检索结果-内蒙古大学图书馆

29th IEEE International Conference on Parallel and distributed Systems, ICPADS 2023

作者： Feng, Jiaxin Zhu, Kun Zhang, Tong Nanjing University of Aeronautics and Astronautics Nanjing China

ISBN: (纸本)9798350330717

With the widespread application of deep learning (DL) technology in the modern Internet of Things (IoT) areas such as autonomous driving, smart cities and homes, embedded real-time systems are increasingly used at the edge of the network to complete various hybrid DNN tasks. Although embedded real-time systems are equipped with heterogeneous CPU and GPU cores to reduce the response time of inference jobs, the computing resources of heterogeneous devices are not fully utilized, and there is still plenty of room for schedulability to be improved. In this paper, we propose a layer-based hybrid deep neural network (DNN) tasks scheduling algorithm in embedded real-time systems (LHTS) that maps DNN layers to CPU and GPU devices and regulates their start time to avoid confliction. We evaluate LHTS through extensive simulations. The experimental results show that LHTS can achieve more sufficient use of heterogeneous CPU and GPU resources in embedded real-time systems, reduce the worst-case execution time and enhance the schedulability performance of hybrid DNN tasks. © 2023 IEEE.

关键词： deep neural network optimization real-time system resource allocation task scheduling

来源：评论

学校读者我要写书评

暂无评论

Fused feature extract method for 40-OTDR event recognition based on VGGish transfer learning

引用

APPLIED OPTICS 2024年第20期63卷 5411-5420页

作者： Gan, Jiaqi Xiao, Yueyu Zhang, Andong Shanghai Univ Key Lab Specialty Fiber Opt Opt Access Networks Sh Shanghai 200444 Peoples R China Shanghai Univ Joint Int Res Lab Specialty Fiber Opt & Adv Commun Shanghai 200444 Peoples R China Shanghai Univ Inst Fiber Opt Shanghai 200444 Peoples R China

Thanks to the development of artificial intelligence algorithms, the event recognition of distributed optical fiber sensing systems has achieved high classification accuracy on many deep learning models. However, the large-scale samples required for the deep learning networks are difficult to collect for the optical fiber vibration sensing systems in actual scenarios. An overfitting problem due to insufficient data in the network training process will reduce the classification accuracy. In this paper, we propose a fused feature extract method suitable for the small dataset of 40-OTDR systems. The high-dimensional features of signals in the frequency domain are extracted by a transfer learning method based on the VGGish framework. Combined with the characteristics of 12 different acquisition points in the space, the spatial distribution characteristics of the signal can be reflected. Fused with the spatial and temporal features, the features undergo a sample feature correction algorithm and are used in a SVM classifier for event recognition. Experimental results show that the VGGish, a pre-trained convolutional network for audio classification, can extract the knowledge features of 40-OTDR vibration signals more efficiently. The recognition accuracy of six types of intrusion events can reach 95.0% through the corrected multi-domain features when only 960 samples are used as the training set. The accuracy is 17.7% higher than that of the single channel trained on VGGish without fine-tuning. Compared to other CNNs, such as ResNet, the feature extract method proposed can improve the accuracy by at least 4.9% on the same dataset. (c) 2024 Optica Publishing Group. All rights, including for text and data mining (TDM), Artificial Intelligence (AI) training, and similar technologies, are reserved.

关键词： Deep learning Machine learning neural networks Optical fibers Pattern recognition Signal processing

来源：评论

学校读者我要写书评

暂无评论

TIPAngle: Traffic Tracking at City Scale by Pose Estimation of Pan and Tilt Intersection Cameras 38

TIPAngle: Traffic Tracking at City Scale by Pose Estimation ...

引用

38th International Conference on VLSI Design and International Conference on Embedded Systems

作者： Jagadeesha, Shreehari Andert, Edward Shrivastava, Aviral Arizona State Univ SCAI Tempe AZ 85287 USA

ISBN: (纸本)9798331522452;9798331522445

Modern cities have hundreds to thousands of traffic cameras distributed across them, many of them with the capability to pan and tilt, but very often these pan and tilt cameras either do not have angle sensors or do not provide camera orientation feedback. This makes it difficult to robustly track traffic using these cameras. Several methods to automatically detect the camera pose have been proposed in literature, with the most popular and robust being deep learning-based approaches. However, they are compute intensive, require large amounts of training data, and generally cannot be run on embedded devices. In this paper, we propose TIPAngle - a Siamese neural network, lightweight training, and a highly optimized inference mechanism and toolset to estimate camera pose and thereby improve traffic tracking even when operators change the pose of the traffic cameras. TIPAngle is 18.45x times faster and 3x more accurate in determining the angle of a camera frame than a ResNet-18 based approach. We deploy TIPAngle to a Raspberry Pi CPU and show that processing an image takes an average of.057s, equating to a frequency of about 17Hz on an embedded device.

关键词： Camera Pose Estimation Traffic Tracking Pan Tilt Traffic Camera Siamese neural network

来源：评论

学校读者我要写书评

暂无评论

ReHy: A ReRAM-Based Digital/Analog Hybrid PIM Architecture for Accelerating CNN Training

引用

IEEE TRANSACTIONS ON PARALLEL AND distributed SYSTEMS 2022年第11期33卷 2872-2884页

作者： Jin, Hai Liu, Cong Liu, Haikun Luo, Ruikun Xu, Jiahong Mao, Fubing Liao, Xiaofei Huazhong Univ Sci & Technol Natl Engn Res Ctr Big Data Technol & Syst Serv Comp Technol & Syst Lab Cluster & Grid Comp LabSch Comp Sci & Technol Wuhan 430074 Peoples R China

processing-In-Memory(PIM) has emerged as a high-performance and energy-efficient computing paradigm for accelerating convolutional neural network (CNN) applications. Resistive random access memory (ReRAM) has been widely used in PIM architectures due to its extremely high efficiency for accelerating matrix-vector multiplications through analog computing. However, because CNN training usually requires high-precision computation in the backward propagation (BP) stage, the limited precision of analog PIM accelerators impedes their adoption in CNN training. In this article, we propose ReHy, a hybrid PIM accelerator to support CNN training in ReRAM arrays. It is composed of Analog PIM (APIM) and Digital PIM (DPIM) modules. ReHy uses APIM to accelerate the feed-forward propagation (FP) stage for high performance, and DPIM to process the BP stage for high accuracy. We exploit the capability of ReRAM for Boolean logic operations to design the DPIM architecture. Particularly, we design floating-point multiplication and addition operators to support matrix multiplications in ReRAM arrays. We also propose a performance model to offload high-precision matrix multiplications to DPIM according to the data parallelism. Experimental results show that ReHy can speed up CNN training by 48.8x and 2.4x, and reduce energy consumption by 35.1x and 2.33x, compared with CPU/GPU architectures (baseline) and the state-of-the-art FloatPIM, respectively.

关键词： Training Convolutional neural networks Computer architecture Parallel processing Resistance Memristors Arrays Resistive random access memory convolutional neural network training digital-analog hybrid accelerator

来源：评论

学校读者我要写书评

暂无评论

Finite time fault tolerant tracking control for multiple unmanned aerial vehicles with actuator faults

引用

INTERNATIONAL JOURNAL OF ADAPTIVE CONTROL AND SIGNAL processing 2022年第12期36卷 3142-3160页

作者： Qian, Moshu Wu, Zhu Zhang, Zhen Bo, Cuimei Nanjing Tech Univ Coll Elect Engn & Control Sci Nanjing 211816 Peoples R China Nanjing Univ Aeronaut & Astronaut State Key Lab Mech & Control Mech Struct Nanjing Peoples R China

An adaptive fast nonsingular integral terminal sliding mode control scheme is proposed for the distributed formation control problem of multiple unmanned aerial vehicles (UAVs) affected by actuator faults and external disturbances. In order to achieve a good cooperative tracking performance of UAVs during mission execution, the adverse effect of the lumped disturbances composed of actuator failures, wind and vortex disturbances is estimated by the cerebellar model articulation neural network, meanwhile the tangent function is introduced to eliminate the chattering of the sliding mode. Then the analysis of superior convergence performance of the fast nonsingular integral terminal sliding manifold is implemented, and the finite time stability of the closed-loop formation flight control system is proved. Finally, the simulation experiments verify that the proposed approach is superior to the nonsingular integral terminal sliding mode control for the formation control system of faulty multiple UAVs.

关键词： cerebellar model articulation neural network nonsingular integral terminal sliding mode control unmanned aerial vehicle

来源：评论

学校读者我要写书评

暂无评论

Design of Real-time Object Recognition Algorithm Based on Deep Convolution neural network

Design of Real-time Object Recognition Algorithm Based on De...

引用

2023 International Conference on Internet of Things, Robotics and distributed Computing, ICIRDC 2023

作者： Wang, Shuqin Zibo Technician College Shandong Zibo255000 China

ISBN: (纸本)9798350375275

In recent years, with the rapid development of computer vision and deep learning technology, real-time object recognition has become more and more important in various application fields. This research is devoted to designing a real-time object recognition algorithm based on Deep Convolution neural network (DCNN), aiming at overcoming the trade-off between accuracy and real-time performance of traditional methods. In this study, a DCNN is trained by using a large-scale labeled data set, so that it can learn rich feature representation. By introducing the hierarchical structure of deep network, the abstract ability of the model to objects is improved, and its generalization performance is enhanced by data enhancement technology. Through careful design of network structure and parameter adjustment, the model performs well in real-time object recognition tasks. In order to optimize real-time processing, we adopt a series of acceleration and optimization strategies to ensure that the algorithm can run quickly and reliably in various environments. Experiments show that the proposed algorithm achieves superior performance on several public data sets. Compared with traditional methods, the accuracy is improved by more than 8.26%, while maintaining high efficiency in real-time. The real-time object recognition algorithm based on DCNN proposed in this study has important potential value in practical application, which provides new technical support for promoting the development and application of intelligent systems. © 2023 IEEE.

关键词： Intelligent systems

来源：评论

学校读者我要写书评

暂无评论

Dynamic Resource Partitioning for Multi-Tenant Systolic Array Based DNN Accelerator 31

Dynamic Resource Partitioning for Multi-Tenant Systolic Arra...

引用

31st Euromicro International Conference on Parallel, distributed and network-Based processing, PDP 2023

作者： Reshadi, Midia Gregg, David School of Computer Science and Statistics Lero Trinity College Dublin Dublin 2 Ireland

ISBN: (纸本)9798350337631

Deep neural networks (DNN) have become a significant applications in both cloud-server and edge devices. Meanwhile, the growing number of DNNs on those platforms raises the need to execute multiple DNNs on the same device. This paper proposes a dynamic partitioning algorithm to perform concurrent processing of multiple DNNs on asystolic-array-based accelerator. Sharing an accelerator's storage and processing resources across multiple DNNs increases resource utilization and reduces computation time and energy consumption. To this end, we propose a partitioned weight stationary dataflow with a minor modification in the logic of the processing element. We evaluate the energy consumption and computation time with both heavy and light workloads. Simulation results show a 35% and 62% improvement in energy consumption and 56% and 44% in computation time under heavy and light workloads, respectively, compared with single tenancy. © 2023 IEEE.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

Multi-scale Frequency Attention Fusion network for Optical and SAR Image Registration 17

Multi-scale Frequency Attention Fusion Network for Optical a...

引用

17th IEEE International Conference on Signal processing, ICSP 2024

作者： Hu, Xin Wu, Yan Li, Zhikang Yang, Zhifei Li, Ming Xidian University School of Electronic Engineering Xi'an China Xidian University National Key Laboratory of Radar Signal Processing Xi'an China

ISBN: (纸本)9798350387384

Image registration is a prerequisite for remote sensing image fusion and classification, and registration accuracy affects the performance of these tasks. However, there are significant nonlinear radiation differences between optical and synthetic aperture radar (SAR) images and speckle noise in SAR, which limits the ability of pure convolutional neural networks (CNNs) to obtain modality-shared features. To solve these problems, this paper proposes a Multi-scale Frequency Attention Fusion network for Optical and SAR image Registration (MFAF-OS-RegNet), which includes a Multi-scale Frequency Channel Attention Fusion (MFCAF) module and a Momentum Contrastive Learning (MCL) strategy. Firstly, we extract repeatable keypoints distributed on the edges and develop a novel MFCAF module to obtain modality-shared features. In the MFCAF module, multi-scale feature fusion is used to improve the positioning accuracy of matching points, and frequency channel attention is used to mine more favorable features. Secondly, we design a MCL strategy to reduce the interference of speckle noise in network training. The momentum teacher corrects handcrafted labels by learning the essential characteristics of the data for robust matching. Finally, we filter outliers and estimate transformation parameters based on the initial matching correspondence output by MFAF-OS-RegNet. Experimental results on a publicly available OS-dataset show that the proposed method outperforms other state-of-the-art methods. © 2024 IEEE.

关键词： Image fusion

来源：评论

学校读者我要写书评

暂无评论

A novel evaluation method for pavement distress based on impact of ride comfort

引用

INTERNATIONAL JOURNAL OF PAVEMENT ENGINEERING 2022年第3期23卷 638-650页

作者： Li, Yishun Liu, Chenglong Du, Yuchuan Jiang, Shengchuan Tongji Univ Minist Educ Key Lab Rd & Traff Engn Shanghai Peoples R China

Assessment of the pavement condition plays a significant role in pavement maintenance and driving comfort enhancement. Current evaluation methods primarily employ manual weights according to the geometric appearance of the distress, which makes it difficult to assess its depth or impact on passengers' experience. This paper proposes a data fusion-based method for pavement distress evaluation, which comprehensively considers the joint effect of distress physical appearance and the corresponding impact on riding comfort. A deep convolutional neural network was employed to automatically detect and locate the pavement distress using image data. A wavelet transform was applied to extract the acceleration effectuated by the defects in the frequency domain using vibration data. Finally, a comfort evaluation index was constructed based on the results of image and vibration data fusion. Furthermore, a mobile vehicle-mounted collective system was designed for rapid evaluation of the pavement distress, which integrated multiple distributed accelerometers, an industrial camera, and a graphics processing unit. The results demonstrated the stability and efficiency of the proposed approach, making it a potential tool to comprehensively evaluate the condition of pavement distress.

关键词： Pavement distress evaluation ride comfort data fusion convolutional neural network wavelet transform

来源：评论

学校读者我要写书评

暂无评论

CUTTANA: Scalable Graph Partitioning for Faster distributed Graph Databases and Analytics 51st

CUTTANA: Scalable Graph Partitioning for Faster Distributed ...

引用

51st International Conference on Very Large Data Bases, VLDB 2025

作者： Hajidehi, Milad Rezaei Sridhar, Sraavan Seltzer, Margo University of British Columbia Vancouver Canada

Graph partitioning plays a pivotal role in various distributed graph processing applications, including graph analytics, graph neural network training, and distributed graph databases. A "good" graph partitioner reduces workload execution time, worker imbalance, and network overhead. Graphs that require distributed settings are often too large to fit in the main memory of a single machine. This challenge renders traditional in-memory graph partitioners infeasible, leading to the emergence of streaming solutions. Streaming partitioners produce lower-quality partitions, because they work from partial information and must make premature decisions before they have a complete view of a vertex’s neighborhood. We introduce CUTTANA, a streaming graph partitioner that partitions massive graphs (Web/Twitter scale) with superior quality compared to existing streaming solutions. CUTTANA uses a novel buffering technique that prevents the premature assignment of vertices to partitions and a scalable coarsening and refinement technique that enables a complete graph view, improving the intermediate assignment made by a streaming partitioner. We implemented a parallel version for CUTTANA that offers nearly the same partitioning latency as existing streaming partitioners. Our experimental analysis shows that CUTTANA consistently yields better partitioning quality than state-of-the-art streaming vertex partitioners in terms of both edge-cut and communication volume metrics. We also evaluate the workload latencies that result from using CUTTANA and other partitioners in distributed graph analytics and databases. CUTTANA outperforms the other methods in most scenarios (algorithms, datasets). In analytics applications, CUTTANA improves runtime performance by up to 59% compared to various streaming partitioners (i.e., HDRF, Fennel, Ginger, HeiStream). In graph database tasks, CUTTANA results in higher query throughput by up to 23%, without hurting tail latency. © 2024, VLDB Endowment. All

关键词： Graph Databases

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：