检索结果-内蒙古大学图书馆

GradSA: Gradient Sparsification and Accumulation for Communication-Efficient Distributed Deep Learning 15th

学校读者我要写书评

暂无评论

GradSA: Gradient Sparsification and Accumulation for Communi...

15th International Conference on Green, Pervasive, and Cloud computing, GPC 2020

作者： Liu, Bo Jiang, Wenbin Zhao, Shaofeng Jin, Hai He, Bingsheng National Engineering Research Center for Big Data Technology and System Service Computing Technology and System Lab Cluster and Grid Computing Lab Huazhong University of Science and Technology Wuhan China Wuhan National Laboratory for Optoelectronics Key Laboratory of Information Storage System Engineering Research Center of Data Storage Systems and Technology Huazhong University of Science and Technology Wuhan China Library Henan University of Economics and Law Zhengzhou China Department of Computer Science School of Computing National University of Singapore Singapore Singapore

ISBN: (纸本)9783030642426

Large-scale distributed deep learning is of great importance in various applications. For distributed training, the inter-node gradient communication often becomes the performance bottleneck. Gradient sparsification has been proposed to reduce the communication overhead. However, the sparsification only arbitrarily selects a small fraction of gradients, while the efficiency of selection dimension has been overlooked. Furthermore, gradient staleness is inevitable after applying the sparsification which will ultimately lead to model divergence. In this paper, we propose a staleness-compensated sparse stochastic gradient descent algorithm, GradSA, to improve the training efficiency. Layer-level gradients are sparsified to reduce the communication overhead, which conform to the characteristics of the network structure, and historical accumulation of the approximated gradients is utilized to speed up convergence. We demonstrate the model convergence acceleration and the efficiency of our layer-level selection over existing state-of-the-art works, such as DGC, TernGrad, and 8-Bit quantization. © 2020, Springer Nature Switzerland AG.

关键词： Efficiency

An Automated Feature Engineering Method for Online Payment Fraud Detection

学校读者我要写书评

暂无评论

Jisuanji Xuebao/Chinese Journal of Computers 2020年第10期43卷 1983-2001页

作者： Wang, Cheng Wang, Chang-Qi Department of Computer Science and Technology School of Electronics and Information Engineering Tongji University Shanghai201804 China The Key Laboratory of Embedded System and Service Computing Ministry of Education Shanghai201804 China Shanghai Institute of Intelligent Science and Technology Tongji University Shanghai200092 China

Internet finance fraud is an increasingly serious social and economic problem. Online payment services (OPSs) are the typical models of Internet finance, and the fraudulent transaction in OPSs is also a typical fraud pattern. The method of identifying fraudulent transactions by constructing a fraud detection model based on machine learning has become a promising idea for online payment anti-fraud. In the process of constructing fraud detection models, the feature engineering is the most critical step. It is also one of the most time-consuming and specialized steps in the relevant area. In the study of feature engineering, the existing online payment fraud detection models are mainly carried out by experts in the form of manual construction based on business knowledge. However, there are many fraud scenarios in OPSs where the process of feature construction is so different. Artificial feature construction methods can no longer meet the increasing demand of anti-fraud. An important way to solve this problem is to automate feature engineering. In the field of Internet financial anti-fraud, the expressibility and interpretability of features play a pivotal role. It is helpful to understand the original source fields and their construction process of important features. This is useful for mining and analyzing the characteristics of fraud methods and follow-up improvement rules engines. These are of great significance for fraud detection models. Therefore, the interpretability of the model method is particularly important. Usually, the optimization of detection accuracy is carried out under the premise of ensuring interpretability. This paper proposed a lightweight, tree-structure, high efficiency and scalable automatic feature engineering method for fraud detection of online payment. The method is as follows: (1) The method has low requirements on the calculation conditions and little dependence on the dataset samples. To realize this advantage, it used the tree structur

关键词： Feature extraction

Modeling and analysis of three properties of mobile interactive systems based on variable Petri nets

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Yang, Ru Ding, Zhijun Jiang, Changjun Zhou, MengChu The Key Laboratory of Embedded System and Service Computing Ministry of Education Tongji University The Department of Computer Science and Technology Tongji University Shanghai201804 China The Institute of Systems Engineering Macau University of Science and Technology 999078 China The Department of Electrical and Computer Engineering New Jersey Institute of Technology NewarkNJ07102 United States

Due to the mobility and frequent disconnections, the correctness of mobile interaction systems, such as mobile robot systems and mobile payment systems, are often difficult to analyze. This paper introduces three critical properties of systems, called system connectivity, interaction soundness and data validity, and presents a related modeling and analysis method, based on a kind of Petri nets called VPN. For a given system, a model including component nets and interaction structure nets is constructed by using VPNs. The component net describes the internal process of each component, while the interaction structure net reflects the dynamic interaction between components. Based on this model, three properties are defined and analyzed. The case study of a practical mobile payment system shows the effectiveness of the proposed method. Copyright © 2021, The Authors. All rights reserved.

关键词： Petri nets

Variable petri nets for mobility

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Ding, Zhijun Yang, Ru Cui, Puwen Zhou, MengChu Jiang, Changjun Key Laboratory of Embedded System and Service Computing Ministry of Education Tongji University Shanghai201804 China Department of Computer Science and Technology Tongji University Shanghai201804 China Institute of Systems Engineering Macau University of Science and Technology Macau999078 China Department of Electrical and Computer Engineering New Jersey Institute of Technology NewarkNJ07102 United States

Mobile computing systems, service-based systems and some other systems with mobile interacting components have recently received much attention. However, because of their characteristics such as mobility and disconnection, it is difficult to model and analyze them by using a structure-fixed model. This work proposes a new Petri net model called Variable Petri Net (VPN) for modeling and analyzing these systems. The definition, firing rule, and related analysis technology of VPN are introduced in detail. In a VPN, the possible interaction interfaces are abstracted as a new kind of places called virtual places, and the occurrences of (dis)connections are described by new functions, which makes it appropriate to describe the component collaboration in systems and realize the scalability and pluggability of systems. Moreover, to overcome the shortcoming that markings cannot reflect link capability of a system, VPNs add a constraint function along with a marking to represent a complete system configuration. Several examples are used to demonstrate the newly proposed model and method. Copyright © 2021, The Authors. All rights reserved.

关键词： embedded systems

Base station network traffic prediction approach based on LMA-DeepAR

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Zhang, Jiachen Zuo, Xingquan Xu, Mingying Han, Jing Zhang, Baisheng Key Laboratory of Trustworthy Distributed Computing and Service Ministry of Education Beijing Uinversity of Posts and Telecommunications Beijing China Key Laboratory of Intelligent Telecommunication Software and Multimedia Beijing Uinversity of Posts and Telecommunications Beijing China Network Interface Virtualization Shanghai System Design Deptment Zhongxing Telecommunication Equipment Corporation Shanghai China

Accurate network traffic prediction of base station cell is very vital for the expansion and reduction of wireless devices in base station cell. The burst and uncertainty of base station cell network traffic makes the network traffic nonlinear and non-stationary, which brings challenges to the long-term prediction of network traffic. In this paper, the traffic model LMA-DeepAR for base station network is established based on DeepAR. Acordding to the distribution characteristics of network traffic, this paper proposes an artificial feature sequence calculation method based on local moving average (LMA). The feature sequence is input into DeepAR as covariant, which makes the statistical characteristics of network traffic near a period of time in the past be considered when updating parameters, and the interference of non-stationary network traffic on model training will be reduced. Experimental results show that the proposed prediction approach (LMA-DeepAR) outperforms other methods in the overall long-term prediction performance and stability of multi cell network traffic. Copyright © 2021, The Authors. All rights reserved.

关键词： Base stations

Robust analog precoding designs for millimeter wave MIMO transceivers 20

学校读者我要写书评

暂无评论

Robust analog precoding designs for millimeter wave MIMO tra...

20th International ITG Workshop on Smart Antennas, WSA 2016

作者： Xia, Pengfei Heath, Robert W. Prelcic, Nuria Gonzalez Key Lab of Embedded System and Service Computing Tongji University Shanghai China University of Texas at Austin AustinTX United States University of Vigo Vigo Pontevedra Spain

ISBN: (纸本)9783800741779

Millimeter wave communication provides high data rates thanks to large arrays at the transmitter and receiver, coupled with large bandwidth channels. Exploiting the arrays is challenging due to the need to configure precoding at the transmitter based on the large frequency selective channel. In this paper we exploit the power iteration principle and propose a robust analog precoding training algorithm that can be applied in both frequency division duplex transmission (FDD) systems and time division duplex transmission (TDD) systems with or without RF calibration. We further analyze the convergence of the proposed algorithm and show how it converges to the singular value decomposition optimality exponentially. We propose null space projection on top of the power iteration to form multiple orthogonal beams at the transmitter and receiver. Strongest tap selection with proper energy pruning is used to collect as much precoding gain as possible from a frequency selective fading channel. The exponential effective SINR mapping performance is evaluated and demonstrates that the overall approach works effectively. Numerical simulation results demonstrate algorithm robustness and the algorithm works effectively not only for the simplified millimeter wave directional channels, but also for more general rich scattering channels. © VDE VERLAG GMBH · Berlin · Offenbach, Germany.

关键词： Millimeter waves

Learning How to Avoiding Obstacles for End-to-End Driving with Conditional Imitation Learning 19

学校读者我要写书评

暂无评论

Learning How to Avoiding Obstacles for End-to-End Driving wi...

Proceedings of the 2019 2nd International Conference on Signal Processing and Machine Learning

作者： Enwei Zhang Hongtu Zhou Yongchao Ding Junqiao Zhao Chen Ye The Key Laboratory of Embedded System and Service Computing Ministry of Education Tongji University School of Electronics and Information Engineering Shanghai P. R. China Tongji University School of Electronics and Information Engineering Shanghai P. R. China Tongji University School of Electronics and Information Engineering P. R. China The Key Laboratory of Embedded System and Service Computing Ministry of Education Tongji University School of Electronics and Information Engineering P. R. China

ISBN: (纸本)9781450372213

Obstacle avoiding is one of the most complex tasks for autonomous driving systems, which was also ignored by many cutting-edge end-to-end learning-based methods. The difficulties stem from the integrated process of detection and interpretation of environment and obstacles and generation of proper behaviors. We make the use of CARLA, a simulator for autonomous driving research, and collect massive human drivers' reactions to obstacles on road subjecting to given driving commands, i.e. follow, go straight, turn left and turn right for about 6 hours. A behavior-Cloning neural network architecture is proposed with the modified loss that enlarge the effects of errors for steer, which indicates the benefit to high an accuracy. We found the data augmentation of the image is crucial to the training of the proposed network. And a reasonable limit allows avoiding unexpected stop. The experiments demonstrate 3 obstacle avoidance cases: for the same type as the training dataset, other automobile and two-wheeled vehicles. Finally, the CARLA benchmark is also tested.

关键词： imitation learning obstacle avoidance end-to-end autonomous vehicles

Low-rank subspace representation from optimal coded-aperture for unsupervised classification of hyperspectral imagery

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Zhu, Jianchen Zhang, Tong Zhao, Shengjie Key Laboratory of Embedded System and Service Computing Ministry of Education Tongji University Shanghai201804 China School of Software Engineering Tongji University Shanghai201804 China School of Electronic and Information Engineering Tongji University Shanghai201804 China

This paper aims at developing a clustering approach with spectral images directly from the compressive measurements of coded aperture snapshot spectral imager (CASSI). Assuming that compressed measurements often lie approximately in low dimensional subspaces corresponding to multiple classes, state of the art methods generally obtains optimal solution for each step separately but cannot guarantee that it will achieve the globally optimal clustering results. In this paper, a low-rank subspace representation (LRSR) algorithm is proposed to perform clustering on the compressed measurements. In addition, a subspace structured norm is added into the objective of low-rank representation problem exploiting the fact that each point in a union of subspaces can be expressed as a sparse linear combination of all other points and that the matrix of the points within each subspace is low rank. Simulation with real dataset illustrates the accuracy of the proposed spectral image clustering approach. Copyright © 2020, The Authors. All rights reserved.

关键词： Spectroscopy

Non-Intrusive Load Disaggregation Using Semi-Supervised Learning Method

学校读者我要写书评

暂无评论

Non-Intrusive Load Disaggregation Using Semi-Supervised Lear...

International Conference on Security, Pattern Analysis, and Cybernetics (SPAC)

作者： Nan Miao Shengjie Zhao Qingjiang Shi Rongqing Zhang The Key Laboratory of Embedded System and Service Computing of Ministry of Education Tongji University Shanghai China School of Software Engineering The Key Laboratory of Embedded System and Service Computing of Ministry of Education Tongji University Shanghai China School of Software Engineering Tongji University Shanghai China

ISBN: (数字)9781728159287

ISBN: (纸本)9781728159294

With the emerging of smart metering around the world, there is a growing demand to analyse the residential energy usage. In this paper, we propose a Deep Neural Network (DNN)-based approach for non-intrusive load monitoring (NILM), which can achieve effective and efficient estimation of individual appliance usage according to a single main meter reading in a non-intrusive manner. Considering practical situations, two training methods are provided. The first training approach is fully supervised learning, which requires a ground truth of label, indicating the state of the appliance (ON/OFF), to build a prediction model. The second training approach is semi-supervised learning, leading to better performance by F-Measure metric while only requiring some more unlabelled training data. Experimental results on the low-sample rate REDD dataset demonstrate the superior performance of our proposed DNN-based method compared with Hidden Markov Model (HMM)based and Graph Signal Processing (GSP)-based approaches.

关键词： Training Hidden Markov models Home appliances Neural networks Convolution Security Pattern analysis