检索结果-内蒙古大学图书馆

LeL-GNN: Learnable Edge Sampling and Line Based Graph neural network for Link Prediction

IEEE ACCESS 2023年 11卷 56083-56097页

作者： Morshed, Md Golam Sultana, Tangina Lee, Young-Koo Kyung Hee Univ Dept Comp Sci & Engn Global Campus Yongin 17104 South Korea Int Univ Business Agr & Technol Dept Comp Sci & Engn Dhaka 1230 Bangladesh Hajee Mohammad Danesh Sci & Technol Univ Dept Elect & Commun Engn Dinajpur 5200 Bangladesh

Graph neural networks lose a lot of their computing power when more network layers are added. As a result, the majority of existing graph neural networks have a shallow depth of learning. Over-smoothing and information loss are two of the key issues that restrict graph neural networks from going deeper. As network depth goes up, the embeddings of all the nodes eventually converge on the same value, which separates output representations from input vectors and causes over-smoothing. Moreover, layers of graph pooling are required in a deep learning model to retrieve specified features for prediction, which results in some degree of information loss. In this research, we present a new and multi-scale approach for overcoming these constraints by using concepts from graph theory, namely learnable edge sampling and line graphs. An edge-sampling mechanism that selects a particular number of edges through a learning parameter before training reduces oversmoothing, and the issue of information loss is alleviated using a line graph technique that converts the original graph into a similar line graph. Our method of edge sampling preserves the core spectral features of the graph without affecting its fundamental structure. Our suggested technique outperforms state-of-the-art models on publicly available datasets of diverse applications while having minimal constraints and great training skills.

关键词： Edge sampling deep graph neural networks line graph link prediction

来源：评论

学校读者我要写书评

暂无评论

Structural Approach to Convolutional neural network Trained With Novel Scaled Matrix Image for Pseudo Real-Time Power Quality Event Monitoring

引用

IEEE ACCESS 2024年 12卷 130833-130856页

作者： Mishra, Sairam Mallick, Ranjan Kumar Gadanayak, Debadatta Amaresh Nayak, Pravati Sharma, Renu Flah, Aymen Prokop, Lukas Kraiem, Habib Kanan, Mohammad Siksha O Anusandhan Deemed be Univ Dept Elect Engn ITER Bhubaneswar 751030 India Siksha O Anusandhan Deemed be Univ Dept Elect & Elect Engn ITER Bhubaneswar 751030 India Univ Gabes Natl Engn Sch Gabes Energy Proc Environm & Elect Syst Unit Gabes 6029 Tunisia Middle East Univ MEU Res Unit Amman 11831 Jordan Univ Gabes Private Higher Sch Appl Sci & Technol Gabes ESSAT Gabes 6029 Tunisia Appl Sci Private Univ Appl Sci Res Ctr Amman 11931 Jordan VSB Tech Univ Ostrava ENET Ctr Ostrava 70800 Czech Republic Northern Border Univ Coll Engn Dept Elect Engn Ar Ar 91431 Saudi Arabia Univ Business & Technol UBT Coll Engn Ind Engn Dept Jeddah 21448 Saudi Arabia

The trend of integrating different distributed generation sources into the existing grid have increased the probability of power quality disturbances to a threatening level. Eventually, detection, protection and mitigation of these disturbances are even more challenging. In this regard, the article presents an intelligent power quality disturbance classification scheme using a 2D convolutional neural network designed from a systematic and structural standpoint. A total of 8 singular, 5 complex power quality events are simulated and voltage data collection is made from a test system designed in MATLAB Simulink environment. The three-phase voltage data is converted to a single signal arrangement through a newly proposed Unique Clark's Transformed Sequence. In addition to that, the scheme completely eliminates the worry of a signal processing stage by proposing a novel scaled matrix image created out of 2-cycle data collected at 6.4kHz sampling frequency that acts as input to the CNN architecture designed in the PYTHON environment. Further, the novelty extended to design a pseudo-real-time setup where the MATLAB environment continuously runs the test system, producing scaled matrix images. These images are saved to a shared directory, enabling a PYTHON loop for prompt event classification through the trained CNN model. The model performance is found to be 100% under ideal conditions. It has also tested under three different noise conditions of 40dB, 30dB and 20dB and obtained an overall accuracy of 98.86% with singular events. Further, the method is also verified for complex and unsymmetrical dataset and found to be equally effective. Additionally, the validation is likewise made with a trivial set of real-time simulated data using OPAL-RT 4510 setup. Finally, the proposed PQ detection scheme is compared with recently published work to express its superiority over other similar studies in terms of classification *** TERMS Convolutional neural network, deep l

关键词： Feature extraction Convolutional neural networks Manuals Power quality Mathematical models Signal processing Data models Deep learning Power distribution networks Convolutional neural network deep learning power quality disturbance classification scaled matrix image unique Clark's transformed sequence

来源：评论

学校读者我要写书评

暂无评论

Scalable K-FAC Training for Deep neural networks With distributed Preconditioning

引用

IEEE TRANSACTIONS ON CLOUD COMPUTING 2023年第3期11卷 2365-2378页

作者： Zhang, Lin Shi, Shaohuai Wang, Wei Li, Bo Hong Kong Univ Sci & Technol Dept Comp Sci & Engn Hong Kong Peoples R China Harbin Inst Technol Sch Comp Sci & Technol Shenzhen 518055 Guangdong Peoples R China

The second-order optimization methods, notably the D-KFAC (distributed Kronecker Factored Approximate Curvature) algorithms, have gained traction on accelerating deep neural network (DNN) training on GPU clusters. However, existing D-KFAC algorithms require to compute and communicate a large volume of second-order information, i.e., Kronecker factors (KFs), before preconditioning gradients, resulting in large computation and communication overheads as well as a high memory footprint. In this article, we propose DP-KFAC, a novel distributed preconditioning scheme that distributes the KF constructing tasks at different DNN layers to different workers. DP-KFAC not only retains the convergence property of the existing D-KFAC algorithms but also enables three benefits: reduced computation overhead in constructing KFs, no communication of KFs, and low memory footprint. Extensive experiments on a 64-GPU cluster show that DP-KFAC reduces the computation overhead by 1.55x-1.65x, the communication cost by 2.79x-3.15x, and the memory footprint by 1.14x-1.47x in each second-order update compared to the state-of-the-art D-KFAC methods.

关键词： Training Computational modeling Clustering algorithms Graphics processing units Memory management Deep learning Convergence distributed deep learning K-FAC performance optimization second-order

来源：评论

学校读者我要写书评

暂无评论

B-ary Tree Push-Pull Method is Provably Efficient for distributed Learning on Heterogeneous Data 38

B-ary Tree Push-Pull Method is Provably Efficient for Distri...

引用

38th Conference on neural Information processing Systems, NeurIPS 2024

作者： You, Runze Pu, Shi The Chinese University of Hong Kong Shenzhen China

This paper considers the distributed learning problem where a group of agents cooperatively minimizes the summation of their local cost functions based on peer-to-peer communication. Particularly, we propose a highly efficient algorithm, termed "B-ary Tree Push-Pull" (BTPP), that employs two B-ary spanning trees for distributing the information related to the parameters and stochastic gradients across the network. The simple method is efficient in communication since each agent interacts with at most (B + 1) neighbors per iteration. More importantly, BTPP achieves linear speedup for smooth nonconvex and strongly convex objective functions with only Õ(n) and Õ(1) transient iterations, respectively, significantly outperforming the state-of-the-art results to the best of our knowledge. Our code is available at https://***/ryou98/BTPP. © 2024 neural information processing systems foundation. All rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Time-wavelength multiplexed photonic neural network accelerator for distributed acoustic sensing systems

引用

Advanced Photonics 2025年第2期 165-178页

作者： Fuhao Yu Kangjian Di Wenjun Chen Sen Yan Yuanyuan Yao Silin Chen Xuping Zhang Yixin Zhang Ningmu Zou Wei Jiang Nanjing University College of Engineering and Applied Sciences Nanjing University Ministry of Education Key Laboratory of Intelligent Optical Sensing and Manipulation Nanjing University School of Integrated Circuits Nanjing University Interdisciplinary Research Center for Future Intelligent Chips (Chip-X)

distributed acoustic sensors（DASs） can effectively monitor acoustic fields along sensing fibers with high sensitivity and high response speed. However, their data processing is limited by the performance of electronic signal processing, hindering real-time applications. The time-wavelength multiplexed photonic neural network accelerator（TWM-PNNA）, which uses photons instead of electrons for operations, significantly enhances processing speed and energy efficiency. Therefore, we explore the feasibility of applying TWMPNNA to DAS systems. We first discuss processing large DAS system data for compatibility with the TWM-PNNA system. We also investigate the effects of chirp on optical convolution in complex tasks and methods to mitigate its impact on classification accuracy. Furthermore, we propose a method for achieving an optical full connection and study the influence of pruning on the full connection to reduce the computational burden of the model. Experimental results indicate that decreasing the ratio of Δλchirp∕Δλ or choosing push–pull modulation can eliminate the impact of chirp on recognition accuracy. In addition, when the full connection parameter retention rate is no less than 60%, it can still maintain a classification accuracy of over 90%.TWM-PNNA provides an innovative computational framework for DAS systems, paving the way for the all-optical fusion of DAS systems with computational systems.

关键词：

来源：评论

学校读者我要写书评

暂无评论

AFCNNM: Accelerating Fully Connected neural network on Mesh-based Optical network-on-Chip 22

AFCNNM: Accelerating Fully Connected Neural Network on Mesh-...

引用

22nd IEEE International Symposium on Parallel and distributed processing with Applications, ISPA 2024

作者： Yang, Wen Liao, Yunting China Three Gorges University Hubei Key Laboratory of Intelligent Vision Based Monitoring for Hydroelectric Engineering Yichang443000 China China Three Gorges University College of Computer and Information Technology Yichang443002 China

ISBN: (纸本)9798331509712

Fully Connected neural network (FCNN) are widely used in image recognition and natural language processing. However, the time cost of training large datasets is high. Optical network-on Chip (ONoC) has been proposed to accelerate the parallel computing of FCNN because of its advantages. Therefore, this paper proposes an accelerated FCNN model based on ONoC. We first design an FCNN-aware mapping strategy, and then propose a group-based inter-core communication scheme with low wavelength requirements according to the distribution of mapping cores. The optimal number of cores in each period is obtained by achieving the trade-off between the communication and computation time. The simulation results show that the proposed scheme has the advantages of low wavelength requirement, short training time and good scalability. © 2024 IEEE.

关键词： Large datasets

来源：评论

学校读者我要写书评

暂无评论

neural network Algorithms in the Construction of Intelligent Financial Sharing Platform 3

Neural Network Algorithms in the Construction of Intelligent...

引用

3rd IEEE International Conference on distributed Computing and Electrical Circuits and Electronics, ICDCECE 2024

作者： Fu, Yuanyuan Weifang Engineering Vocational College Shandong Qingzhou China

ISBN: (纸本)9798350318609

The current financial sharing platform has no advantages when dealing with nonlinear relationships. The purpose of this research is to apply neural network algorithms to build an intelligent financial sharing platform to improve the accuracy of financial data processing and forecasting. In this study, a multi-layer perceptron is used as the main neural network algorithm, and a financial data set is used for model training and verification. It performed preprocessing and feature engineering of financial data, and then divided the data into training sets and test sets. After the division is completed, the paper continued to build the multilayer perceptron model, added cross-verification and parameter adjustment techniques to optimize the performance and generalization capabilities of the model, and then evaluated the accuracy of the model after the platform was built. The experimental results show that the accuracy rate of the multilayer perceptron algorithm is between 92% and 98%, and the intelligent financial sharing platform built on the multilayer perceptron algorithm has achieved significant improvements in financial data classification and prediction tasks. © 2024 IEEE.

关键词： Multilayer neural networks

来源：评论

学校读者我要写书评

暂无评论

A Novel Framework for Quantum-Enhanced Federated Learning with Edge Computing for Advanced Pain Assessment Using ECG Signals via Continuous Wavelet Transform Images

引用

SENSORS 2025年第5期25卷 1436-1436页

作者： Balasubramani, Madankumar Srinivasan, Monisha Jean, Wei-Horng Fan, Shou-Zen Shieh, Jiann-Shing Yuan Ze Univ Dept Mech Engn Taoyuan 320 Taiwan Far Eastern Mem Hosp Dept Anesthesiol New Taipei City 220 Taiwan Natl Taiwan Univ Coll Med Dept Anesthesiol Taipei 100 Taiwan En Chu Kong Hosp Dept Anesthesiol New Taipei City 237 Taiwan

Our research introduces a framework that integrates edge computing, quantum transfer learning, and federated learning to revolutionize pain level assessment through ECG signal analysis. The primary focus lies in developing a robust, privacy-preserving system that accurately classifies pain levels (low, medium, and high) by leveraging the intricate relationship between pain perception and autonomic nervous system responses captured in ECG signals. At the heart of our methodology lies a signal processing approach that transforms one-dimensional ECG signals into rich, two-dimensional Continuous Wavelet Transform (CWT) images. These transformations capture both temporal and frequency characteristics of pain-induced cardiac variations, providing a comprehensive representation of autonomic nervous system responses to different pain intensities. Our framework processes these CWT images through a sophisticated quantum-classical hybrid architecture, where edge devices perform initial preprocessing and feature extraction while maintaining data privacy. The cornerstone of our system is a Quantum Convolutional Hybrid neural network (QCHNN) that harnesses quantum entanglement properties to enhance feature detection and classification robustness. This quantum-enhanced approach is seamlessly integrated into a federated learning framework, allowing distributed training across multiple healthcare facilities while preserving patient privacy through secure aggregation protocols. The QCHNN demonstrated remarkable performance, achieving a classification accuracy of 94.8% in pain level assessment, significantly outperforming traditional machine learning approaches.

关键词： quantum convolutional hybrid neural network (QCHNN) federated learning edge computing pain assessment quantum transfer learning

来源：评论

学校读者我要写书评

暂无评论

ACTIVE LEARNING FOR SOUND EVENT CLASSIFICATION USING BAYESIAN neural networkSWITH GAUSSIAN VARIATIONAL POSTERIOR 49

ACTIVE LEARNING FOR SOUND EVENT CLASSIFICATION USING BAYESIA...

引用

49th IEEE International Conference on Acoustics, Speech, and Signal processing (ICASSP)

作者： Shishkini, Stepan Hollosi, Danilo Goetze, Stefan Doclo, Simon Fraunhofer Inst Digital Media Technol IDMT Oldenburg Branch Hearing Speech & Audio Technol H Oldenburg Germany Univ Sheffield Dept Comp Sci Speech & Hearing SPandH Sheffield S Yorkshire England Carl von Ossietzky Univ Oldenburg Dept Med Phys & Acoust Oldenburg Germany Carl von Ossietzky Univ Oldenburg Cluster Excellence Hearing4all Oldenburg Germany

ISBN: (纸本)9798350344868;9798350344851

Manual annotation of audio material is cumbersome. Active learning aims at minimizing the annotation effort by iteratively selecting an acquisition batch of unlabeled data, asking a human to annotate the selected data and re-training a classifier until an annotation budget is depleted. In this paper we propose the Gaussian-dense active learning (GDAL) algorithm to train a sound event classifier. The classifier is a Bayesian neural network where the weights are normally distributed. This is in contrast to conventional neural networks where weights are not distributed, but have assigned values. The Bayesian nature of the classifier empowers GDAL to select acquisition batches from a set of unlabeled audio clips based on their estimated informativeness. Evaluation results on the UrbanSound8k dataset show that GDAL outperforms a state-of-the-art algorithm based on medoid active learning for all considered annotation budgets and an algorithm based on dropout active learning for sufficiently large annotation budgets.

关键词： sound event classification active learning Bayesian neural networks PANN embeddings

来源：评论

学校读者我要写书评

暂无评论

Preprocessing and frame level classification framework for cardiac phase detection in 2D echocardiography

引用

BIOMEDICAL SIGNAL processing AND CONTROL 2025年 107卷

作者： Singh, Gajraj Darji, Anand D. Sarvaiya, Jignesh N. Patnaik, Suprva Sardar Vallabhbhai Natl Inst Technol Dept Elect Engn Ichhanath Rd Surat 395007 Gujrat India DY Patil Deemed Univ Ramrao Adik Inst Technol Dept Informat Technol Navi Mumbai 400706 Maharashtra India Kalinga Inst Ind Technol Sch Elect Engn KIIT Rd Bhubaneswar 751024 Orissa India

Accurate detection of end-diastole (ED) and end-systole (ES) frames is a crucial step in cardiac function analysis, enabling precise measurement of ventricular volume, ejection fraction (EF), and stroke volume (SV). However, this task is challenging due to variations in cardiac structure, heart rate fluctuations associated with clinical conditions, and the low-resolution nature of echocardiographic sequences. This study addresses these challenges by introducing three preprocessing steps - noise reduction via heart rate formulation, video frame synchronization, and non-oscillating mean absolute frame difference - to denoise and enhance the EchoNetDynamic dataset. Additionally, the echo phase detection problem is reformulated as a frame-level binary classification task to mitigate class imbalance between diastole and systole phases. The proposed architecture employs a time-distributed convolutional neural network (CNN) to extract spatial features, followed by a bidirectional long short-term memory (BiLSTM) network to capture temporal dynamics, and a classification layer for phase prediction. The model achieves an average absolute frame distance of 1.02 and 1.04 frames for ED and ES frames, respectively, on the preprocessed EchoNet-Dynamic dataset. To ensure better generalization, the model was also validated on the CAMUS dataset and private data, where it demonstrated consistent performance and robust results. These findings significantly enhance the reliability of cardiac metrics, offering clinicians a precise and efficient tool for echocardiographic analysis.

关键词： Echocardiography Diastole and systole phase Time-distributed CNN Bidirectional long short-term memory Average absolute frame distance

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：