息
Outlier detection is an important data mining task, and developing effective methods to detect outliers is challenging in cases where there is insufficient labeled data. Manually labeling the data is labor-intensive and time-consuming. Because of a limited number of labeled samples, the classes are unbalanced, resulting in a class-imbalance problem. Existing methods fail to address these aforementioned issues holistically and fall short in generating quality outlier samples for effective outlier detection accuracy. In this paper, we propose a new solution that tackles these problems. We propose a. Generative Adversarial Active Learning method (DIR-GAAL), which generates Diverse, Informative, and Representative outlier samples through active learning, and employs the mini-max game between the generator and discriminator in a generative adversarial network. We conducted extensive experiments on several benchmark datasets to evaluate the performance of our method. When compared to other benchmark methods, our method consistently demon-strates better outlier detection accuracy without being negatively affected by the class-imbalance problem.
With the rise of the cloud computing, saving energy consumed by cloud systems has become a tricky issue nowadays. How to place data efficiently and schedule the nodes effectively in a cloud platform are very important...
详细信息
With the rise of the cloud computing, saving energy consumed by cloud systems has become a tricky issue nowadays. How to place data efficiently and schedule the nodes effectively in a cloud platform are very important issues from the view of the energy-saving. However, the state-of-the-art node-scheduling strategies can't save large amount of energy for the cloud computing platforms significantly. This paper proposes a heuristic data placement algorithm and two node scheduling strategies for cloud platforms to save energy with tasks guaranteed. The Cloudsim is employed to simulate a private cloud system. Energy-saving is achieved by turning on minimum nodes to cover maximum data blocks. The problem of covering data block with computing nodes is abstracted as a set cover problem, and a greedy algorithm is utilized to solve this problem. This approach is practical to any cloud computing infrastructure. The designed experiment verifies the efficiency of the data placement algorithm and node scheduling strategies proposed in this paper.
As more and more power is consumed in servers, growing attention has been paid to power management. However, little work has been dated on the power-efficiency of query execution plans generated by the DBMS. Tradition...
详细信息
As more and more power is consumed in servers, growing attention has been paid to power management. However, little work has been dated on the power-efficiency of query execution plans generated by the DBMS. Traditional optimal query execution plan does not take power-efficiency into account. This paper tries to generate a power-efficient query execution plan by redesigning the query optimizer of the database. We build a power model to estimate the power consumption of a query execution plan and propose an algorithm to generate the power-efficient query execution plan. Using two experiments based on the TPC-H benchmark, we evaluate the accuracy of our power model and compare the power-efficient query execution plan with the traditional optimal query execution plan.
暂无评论