检索结果-内蒙古大学图书馆

Towards adaptive semantic segmentation by progressive feature refinement

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Zhang, Bin Zhao, Shengjie Zhang, Rongqing School of Software Engineering Tongji University Shanghai China Key Laboratory of Embedded System and Service Computing Tongji University Shanghai China

As one of the fundamental tasks in computer vision, semantic segmentation plays an important role in real world applications. Although numerous deep learning models have made notable progress on several mainstream datasets with the rapid development of convolutional networks, they still encounter various challenges in practical scenarios. Unsupervised adaptive semantic segmentation aims to obtain a robust classifier trained with source domain data, which is able to maintain stable performance when deployed to a target domain with different data distribution. In this paper, we propose an innovative progressive feature refinement framework, along with domain adversarial learning to boost the transferability of segmentation networks. Specifically, we firstly align the multi-stage intermediate feature maps of source and target domain images, and then a domain classifier is adopted to discriminate the segmentation output. As a result, the segmentation models trained with source domain images can be transferred to a target domain without significant performance degradation. Experimental results verify the efficiency of our proposed method compared with state-of-the-art methods. Copyright © 2020, The Authors. All rights reserved.

关键词： Semantic Segmentation

Adaptive finite time stability of delayed systems via aperiodically intermittent control and quantized control

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Liu, Xiwei Ma, Hailian Department of Computer Science and Technology Tongji University Key Laboratory of Embedded System and Service Computing Ministry of Education Shanghai201804 China

In this brief, we set up the finite time stability (FnTSta) theory for dynamical systems with bounded timevarying delays via aperiodically intermittent control (AIC) and quantized control (QC). A more general QC is designed in this brief. The bound of time-varying delay is required to be less than the infimum in AIC. Two-phases-method (2PM) is applied to solve the FnTSta for delayed system, i.e., it divides the whole proof process into two phases, one phase is that the process for the norm of system error evolving from initial values to 1, and the other is the process for the norm of system error evolving from 1 to 0. By proving that these two phases both use FnT to realize, the whole FnTSta for system is proved. We also design the adaptive rules and prove its validity rigorously. Furthermore, the obtained theories are used to discuss the finite time synchronization (FnTSyn) of neural networks as an application. Finally, some simulations are given to illustrate the effectiveness of our theoretical result. Copyright © 2020, The Authors. All rights reserved.

关键词： Dynamical systems

Adaptive finite time stability of delayed systems with applications to network synchronization

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Liu, Xiwei Department of Computer Science and Technology Tongji University Key Laboratory of Embedded System and Service Computing Ministry of Education Shanghai201804 China

The finite time stability (FnTSta) theory of delayed systems has not been set up until now. In this paper, we propose a two-phases-method (2PM), to achieve this object. In the first phase, we prove that the time for norm of system error evolving from initial values to 1 is finite;then in the second phase, we prove that the time for norm of the error evolving from 1 to 0 is also finite, thus FnTSta is obtained. Considering the cost and complexity of controller, we use only two simple terms to realize this aim. For the proposed 2PM, time delays can be the same or asynchronous, bounded or unbounded, etc;the norm can be 2-norm, 1-norm, ∞-norm, which show that 2PM is powerful and has a wide scope of applications. Furthermore, we also prove the adaptive finite time stability (AFnTSta) theory of delayed systems using 2PM. As an application of the obtained FnTSta theory, we consider the finite time outer synchronization (FnTOSyn) for complex networks with asynchronous unbounded time delays, corresponding criteria are also obtained. Finally, two numerical examples are given to demonstrate the validity of our theories. Copyright © 2020, The Authors. All rights reserved.

关键词： Synchronization

Global µ-stability and finite-time control of octonion-valued neural networks with unbounded delays✡

学校读者我要写书评

暂无评论

arXiv 2020年

作者： Wang, Jingzhu Liu, Xiwei Department of Computer Science and Technology Tongji University Key Laboratory of Embedded System and Service Computing Ministry of Education Shanghai201804 China

Octonion-valued neural networks (OVNNs) are a type of neural networks for which the states and weights are octonions. In this paper, the global µ-stability and finite-time stability problems for octonion-valued neural networks are considered under unbounded and asynchronous time-varying delays. To avoid the non-communicative and non-associative multiplication feature of the octonions, we firstly decompose the OVNNs into eight real-valued neural networks (RVNNs) equivalently. Through the use of generalized norm and the Cauchy convergence principle, we obtain the sufficient criteria which assure the existence, uniqueness of the equilibrium point and global µ-stability of OVNNs. By adding controllers, the criteria to ensure the finite-time stability for OVNNs are presented by dividing the analysis of finite-time stability process into two phases. Furthermore, we also prove the adaptive finite time stability theory of above networks. At last, the simulation results of specified examples is given to substantiate the effectiveness and correctness of the theoretical results. Copyright © 2020, The Authors. All rights reserved.

关键词： Timing circuits

A hierarchical clustering strategy of processing class imbalance and its application in fraud detection 21

学校读者我要写书评

暂无评论

A hierarchical clustering strategy of processing class imbal...

21st IEEE International Conference on High Performance computing and Communications, 17th IEEE International Conference on Smart City and 5th IEEE International Conference on Data Science and systems, HPCC/SmartCity/DSS 2019

作者： Zhang, Youjun Liu, Guanjun Zheng, Lutao Yan, Chungang Key Laboratory of the Ministry of Education for Embedded System and Service Computing Shanghai Electronic Transactions and Information Service Collaborative Innovation Center Tongji University Shanghai China

ISBN: (纸本)9781728120584

With the Internet and mobile communications becoming an indispensable part of people's daily lives, online transactions have become one of the most common payment methods. However, transaction fraud incidents also occur frequently, causing a larger number of economic losses. Therefore, transaction fraud detection is significant. The methods of machine learning are often used to detect fraudulent transactions from a larger amount of transaction data. However, the class imbalance problem reduces the performance of these methods. There are mainly four factors causing this problem: imbalanced class distribution, sample size, class separability and within-class concept. The existing improvement strategies for class imbalance mainly focus on the first factor but omit other three ones. This paper considers the four factors to propose a comprehensive model called clustering tree. Constructing a clustering tree includes two steps: 1) we first select a clustering algorithm considering the class separability;and then 2) we use this clustering algorithm to construct a tree that can be used to determine if an incoming transaction is illegal. The root node of this tree contains all samples, and we consider both the imbalanced class distribution and the within-class concept when samples are hierarchically divided into sub-nodes during the constructing process. We compare the proposed method with five state-of-the-art ones on two real transaction datasets, and the experimental results show that our method works better. © 2019 IEEE.

关键词： Clustering algorithms

An Accuracy-and-Diversity-based Ensemble Method for Concept Drift and Its application in Fraud Detection

学校读者我要写书评

暂无评论

An Accuracy-and-Diversity-based Ensemble Method for Concept ...

IEEE International Conference on Data Mining Workshops (ICDM Workshops)

作者： Shujie Yin Guanjun Liu Zhenchuan Li Chungang Yan Changjun Jiang Key Laboratory of the Ministry of Education for Embedded System and Service Computing Shanghai Electronic Transactions and Information Service Collaborative Innovation Center Tongji University Shanghai China

ISBN: (数字)9781728190129

ISBN: (纸本)9781728190136

Concept Drift is one of the most challenging issues in various applications such as fraud detection, spam filtering and sensor networks, causing the performance degradation of learning algorithms which are dedicated to processing static data such as Naive Bayes, due to their poor adaptability to changes. Several techniques and algorithms have been proposed to address the concept drift problem, e.g. the ensemble methods that have already achieved great success. However, most of the ensemble methods consider either accuracy or diversity of a trained model when adapting to new concepts, and thus they can hardly deal with different types of drifts. This paper proposes a novel ensemble method that considers both accuracy and diversity. We linearly combine accuracy and diversity with different coefficients, and propose an entropy-based policy to compute these coefficients. Experiments on 10 synthetic datasets and 4 public datasets demonstrate that our method outperforms the state-of-the-art ones on dealing with different concept drifts. Our method is also applied to the electronic transaction fraud detection and achieves an excellent performance.

关键词： Degradation Adaptation models Filtering Design methodology Conferences Computational modeling Data mining

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...

2019 International Conference on Security, Pattern Analysis, and Cybernetics, SPAC 2019

作者： Miao, Nan Zhao, Shengjie Shi, Qingjiang Zhang, Rongqing Tongji University Key Laboratory of Embedded System and Service Computing of Ministry of Education Shanghai China Tongji University Sch. of Software Eng. the Key Lab. of Embedded System and Service Computing of Ministry of Education Shanghai China Tongji University School of Software Engineering Shanghai China

ISBN: (纸本)9781728159287

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. © 2019 IEEE.

关键词： Deep neural networks

Detecting Data-flow Errors Based on Petri Nets With Data Operations

学校读者我要写书评

暂无评论

IEEE/CAA Journal of Automatica Sinica 2018年第1期5卷 251-260页

作者： Dongming Xiang Guanjun Liu Chungang Yan Changjun Jiang Key Laboratory of Embedded System and Service Computing Ministry of Education Tongji University Shanghai 201804 China IEEE

In order to guarantee the correctness of business processes, not only control-flow errors but also data-flow errors should be considered. The control-flow errors mainly focus on deadlock, livelock, soundness, and so on. However, there are not too many methods for detecting data-flow errors. This paper defines Petri nets with data operations(PN-DO) that can model the operations on data such as read, write and delete. Based on PN-DO, we define some data-flow errors in this paper. We construct a reachability graph with data operations for each PN-DO, and then propose a method to reduce the reachability graph. Based on the reduced reachability graph, data-flow errors can be detected rapidly. A case study is given to illustrate the effectiveness of our methods.

关键词： Business process modeling data-flow errors Petri nets reachability graph

Low-rate non-intrusive appliance load monitoring based on graph signal processing

学校读者我要写书评

暂无评论

Low-rate non-intrusive appliance load monitoring based on gr...

2019 International Conference on Security, Pattern Analysis, and Cybernetics, SPAC 2019

作者： Zhang, Bing Zhao, Shengjie Shi, Qingjiang Zhang, Rongqing Tongji University Key Laboratory of Embedded System and Service Computing of Ministry of Education Shanghai China Tongji University Sch. of Software Eng. the Key Lab. of Embedded System and Service Computing of Ministry of Education Shanghai China Tongji University School of Software Engineering Shanghai China

ISBN: (纸本)9781728159287

Thanks to the large-scale smart meters deployments around the world, non-intrusive appliance load monitoring (NILM) is receiving popularity. It aims to disaggregate the total electricity load of a home into individual appliances without resorting to any specific appliance power monitors. NILM is worthy of broad attention owing to its facilitation in energy savings. This paper regards NILM as a classification task and proposes a two-step method based on graph signal processing (GSP). In the first step, a smoothest solution is obtained by minimizing the regularization term. In the second step, gradient projection method, which uses the obtained minimizer as a start point, is adopted to optimize the while objective function, where NILM is regarded as a constrained nonlinear programming problem. The experiment results based on the open-access data set REDD clearly demonstrate that the proposed GSP-based method achieves improved performance compared with other state-of-the-art low-rate NILM approaches. © 2019 IEEE.

关键词： Signal processing