检索结果-内蒙古大学图书馆

arXiv 2018年

作者： Cheng, Yong Tu, Zhaopeng Meng, Fandong Zhai, Junjie Liu, Yang Tencent AI Lab China State Key Laboratory of Intelligent Technology and Systems Beijing National Research Center for Information Science and Technology Department of Computer Science and Technology Tsinghua University Beijing Advanced Innovation Center for Language Resources Beijing China

Small perturbations in the input can severely distort intermediate representations and thus impact translation quality of neural machine translation (NMT) models. In this paper, we propose to improve the robustness of NMT models with adversarial stability training. The basic idea is to make both the encoder and decoder in NMT models robust against input perturbations by enabling them to behave similarly for the original input and its perturbed counterpart. Experimental results on Chinese-English, English-German and English-French translation tasks show that our approaches can not only achieve significant improvements over strong NMT systems but also improve the robustness of NMT models. Copyright © 2018, The Authors. All rights reserved.

关键词： Neural machine translation

来源：评论

学校读者我要写书评

暂无评论

Texture classification in extreme scale variations using GANet

arXiv

引用

arXiv 2018年

作者： Liu, Li Chen, Jie Zhao, Guoying Fieguth, Paul Chen, Xilin Pietikäinen, Matti Center for Machine Vision and Signal Analysis University of Oulu Oulu90014 Finland Department of Systems Design Engineering University of Waterloo WaterlooONN2L 3G1 Canada Key Lab of Intelligent Information Processing of CAS Institute of Computing Technology CAS Beijing100190 China

research in texture recognition often concentrates on recognizing textures with intraclass variations such as illumination, rotation, viewpoint and small scale changes. In contrast, in real-world applications a change in scale can have a dramatic impact on texture appearance, to the point of changing completely from one texture category to another. As a result, texture variations due to changes in scale are amongst the hardest to handle. In this work we conduct the first study of classifying textures with extreme variations in scale. To address this issue, we first propose and then reduce scale proposals on the basis of dominant texture patterns. Motivated by the challenges posed by this problem, we propose a new GANet network where we use a Genetic Algorithm to change the units in the hidden layers during network training, in order to promote the learning of more informative semantic texture patterns. Finally, we adopt a FV-CNN (Fisher Vector pooling of a Convolutional Neural Network filter bank) feature encoder for global texture representation. Because extreme scale variations are not necessarily present in most standard texture databases, to support the proposed extreme-scale aspects of texture understanding we are developing a new dataset, the Extreme Scale Variation Textures (ESVaT), to test the performance of our framework. It is demonstrated that the proposed framework significantly outperforms gold-standard texture features by more than 10% on ESVaT. We also test the performance of our proposed approach on the KTHTIPS2b and OS datasets and a further dataset synthetically derived from Forrest, showing superior performance compared to the state of the art. Copyright © 2018, The Authors. All rights reserved.

关键词： Textures

来源：评论

学校读者我要写书评

暂无评论

An Improved FTRLS Filtering Algorithm and Its Simulation Analysis

An Improved FTRLS Filtering Algorithm and Its Simulation Ana...

引用

2017 17th IEEE International Conference on Communication Technology (ICCT 2017)

作者： Jun Zhu Jingjing Zhang Qiang Chen Key Lab of Intelligent Computing and Signal Processing Ministry of Education China School of Electronics and Information Engineering Anhui University

Due to the large error of LMS algorithm and the slow convergence rate, the recursive least squares(RLS) algorithm is proposed. Although the recursive estimation error is greatly reduced, the convergence rate is one order of magnitude higher than that of the general LMS filter. When the order N increases, the amount of calculation for a single iteration of the RLS algorithm is increased significantly. Aiming at these problems, this paper proposes an improved FTRLS filtering algorithm, which is to find out the amount of large error and accumulate the error, and then make the error feedback to make the algorithm more stable. The analysis of MATlab simulation results show that the improved algorithm can improve the convergence speed and stability of the algorithm, and effectively reduce the convergence of the noise.

关键词： adaptive filtering Least Mean Squares(LMS) Recursive Least Squares(RLS) Fast Transverse Recursive Least Squares(FTRLS) improved FTRLS algorithm convergence rate

来源：评论

学校读者我要写书评

暂无评论

Sparse DNNs with improved adversarial robustness 18

Sparse DNNs with improved adversarial robustness

引用

Proceedings of the 32nd International Conference on Neural information processing Systems

作者： Yiwen Guo Chao Zhang Changshui Zhang Yurong Chen Intel Labs China Academy for Advanced Interdisciplinary Studies Center for Data Science Peking University Institute for Artificial Intelligence Tsinghua University (THUAI) State Key Lab of Intelligent Technologies and Systems Beijing National Research Center for Information Science and Technology (BNRis) Department of Automation Tsinghua University

Deep neural networks (DNNs) are computationally/memory-intensive and vulnerable to adversarial attacks, making them prohibitive in some real-world applications. By converting dense models into sparse ones, pruning appears to be a promising solution to reducing the computation/memory cost. This paper studies classification models, especially DNN-based ones, to demonstrate that there exists intrinsic relationships between their sparsity and adversarial robustness. Our analyses reveal, both theoretically and empirically, that nonlinear DNN-based classifiers behave differently under l2 attacks from some linear ones. We further demonstrate that an appropriately higher model sparsity implies better robustness of nonlinear DNNs, whereas over-sparsified models can be more difficult to resist adversarial examples.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Anisotropic energy accumulation for stereoscopic image seam carving

Anisotropic energy accumulation for stereoscopic image seam ...

引用

International Conference on 3D Imaging (IC3D)

作者： Yuqing Lin Yuzhen Niu Shuai Zhang Junhao Chen College of Mathematics and Computer Science Fuzhou University Fujian Provincial Key Lab of the Network Computing and Intelligent Information Processing Fujian China

Seam carving is an image resizing method that aims at adapting the image to various display screens while reducing the distortion as much as possible. Severe visual distortion may be introduced by repeated removal or insertion of seams within a concentrated region of the image. To reduce such visual distortion, we propose a new AESSC method. First, we distribute the energy of each pixel on the seam to its adjacent 8-connected pixels when removing or inserting a seam. Second, since the information of each pixel is anisotropic, we use the Sobel operator to detect the direction that has the maximum edge information and continue the energy accumulation along this direction. Besides, we incorporate a 3D structure consistency constraint in the energy function and adopt a pixel visibility maintenance method. Experimental results show that the proposed method can effectively reduce visual distortion for stereo images while maintaining the geometric consistency.

关键词： Distortion Visualization Three-dimensional displays Stereo image processing Image edge detection Histograms Maintenance engineering

来源：评论

学校读者我要写书评

暂无评论

Unscented Kalman filter for arbitrary step randomly delayed measurements

Unscented Kalman filter for arbitrary step randomly delayed ...

引用

Indian Control Conference (ICC)

作者： Ajay Kumar Yadav Vikas Kumar Mishra Abhinoy Kumar Singh Shovan Bhaumik Ministry of Education Key Lab of Intelligent Computing and Signal Processing China School of Electronics and Information Engineering Anhui University Hefei China

ISBN: (纸本)9781509017966

The conventional Bayesian framework of filtering is based on the assumption that the measurements are available at each time-step without any delay. But in real-life problems, measurements may be randomly delayed in time. In this paper, we modified the unscented Kalman filter (UKF) for arbitrary time delayed measurements. With the help of simulation results, it has been shown that the proposed filter provides more accurate estimation compared to the ordinary UKF in presence of randomly delayed measurements.

关键词： Delays Kalman filters Noise measurement Bayes methods Mathematical model Estimation

来源：评论

学校读者我要写书评

暂无评论

Color correction for stereoscopic image based on matching and optimization

Color correction for stereoscopic image based on matching an...

引用

International Conference on 3D Imaging (IC3D)

作者： Xiaohua Zheng Yuzhen Niu Junhao Chen Yuzhong Chen College of Mathematics and Computer Science Fuzhou University Fujian Provincial Key Lab of the Network Computing and Intelligent Information Processing Fujian China

Stereoscopic 3D (S3D) image color correction is a major issue in the field of image processing. However, existing color correction algorithms have limitations. Global color correction algorithms cannot handle local color discrepancies, and local color correction algorithms are sensitive to matching quality between reference and target images. In this study, we propose an S3D image color correction algorithm that combines global and local color information to correct color discrepancies between S3D images. Sparse feature matching usually generates only a few matching features, producing error correction results in some local regions. Our algorithm uses dense stereo matching and global color correction algorithms to initialize color values, and improves the local color smoothness and global color consistency of the resulting image, while maintaining the initial color in that image as much as possible. Experimental results show that our algorithm performs better than do five state-of-the-art color correction algorithms.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Low-complexity linear precoding for pilot contamination mitigation in multi-cell massive MIMO systems

Low-complexity linear precoding for pilot contamination miti...

引用

IEEE International Symposium on intelligent Signal processing and Communication Systems (ISPACS)

作者： Manman Feng Yaohua Xu Ministry of Education Key Lab of Intelligent Computing and Signal Processing China School of Electronics and Information Engineering Anhui University Hefei China

The pilot contamination is caused by non-orthogonal pilot sequences reuse in uplink, which affects the performance of Massive MIMO systems seriously, so it is necessary to mitigate pilot contamination. In this paper, we propose a pilot contamination precoding scheme to mitigate multi-cell pilot contamination. In the uplink, we propose a cell-defined training scheme, where the same pilot sequence is used in the same cell, and different cells use orthogonal pilot sequences, which eliminates inter-cell interference and introduces intra-cell interference artificially. In the downlink, we adopt Truncated Polynomial Expansion (TPE) precoding to reduce intra-cell interference, since the truncated polynomial of TPE precoding can replace the matrix inversion of Regularized Zero-Forcing (RZF) precoding, which reduces the complexity of RZF precoding and approximates the performance of RZF precoding by suitable truncation orders. Simulation results show the effectiveness of the proposed scheme.

关键词： Precoding Base stations Contamination Interference Partial transmit sequences Antennas Complexity theory

来源：评论

学校读者我要写书评

暂无评论

A Novel Compact Broadband Circularly Polarized Antenna for GNSS

A Novel Compact Broadband Circularly Polarized Antenna for G...

引用

IEEE Asia-Pacific Conference on Antennas and Propagation

作者： Aoyun Zhang Zhongxiang Zhang Wensheng Yu Xianliang Wu Key Lab of Intelligent Computing & Signal Processing Ministry of Education Anhui University School of Electronic and Information Engineering Hefei Normal University

ISBN: (纸本)9781538616093

A novel compact broadband circularly polarized (CP) antenna is proposed for use in the Global Navigation Satellite System (GNSS). Circular polarization for the presented antenna is achieved by introducing two L-shaped branches in the ground plane and inverted L-shaped microstrip fed. The antenna exhibits a wide impedance bandwidth of 44.9% (1.14-1.8GHz) for reflection coefficient<-10dB and an axial ratio (AR) bandwidth of 36.9% (1.15-1.67GHz) for AR<3dB. The CP bandwidth (reflection coefficient<-10dB and AR<3dB) can cover all GNSS frequency bands. Although simple in structure, the antenna can meet the requirements for handheld wireless terminals.

关键词： Broadband Circular polarization Compact GNSS

来源：评论

学校读者我要写书评

暂无评论

Isometrically subdivisible tetrahedra

引用

Computer Aided Drafting,Design and Manufacturing 2017年第2期27卷 24-31页

作者： Li Erbo Li Zhi Li Hua EON Reality Inc Liaoning Center for Disease Prevention and Control Key Lab of Intelligent Information Processing Institute of Computing TechnologyChinese Academy of Sciences University of Chinese Academy of Sciences

It is well known that a triangle can be divided by mid-point refinement into four sub-triangles with the same shape. Similarly, a tetrahedron can be parted into eight subtetrahedra, which are generally not uniform in shape. This paper proves that there exist a set of tetrahedra, which is called isometrically subdivisible tetrahedra(IST) and can be divided into eight isometric subtetrahedra, including identical and reflection ones. And a new classification of tetrahedra is put forward, based on which all tetrahedra can be categorized into 26 classes according to both the number of maximum equal edges and topological relations. The IST belongs only to three of the classes. That result provides a new viewpoint of spatial structure and may be used to tile or subdivide space uniformly or isometrically.

关键词： regular subdivision isometric identical tetrahedra classification tetrahedra direction

来源：评论

学校读者我要写书评

暂无评论

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案：

请选择收藏分类：

通借通还

建议与咨询 留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

时间限定

文献类型

馆藏选择

核心期刊

语言

文献类型

帮助

文字说明：

检索规则说明：

检索范例：

分类表

所选分类

限定检索结果

文献类型

馆藏范围

日期分布

学科分类号

主题

机构

作者

语言

请选择保存的检索档案： 新增检索档案 确定 取消

请选择收藏分类： 新增自定义分类 确定 取消

通借通还

建议与咨询留下您的常用邮箱和电话号码，以便我们向您反馈解决方案和替代方法

请选择保存的检索档案：

请选择收藏分类：