检索结果-内蒙古大学图书馆

Forward and backward information retention for accurate binary neural networks

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Qin, Haotong Gong, Ruihao Liu, Xianglong Shen, Mingzhu Wei, Ziran Yu, Fengwei Song, Jingkuan State Key Lab of Software Development Environment Beihang University Beijing Advanced Innovation Center for Big Data-Based Precision Medicine Beihang University SenseTime Research Beijing University of Posts and Telecommunications Center for Future Media University of Electronic Science and Technology of China

Weight and activation binarization is an effective approach to deep neural network compression and can accelerate the inference by leveraging bitwise operations. Although many binarization methods have improved the accuracy of the model by minimizing the quantization error in forward propagation, there remains a noticeable performance gap between the binarized model and the full-precision one. Our empirical study indicates that the quantization brings information loss in both forward and backward propagation, which is the bottleneck of training accurate binary neural networks. To address these issues, we propose an Information Retention network (IR-Net) to retain the information that consists in the forward activations and backward gradients. IR-Net mainly relies on two technical contributions: (1) Libra Parameter Binarization (Libra-PB): simultaneously minimizing both quantization error and information loss of parameters by balanced and standardized weights in forward propagation;(2) Error Decay Estimator (EDE): minimizing the information loss of gradients by gradually approximating the sign function in backward propagation, jointly considering the updating ability and accurate gradients. We are the first to investigate both forward and backward processes of binary networks from the unified information perspective, which provides new insight into the mechanism of network binarization. Comprehensive experiments with various network structures on CIFAR-10 and ImageNet datasets manifest that the proposed IR-Net can consistently outperform state-of-the-art quantization methods. Copyright © 2019, The Authors. All rights reserved.

关键词： Chemical activation

Robust and Passive Motion Detection with COTS WiFi Devices

学校读者我要写书评

暂无评论

Tsinghua science and technology 2017年第4期22卷 345-359页

作者： Hai Zhu Fu Xiao Lijuan Sun Xiaohui Xie Panlong Yang Ruchuan Wang the College of Computer Nanjing University of Posts and Telecommunications the Jiangsu High Technology Research Key Laboratory for Wireless Sensor Networks Key Lab of Broadband Wireless Communication and Sensor Network Technology of Ministry of Education and College of Computer Nanjing University of Posts and Telecommunications the College of Computer Science and Technology University of Science and Technology of China

Device-free Passive（DfP） detection has received increasing attention for its ability to support various pervasive applications. Instead of relying on variable Received Signal Strength（RSS）, most recent studies rely on finer-grained Channel State Information（CSI）. However, existing methods have some limitations, in that they are effective only in the Line-Of-Sight（LOS） or for more than one moving individual. In this paper, we analyze the human motion effect on CSI and propose a novel scheme for Robust Passive Motion Detection（R-PMD）. Since traditional low-pass filtering has a number of limitations with respect to data denoising, we adopt a novel Principal Component Analysis（PCA）-based filtering technique to capture the representative signals of human motion and extract the variance profile as the sensitive metric for human detection. In addition, existing schemes simply aggregate CSI values over all the antennas in MIMO systems. Instead, we investigate the sensing quality of each antenna and aggregate the best combination of antennas to achieve more accurate and robust detection. The R-PMD prototype uses off-the-shelf WiFi devices and the experimental results demonstrate that R-PMD achieves an average detection rate of 96.33% with a false alarm rate of 3.67%.

关键词： device-free passive detection Received Signal Strength（RSS） channel state information MIMO

Advanced Information technology and Global Business Competition: Preface for ITQM 2018

学校读者我要写书评

暂无评论

Procedia Computer science 2018年 139卷 1-8页

作者： Shi, Yong Wolcott, Peter Kwak, Wikil Chen, Zhengxin Tian, Yingjie Lee, Heeseok College of Information Science and Technology University of Nebraska at Omaha OmahaNE68182 United States School of Economics and Management University of Chinese Academy of Sciences Key Lab of Big Data Mining and Knowledge Management Chinese Academy of Sciences Beijing100190 China College of Business Administration University of Nebraska at Omaha OmahaNE68182 United States College of Business Korea Advanced Institute Science and Technology Dongdaemum-gu Seoul130-722 Korea Republic of

A Fatigue Driving Detection Method based on Deep Learning and Image Processing

学校读者我要写书评

暂无评论

Journal of Physics: Conference Series 2020年第1期1575卷

作者： Zhong Wang Peibei Shi Chao Wu School of Computer Science and Technology Hefei Normal University No. 1688 Lianhua Road Hefei Anhui China. Anhui Province Key Laboratory of Big Data Analysis and Application University of Science and Technology of China No. 96 Jinzhai Road Hefei Anhui China. Network and Information Center University of Science and Technology of China No. 96 Jinzhai Road Hefei Anhui China

Driving fatigue is one of the important causes of traffic accidents. It is of great significance to study fatigue driving detection algorithms to improve human life and property safety. This paper proposes a fatigue driving detection method based on deep learning and image processing. First, the driver's face image is obtained in real time through the camera and the face image is detected using the MTCNN model. Next the image processing is performed on the face image, including three steps: grayscale processing, binarization processing, and human eye detection. Then we check the legitimacy of the human eye image and calculate the eye closure rate, and finally use the PERCLOS principle to analyze the fatigue state of the driver. The experimental results show that the proposed method has high detection rate and low false alarm rate, and has strong practicality.

关键词：

On the relationship between energy complexity and other boolean function measures

学校读者我要写书评

暂无评论

arXiv 2018年

作者： Sun, Xiaoming Sun, Yuan Wu, Kewen Xia, Zhiyu CAS Key Lab of Network Data Science and Technology Institute of Computing Technology Chinese Academy of Sciences Beijing China University of Chinese Academy of Sciences Beijing China School of Electronics Engineering and Computer Science Peking University China

In this work we investigate energy complexity, a Boolean function measure related to circuit complexity. Given a circuit C over the standard basis {∨2, ∧2, ¬}, the energy complexity of C, denoted by EC(C), is the maximum number of its activated inner gates over all inputs. The energy complexity of a Boolean function f, denoted by EC(f), is the minimum of EC(C) over all circuits C computing f. Recently, K. Dinesh et al. [3] gave EC(f) an upper bound in terms of the decision tree complexity, EC(f) = O(D(f)3). They also showed that EC(f) ≤ 3n−1, where n is the input size. For the lower bound, they show that EC(f) ≥ 13 psens(f), where psens(f) is the positive sensitivity. They asked whether EC(f) can be lower bounded by a polynomial of D(f). We improve both the upper and lower bounds in this paper. For upper bounds, We show that EC(f) ≤ min{ 12 D(f)2 + O(D(f)), n + 2 D(f) − 2}. For the lower bound, we answer Dinesh et al.'s question by proving that EC(f) = Ω(pD(f)). For non-degenerated functions, we also give another lower bound EC(f) = Ω(log n) where n is the input size. These two lower bounds are incomparable to each other. Besides, we examine the energy complexity of OR functions and ADDRESS functions, which implies the tightness of our two lower bounds respectively. In addition, the former one answers another open question in [3] asking for non-trivial lower bound for energy complexity of OR functions. Copyright © 2018, The Authors. All rights reserved.

关键词： Decision trees

A generalization theory based on independent and task-identically distributed assumption

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Zheng, Guanhua Sang, Jitao Li, Houqiang Yu, Jian Xu, Changsheng University of Science and Technology of China School of Computer and Information Technology Beijing Key Laboratory of Traffic Data Analysis and Mining Beijing Jiaotong University Beijing100044 China Chinese Academy of Sciences Key Laboratory of Technology in Geo-Spatial Information Processing and Application System Hefei230026 China National Lab of Pattern Recognition Institute of Automation CAS Beijing100190 China University of Chinese Academy of Sciences

—Existing generalization theories analyze the generalization performance mainly based on the model complexity and training process. The ignorance of the task properties, which results from the widely used IID assumption, makes these theories fail to interpret many generalization phenomena or guide practical learning tasks. In this paper, we propose a new Independent and Task-Identically Distributed (ITID) assumption, to consider the task properties into the data generating process. The derived generalization bound based on the ITID assumption identifies the significance of hypothesis invariance in guaranteeing generalization performance. Based on the new bound, we introduce a practical invariance enhancement algorithm from the perspective of modifying data distributions. Finally, we verify the algorithm and theorems in the context of image classification task on both toy and real-world datasets. The experimental results demonstrate the reasonableness of the ITID assumption and the effectiveness of new generalization theory in improving practical generalization performance. Copyright © 2019, The Authors. All rights reserved.

关键词： Classification (of information)

A comprehensive analysis of video service quality on IQIYI from large-scale data sets 13th

学校读者我要写书评

暂无评论

A comprehensive analysis of video service quality on IQIYI f...

13th International Conference on Heterogeneous networking for Quality, Reliability, Security and Robustness, QShine 2017

作者： Guo, Yao Lv, Qiujian Liu, Fang Yang, Jie Gao, Zhe Beijing Laboratory of Advanced Information Networks Beijing Key Laboratory of Network System Architecture and Convergence Center for Data Science BUPT Beijing100876 China Institute of Information Engineering Chinese Academy of Sciences Beijing100093 China Technology Research Institute Aisino Corporation Beijing100086 China

ISBN: (纸本)9783319780771

With the proliferation of online video, measuring the quality of the video service has become a vital aspect for improving user’s experience. Recent work shows that measurable quality metrics such as buffering, bitrate, and video resolutions impact user’s experience, but none of them reveal the real relationships between these metrics and user’s actual experience. This paper attempts to solve the problem above. We use IQIYI as the sample, and our large-scale dataset consists of 7 days real Internet traffic in a northern city of China. We quantify user’s experience at per-video level (or view). Using Apache Spark, we extract some video events and calculate several quality metrics. In order to investigate the relationship between the metrics and user’s experience, we use the FP-Growth algorithm to mining the implicit association rules and get some interesting results. © ICST Institute for Computer sciences, Social Informatics and Telecommunications Engineering 2018.

关键词： Large dataset

Author Correction: Enhanced oxygen reduction with single-atomic-site iron catalysts for a zinc-air battery and hydrogen-air fuel cell

学校读者我要写书评

暂无评论

Nature communications 2022年第1期13卷 921页

作者： Yuanjun Chen Shufang Ji Shu Zhao Wenxing Chen Juncai Dong Weng-Chon Cheong Rongan Shen Xiaodong Wen Lirong Zheng Alexandre I Rykov Shichang Cai Haolin Tang Zhongbin Zhuang Chen Chen Qing Peng Dingsheng Wang Yadong Li Department of Chemistry Tsinghua University 100084 Beijing China. Beijing Guyue New Materials Research Institute Beijing University of Technology 100124 Beijing China. Beijing Synchrotron Radiation Facility Institute of High Energy Physics Chinese Academy of Sciences 100049 Beijing China. State Key Laboratory of Coal Conversion Institute of Coal Chemistry Chinese Academy of Sciences 030001 Taiyuan China. Mössbauer Effect Data Center Dalian Institute of Chemical Physics Chinese Academy of Sciences 116023 Dalian China. State Key Laboratory of Advanced Technology for Materials Synthesis and Processing Wuhan University of Technology 430070 Wuhan China. State Key Lab of Organic-Inorganic Composites and Beijing Advanced Innovation Center for Soft Matter Science and Engineering Beijing University of Chemical Technology 100029 Beijing China. Department of Chemistry Tsinghua University 100084 Beijing China. wangdingsheng@***.

Exploring Structural Sparsity of Deep networks via Inverse Scale Spaces

学校读者我要写书评

暂无评论

arXiv 2019年

作者： Fu, Yanwei Liu, Chen Li, Donghao Zhong, Zuyuan Sun, Xinwei Zeng, Jinshan Yao, Yuan School of Data Science Shanghai Key Lab of Intelligent Information Processing Fudan University Fudan ISTBI-ZJNU Algorithm Centre for Brain-Inspired Intelligence Zhejiang Normal University Jinhua China The School of Computer and Information Engineering Jiangxi Normal University China Hong Kong University of Science and Technology Hong Kong

The great success of deep neural networks is built upon their over-parameterization, which smooths the optimization landscape without degrading the generalization ability. Despite the benefits of over-parameterization, a huge amount of parameters makes deep networks cumbersome in daily life applications. On the other hand, training neural networks without over-parameterization faces many practical problems, e.g., being trapped in the local optimal. Though techniques such as pruning and distillation are developed, they are expensive in fully training a dense network as backward selection methods;and there is still a void on systematically exploring forward selection methods for learning structural sparsity in deep networks. To fill in this gap, this paper proposes a new approach based on differential inclusions of inverse scale spaces. Specifically, our method can generate a family of models from simple to complex ones along the dynamics via coupling a pair of parameters, such that over-parameterized deep models and their structural sparsity can be explored simultaneously. This kind of differential inclusion scheme has a simple discretization, dubbed Deep structure splitting Linearized Bregman Iteration (DessiLBI), whose global convergence in learning deep networks could be established under the Kurdyka-Lojasiewicz framework. Particularly, we explore several applications of DessiLBI, including finding sparse structures of networks directly via the coupled structure parameter and growing networks from simple to complex ones progressively. Experimental evidence shows that our method achieves comparable and even better performance than the competitive optimizers in exploring the sparse structure of several widely used backbones on the benchmark datasets. Remarkably, with early stopping, our method unveils "winning tickets" in early epochs: the effective sparse network structures with comparable test accuracy to fully trained over-parameterized models, that are further t

关键词： Linearization