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： Li, Xiaoyu Wu, Jingzheng Ling, Xiang Luo, Tianyue Wu, Yanjun Institute of Software Chinese Academy of Sciences Intelligent Software Research Center China University of Chinese Academy of Sciences China

ISBN: (纸本)9798350330717

In recent years, deep learning has been widely applied in the research of automatic program repair, namely, learning-based program repair, which treats the program repair task as either a neural machine translation problem or a transformation toward the abstract syntax tree of the program. However, all existing learning-based program repair solutions have not adequately modeled the deep semantic information of the programs. To be specific, those solutions based on neural machine translation do not involve any semantic information of the programs at all, while other solutions based on abstract syntax tree transformations also do not involve information such as control flow and data flow. Additionally, the neural networks they use are also unable to sufficiently learn the semantic information. As a result, they cannot fix some bugs that involve semantic *** this paper, we propose a new learning-based program repair approach called SGEPR. SGEPR uses a novel intermediate representation named sequence code property graph (SCPG) to model program semantic information, which can cover multiple types of semantic information in the program. SGEPR then utilizes a graph neural network based on the attention mechanism to learn the semantic information within SCPG. Finally, a well-designed prediction network is employed to generate the repair patches for the buggy program. To evaluate the performance of SGEPR, we conducted comparative experiments on a real-world dataset and the results show that SGEPR can achieve a top-3 accuracy of 56.85%, which is 15.69% higher than the state-of-the-art learning-based program repair method. © 2023 IEEE.

关键词： automatic program repair graph neural network program representation

来源：评论

学校读者我要写书评

暂无评论

SmartPipe: Intelligently Freezing Layers in Pipeline Parallelism for distributed DNN Training 29

SmartPipe: Intelligently Freezing Layers in Pipeline Paralle...

引用

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

作者： Niknami, Nadia Sawwan, Abdalaziz Wu, Jie Temple University Center for Networked Computing United States

ISBN: (纸本)9798350330717

Deep neural network (DNN) models have been widely utilized in various applications. However, the growing complexity of DNNs has led to increased challenges and prolonged training durations. Despite the availability of high-performance computing systems, certain DNNs still require several days for successful training. This study aims to address this issue by proposing a method for significantly reducing the training time of deep learning models while maintaining test accuracy. Existing approaches primarily concentrate on optimizing training efficiency through computational and communication overlap/scheduling. In contrast, this research takes a step further by inspiring transfer learning. Transfer learning is a useful way to quickly retrain a model on new data without having to retrain the entire network. During transfer learning, the first layers of the network are frozen while leaving the end layers open to modification. By doing so, computation and communication requirements in these frozen layers are eliminated. This intelligent approach involves freezing some of the specific DNN layers and allocating resources to the remaining active layers during the training process, thereby minimizing DNN training time. To achieve this objective, we propose an intelligently freezing DNN using pipeline parallelism. Through trace-based simulation results, our scheme has demonstrated its effectiveness in efficiently reducing the time cost of a training iteration. © 2023 IEEE.

关键词： DNN training gradient descent intelligent freezing machine learning neural networks pipeline

来源：评论

学校读者我要写书评

暂无评论

Fault Identification in Honda Scoopy 110CC Continuous Variable Transmission Using Backpropagation neural networks 12

Fault Identification in Honda Scoopy 110CC Continuous Variab...

引用

12th IEEE International Conference on Communication, networks and Satellite, COMNETSAT 2023

作者： Santoso, Rizky Dwi Pradana, Zein Hanni Wibisono, Gunawan Institute Technology Telkom Electrical Engineering Central Java Purwokerto Indonesia Institute Technology Telkom Telecommunications Engineering Central Java Purwokerto Indonesia

ISBN: (纸本)9798350341102

Intensive research in the field of signal processing has driven remarkable advancements in communication technology, particularly in the realm of voice recognition. Voice recognition concepts find application across various domains, with one such application being sound recognition within the context of Continuous Variable Transmission (CVT) for 110cc motor *** study aims to identify potential issues in CVT systems by employing artificial neural networks using Learning Predictive Coding (LPC), Mel Frequency Cepstral Coefficient (MFCC), the Artificial neural network (ANN) Backpropagation method to classify distinct sounds emanating from Honda Scoopy 110cc motor scooters. The dataset used comprises 100 CVT engine sound recordings, equally distributed between 50 samples of normal engine sounds and 50 samples of damaged engine *** research findings reveal the highest level of accuracy achieved with order 16 and 16 hidden neurons, resulting in a testing accuracy of 81.3%, a validation accuracy of 100.00%, and a testing accuracy of 90%. This data strongly supports the effectiveness of the backpropagation artificial neural network method for precise CVT issue identification. © 2023 IEEE.

关键词： Backpropagation

来源：评论

学校读者我要写书评

暂无评论

Selection of Deep neural network Models for IoT Anomaly Detection Experiments 29

Selection of Deep Neural Network Models for IoT Anomaly Dete...

引用

29th Euromicro International Conference on Parallel, distributed and network-Based processing (PDP)

作者： Gaifulina, Diana Kotenko, Igor Russian Acad Sci Russian Acad Sci SPC RAS St Petersburg Inst Informat & Automat St Petersburg Fed Res Ctr 14th Liniya St Petersburg Russia

ISBN: (纸本)9781665414555

This research is about selection of deep neural network models for anomaly detection in Internet of Things network traffic. We are experimentally evaluating deep neural network models using the same software, hardware and the same subsets of the UNSW-NB 15 dataset for training and testing. The assessment results are quality metrics of anomaly detection and the time spent on training models.

关键词： deepp learning deep neural networks anomaly detection Internet of Things information security

来源：评论

学校读者我要写书评

暂无评论

An Intelligent Wearable ECG Sensor in Intra-medical Virtual Chain network and Inter-medical Virtual Chain network

引用

SN Computer Science 2024年第4期5卷 329页

作者： Bhattarai, Adarsha Peng, Dongming Department of Electrical and Computer Engineering University of Nebraska Lincoln NE United States

Today, distributed medical network necessitates considering various dimensions of decentralized diagnosis environments to improve present wearable ECG sensors in terms of their versatility and security. This paper proposes five important mechanisms for the enhancement of ECG sensors. The proposed mechanisms are the partition of diagnosis, switch modes as per the communication status and severity, feedback channel, emergency alert system, and medical virtual chain (MVC). The proposed partition of diagnosis is based on the involvement of processing nodes according to the computational need. Two switch modes open-loop and closed-loop switch modes are put forward to deal with unpredictable communication scenarios and the severity of the patient. An MVC-based medical network is proposed where communication exists in and between medical organizations. The non-partitioned diagnosis used a CNN inference module with low computational complexity to give a diagnostic accuracy of 95.67% at the local ECG sensor, whereas the partitioned diagnosis was completed with an accuracy of 99.17% at the edge tier. In case of a poor communication scenario and low severity level of the patient, switch mode was activated to open loop to find that it was nearly six times more efficient than the regular transmission. The average delay observed for an intelligent sensor accessing a copy of MVC was only 0.3708 ms. Likewise, the average delay to record medical sessions initiated by the respective intelligent sensors in multiple medical institutions was 14.17 ms. These simulation results for the partition of diagnosis in multiple layers, switch mode initiation and MVC-based node interactions indicated that the proposed mechanisms for the intelligent ECG sensors are efficient and effective and hence can be implemented in a real-world scenario. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd 2024.

关键词： Convolutional neural network (CNN) Internet of medical things (IoMT) Medical virtual chain (MVC) Wearable electrocardiogram (ECG) sensor

来源：评论

学校读者我要写书评

暂无评论

PCM Enabled Low-Power Photonic Accelerator for Inference and Training on Edge Devices

PCM Enabled Low-Power Photonic Accelerator for Inference and...

引用

1st International Conference on Smart Energy Systems and Artificial Intelligence (SESAI)

作者： Curry, Juliana Louri, Ahmed Karanth, Avinash Bunescu, Razvan George Washington Univ Dept Elect & Comp Engn Washington DC 20052 USA Ohio Univ Sch Elect Engn & Comp Sci Athens OH 45701 USA Univ N Carolina Dept Comp Sci Charlotte NC USA

ISBN: (纸本)9798350364613;9798350364606

The convergence of edge computing and artificial intelligence requires that inference is performed on-device to provide rapid response with low latency and high accuracy without transferring large amounts of data to the cloud. However, power and size limitations make it challenging for electrical accelerators to support both inference and training for large neural network models. To this end, we propose Trident, a low-power photonic accelerator that combines the benefits of phase change material (PCM) and photonics to implement both inference and training in one unified architecture. Emerging silicon photonics has the potential to exploit the parallelism of neural network models, reduce power consumption and provide high bandwidth density via wavelength division multiplexing, making photonics an ideal candidate for on-device training and inference. As PCM is recoilfigurable and non-volatile, we utilize it for two distinct purposes: (i) to maintain resonant wavelength without expensive electrical or thermal heaters, and (ii) to implement non-linear activation function, which eliminates the need to move data between memory and compute units. This multi-purpose use of PCM is shown to lead to significant reduction in energy consumption and execution time. Compared to photonic accelerators DEAP-CNN, CrossLight, and PIXEL, Trident improves energy efficiency by up to 43% and latency by up to 150% on average. Compared to electronic edge AI accelerators Google Coral which utilizes the Google Edge TPU and Bearkey TB96-AL Trident improves energy efficiency by 11% and 93% respectively. While NVIDIA AGX Xavier is more energy efficient, the reduced data movement and GST activation of Trident reduce latency by 107% on average compared to the NVIDIA accelerator. When compared to the Google Coral and the Bearkey TB96-AI, Trident reduces latency by 1413% and 595% on average.

关键词： neural networks photonic accelerators phase change material training inference

来源：评论

学校读者我要写书评

暂无评论

Fake News Detection Platform-Conceptual Architecture and Prototype

引用

LOGIC JOURNAL OF THE IGPL 2022年第6期30卷 1005-1016页

作者： Kozik, Rafal Pawlicki, Marek Kula, Sebastian Choras, Michal Bydgoszcz Univ Sci & Technol PL-85796 Bydgoszcz Poland Bydgoszcz Univ Sci & Technol PL-85064 Bydgoszcz Poland UKW Bydgoszcz Kazimierz Wielki Univ Bydgoszcz Poland

Countering the fake news phenomenon has become one of the most important challenges for democratic societies, governments and non-profit organizations, as well as for the researchers coming from several domains. This is not a local problem and demands a holistic approach to analyzing heterogeneous data and storing the results. The research problem we face in this paper is the proposition of an innovative distributed architecture to tackle the above-mentioned problems. The architecture uses state-of-the-art technologies with a focus on efficiency, scalability and also openness, so that community-created components and digital content analyzers could be added. Moreover, we prove the usability of the prototype on Kaggle Fake News dataset. In particular, we consider different configurations of the proposed deep neural network and present the results reflecting the effectiveness, scalability and transferability of the proposed solution.

关键词： fake news detection natural language processing neural networks machine learning

来源：评论

学校读者我要写书评

暂无评论

Projective Synchronization Analysis of Fractional-Order neural networks With Mixed Time Delays

引用

IEEE TRANSACTIONS ON CYBERNETICS 2022年第7期52卷 6798-6808页

作者： Liu, Peng Kong, Minxue Zeng, Zhigang Zhengzhou Univ Light Ind Sch Elect & Informat Engn Zhengzhou 450002 Peoples R China Zhengzhou Univ Light Ind Henan Key Lab Informat Based Elect Appliances Zhengzhou 450002 Peoples R China Huazhong Univ Sci & Technol Sch Artificial Intelligence & Automat Wuhan 430074 Peoples R China Huazhong Univ Sci & Technol Key Lab Image Proc & Intelligent Control Educ Minist China Wuhan 430074 Peoples R China

In this article, we analyze the projective synchronization of fractional-order neural networks with mixed time delays. By introducing an extended Halanay inequality that is applicable for the case of fractional differential equations with arbitrary initial time and multiple types of delays, sufficient criteria are deduced for ensuring the projective synchronization of fractional-order neural networks with both discrete time-varying delays and distributed delays. Furthermore, sufficient criteria are presented for ensuring the projective synchronization in the Mittag-Leffler sense if there is no delay in fractional-order neural networks. The results derived herein include complete synchronization, anti-synchronization, and stabilization of fractional-order neural networks as particular cases. Moreover, the testable criteria in this article are a meaningful extension of projective synchronization of neural networks with mixed time delays from integer-order to fractional-order ones. A numerical simulation with four cases is provided to verify the validity of the obtained results.

关键词： Synchronization Delays Delay effects Biological neural networks Differential equations Image processing Fractional order mixed time delays neural network projective synchronization

来源：评论

学校读者我要写书评

暂无评论

SGCNAX: A Scalable Graph Convolutional neural network Accelerator With Workload Balancing

引用

IEEE TRANSACTIONS ON PARALLEL AND distributed SYSTEMS 2022年第11期33卷 2834-2845页

作者： Li, Jiajun Zheng, Hao Wang, Ke Louri, Ahmed George Washington Univ Dept Elect & Comp Engn Washington DC 20052 USA

Convolutional neural networks (GCNs) have emerged as promising tools for graph-based machine learning applications. Given that GCNs are both compute- and memory-intensive, this constitutes a major challenge for the underlying hardware to efficiently process large-scale GCNs. In this article, we introduce SGCNAX, a scalable GCN accelerator architecture for the high-performance and energy-efficient acceleration of GCNs. Unlike prior GCN accelerators that either employ limited loop optimization techniques, or determine the design variables based on random sampling, we systematically explore the loop optimization techniques for GCN acceleration and propose a flexible GCN dataflow that adapts to different GCN configurations to achieve optimal efficiency. We further propose two hardware-based techniques to address the workload imbalance problem caused by the unbalanced distribution of zeros in GCNs. Specifically, SGCNAX exploits an outer-product-based computation architecture that mitigates the intra-PE (processing Elements) workload imbalance, and employs a group-and-shuffle approach to mitigate the inter-PE workload imbalance. Simulation results show that SGCNAX performs 9.2x, 1.6x and 1.2x better, and reduces DRAM accesses by a factor of 9.7x, 2.9x and 1.2x compared to HyGCN, AWB-GCN, and GCNAX, respectively.

关键词： Random access memory Optimization Engines Registers neural networks Computational modeling Accelerator architectures Graph convolutional neural networks dataflow accelerators domain-specific accelerators memory access optimization

来源：评论

学校读者我要写书评

暂无评论

HKTGNN: Hierarchical Knowledge Transferable Graph neural network-based Supply Chain Risk Assessment 21

HKTGNN: Hierarchical Knowledge Transferable Graph Neural Net...

引用

21st IEEE International Symposium on Parallel and distributed processing with Applications, 13th IEEE International Conference on Big Data and Cloud Computing, 16th IEEE International Conference on Social Computing and networking and 13th International Conference on Sustainable Computing and Communications, ISPA/BDCloud/SocialCom/SustainCom 2023

作者： Zhou, Zhanting Bi, Kejun Zhong, Yuyanzhen Tang, Chao Li, Dongfen Ying, Shi Wang, Ruijin University of Electronic Science and Technology of China School of Information Science and Engineering Chengdu China Sichuan University School of Computer Science Chengdu China Chongqing Ceprei Industrial Technology Research Institute Co. Ltd. Chongqing China Sichuan Changhong Electronic Holding Group Co. Ltd. Chengdu China Chengdu University of Technology College of Computer Science and Cyber Security Chengdu China Wuhan University School of Computer Science Wuhan China

ISBN: (纸本)9798350329223

The strength of a supply chain is an important measure of a country's or region's technical advancement and overall competitiveness. Establishing supply chain risk assessment models for effective management and mitigation of potential risks has become increasingly crucial. As the number of businesses grows, the important relationships become more complicated and difficult to measure. This emphasizes the need of extracting relevant information from graph data. Previously, academics mostly employed knowledge inference to increase the visibility of links between nodes in the supply chain. However, they have not solved the data hunger problem of single node feature characteristics. We propose a hierarchical knowledge transferable graph neural network-based (HKTGNN1) supply chain risk assessment model to address these issues. Our approach is based on current graph embedding methods for assessing corporate investment risk assessment. We embed the supply chain network corresponding to individual goods in the supply chain using the graph embedding module, resulting in a directed homogeneous graph with just product nodes. This reduces the complicated supply chain network into a basic product network. It addresses difficulties using the domain difference knowledge transferable module based on centrality, which is presented by the premise that supply chain feature characteristics may be biased in the actual world. Meanwhile, the feature complement and message passing will alleviate the data hunger problem, which is driven by domain differences. Our model outperforms in experiments on a real-world supply chain dataset. We will give an equation to prove that our comparative experiment is both effective and fair.1Accepted by The 21st IEEE International Symposium on Parallel and distributed processing with Applications (IEEE ISPA 2023). © 2023 IEEE.

关键词： Supply chains

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：