检索结果-内蒙古大学图书馆

36th IEEE International Parallel and distributed processing Symposium (IEEE IPDPS)

作者： Koyama, Sohei Tatebe, Osamu Univ Tsukuba Tsukuba Ibaraki Japan

ISBN: (纸本)9781665497473

Graph neural networks (GNNs) operate on data represented as graphs, and are useful for a wide variety of tasks from chemical reaction and protein structure prediction to content recommendation systems. However, training for large graphs and improving training performance remain significant challenges. Existing distributed training systems partition a graph among all compute nodes to train for large graphs;however, this results in a communication overhead to degrade training performance. In this study, to solve these two problems, we propose a scalable data-parallel distributed GNN training system designed to partition a graph redundantly. It is implemented using remote direct memory access (RDMA) and nonblocking active messages to efficiently utilize network performance and hide communication overhead by overlapping with the training computation. Experimental results are presented to show the strong scalability of the proposed approach, which achieved parallel efficiencies of 0.93 using eight compute nodes for the ogbn-products dataset in the Open Graph Benchmark (OGB) and 0.95 based on two compute nodes using 32 compute nodes for the ogbn-papersl00M dataset. The proposed system exhibited a training performance 18.9% better than the state-of-the-art DistDGL, even with only a single compute node. The results demonstrate that the proposed approach may be considered a promising method to achieve scalable training performance for large graphs.

关键词： Graph neural networks distributed training High Performance Computing Deep Graph Library

来源：评论

学校读者我要写书评

暂无评论

CNN-LSTM based Multimodal Models for Music Generation 22

CNN-LSTM based Multimodal Models for Music Generation

引用

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

作者： Zhang, Man Liu, Dongning South China University of Technology School of Electronics and Information Guangzhou China Guangdong University of Technology Guangzhou School of Computer Science and Technology Guangzhou China

ISBN: (纸本)9798331509712

In this article, we tackle the dual challenges of efficiency and quality in music generation. We aim to create a model that produces high-quality music efficiently while keeping the model lightweight and the music authenticity and creativity. This research proposes a model that combines Long Short-Term Memory (LSTM) and Convolutional neural network (CNN) for music generation. LSTM units capture long-term dependencies in music sequences. CNN units are included to extract musical features, enhancing the model's grasp of structure and style. We train and valid the model on the ADL Piano MIDI datasets which is a sub set of the Lakh MIDI datasets. Finding that our CNN-LSTM model excels in music creation. It reduces computational complexity, boosting operational efficiency. This paves the way for deployment on resource-limited devices like mobiles or edge computing platforms. Combining these methods ensures the model learns form rich musical features, generating coherent, stylistically consistent pieces. © 2024 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

A Neurodynamic Algorithm for Energy Scheduling Game in Microgrid Distribution networks

引用

neural processing LETTERS 2022年第1期54卷 369-385页

作者： Wen, Shifan He, Xing Southwest Univ Coll Elect & Informat Engn Chongqing Key Lab Nonlinear Circuits & Intelligen Chongqing 400715 Peoples R China Southwestern Univ Finance & Econ Sch Econ Informat & Engn Chengdu 611130 Peoples R China

In this paper, a competitive energy scheduling strategy game of N-microgrids (MGs) inside a distributed network is considered. Each microgrid (MG) aims to maximize its profit under the noncooperative game frame. The strategy-making of each MG depends on its equipment constraints, the aggregate energy supplies of all MGs, and the energy balance of supplies and demands. To solve above discussed problem, a noncooperative game with linear coupled constraints and a distributed neurodynamic algorithm are proposed to seek the generalized Nash equilibrium (GNE). Besides, the correctness and convergence of the proposed algorithm are analyzed in detail. The effectiveness and feasibility of the proposed method are also illustrated via the simulation example.

关键词： Neurodynamic algorithm Energy scheduling game Microgrid distribution network

来源：评论

学校读者我要写书评

暂无评论

Facial Expression Recognition Based on Spatial-Temporal Fusion with Attention Mechanism

引用

neural processing LETTERS 2023年第5期55卷 6109-6124页

作者： Zhang, Lifeng Zheng, Xiangwei Chen, Xuanchi Ren, Xiuxiu Ji, Cun Shandong Normal Univ Sch Informat Sci & Engn Jinan Peoples R China Shandong Prov Key Lab Distributed Comp Software No Jinan Peoples R China

Facial expression recognition (FER) plays an important role in human-computer interaction and has been introduced in fatigue detection, human-computer interactive games, social robot, teaching effect analysis, and so on. However, not all the features extracted from facial images are suitable for FER. Moreover, individual spatial features or temporal features have certain limitations in characterizing facial expressions. Therefore, it is necessary to extract effective features suitable for facial expression recognition and to leverage effective fusion method to improve the accuracy of FER. In this paper, we propose a facial expression recognition method based on spatial-temporal fusion with attention mechanism(STAFER) which is composed of the spatial feature extractor (SFE), the temporal feature extractor (TFE), and spatial-temporal fusion (STF). Firstly, a 10 layers neural network of pretrained VGG16 is applied to the backbone of SFE to extract the spatial features of facial expression. To filter out the features that are irrelevant of facial expression, the attention mechanism is applied to the first three convolutional blocks. Secondly, TFE is constructed with convolutional blocks and LSTM. Facial expression sequences are fed to the convolutional modules to extract low-level features, and then these features are fed into the LSTM to obtain temporal features. Finally, a decision-level fusion strategy is utilized to fuse spatial and temporal features. The experiemntal results demonstrated that our proposed method achieves an accuracy of 98.05% on CK+ and 88.34% on Oulu-CASIA, which is competitive in comparison with some state-of-art methods.

关键词： Facial expression recognition Spatial-temporal feature fusion Attention mechanism Decision fusion

来源：评论

学校读者我要写书评

暂无评论

Efficient Radar-Target Assignment in Low Probability of Intercept Radar networks: A Machine-Learning Approach

IEEE OPEN JOURNAL OF THE COMMUNICATIONS SOCIETY

引用

IEEE OPEN JOURNAL OF THE COMMUNICATIONS SOCIETY 2023年 4卷 2165-2175页

作者： Amiriara, Hamid Andargoli, Seyed Mehdi Hosseini Meghdadi, Vahid Noshirvani Univ Technol Dept Elect & Comp Engn Babol *** Iran Univ Limoges UMR CNRS Lab 7252 XLIM F-87068 Limoges France

To achieve low probability of intercept (LPI) in radar networks for multiple target detection, it is necessary to find the optimal assignment of distributed radars to targets. The multi-radar to multi-target assignment (MRMTA) problem aims to find the best radar combination, but its brute-force (BF)-based approach over all possible sensor combinations has exponential complexity, making it challenging to implement in networks with a large number of radars or targets. This limits the implementation of the BF approach in networks that prioritize low latency and complexity. To address this challenge, we propose a supervised machine-learning (ML)-based solution for the MRMTA problem. Our proposed implementation scheme performs the training procedure offline, leading to a significant reduction in assignment complexity and processing latency. We conducted extensive numerical simulations to design an ML structure with high accuracy, convergence speed, and scalability. Simulation results demonstrate the efficiency and effectiveness of our proposed ML-based MRMTA solution, which achieves near-optimal LPI performance with considerably lower computation time than benchmark schemes. Our proposed solution has the potential to optimize the assignment of distributed radars to targets in LPI radar networks and improve the performance of complex networks with low latency and complexity requirements.

关键词： Multi-radar to multi-target assignment (MRMTA) supervised machine-learning (ML) low probability intercept radars feed-forward neural network (FNN) radar network information fusion

来源：评论

学校读者我要写书评

暂无评论

Wrht: Efficient All-reduce for distributed DNN Training in Optical Interconnect Systems 23

Wrht: Efficient All-reduce for Distributed DNN Training in O...

引用

52nd International Conference on Parallel processing (ICPP)

作者： Dai, Fei Chen, Yawen Huang, Zhiyi Zhang, Haibo Univ Otago Dunedin New Zealand

ISBN: (纸本)9798400708435

Communication efficiency is crucial for accelerating distributed deep neural network (DNN) training. All-reduce, a vital communication primitive, is responsible for reducing model parameters in distributed DNN training. However, most existing All-reduce algorithms, designed for traditional electrical interconnect systems, fall short due to bandwidth limitations. Optical interconnects, with superior bandwidth, low transmission delay, and less power consumption, emerge as viable alternatives. We propose Wrht (Wavelength Reused Hierarchical Tree), an efficient scheme for implementing the All-reduce operation in optical interconnect systems. Wrht leverages wavelength-division multiplexing (WDM) to minimize the communication time in distributed data-parallel DNN training. We calculate the required wavelengths, minimum communication steps, and optimal communication time, considering optical communication constraints. Simulations with real-world DNN models indicate that Wrht notably reduces communication time. On average, compared with three conventional All-reduce algorithms, Wrht achieves reductions of 65.23%, 43.81%, and 82.22% respectively in optical interconnect systems, and 61.23% and 55.51% compared with two algorithms in electrical systems. This highlights Wrht's potential to enhance communication efficiency in DNN training using optical interconnects.

关键词： Optical interconnects All-reduce distributed training DNN WDM

来源：评论

学校读者我要写书评

暂无评论

Opto-Electronic Smart Home: Heterogeneous Optical Sensors Approaches and Artificial Intelligence for Novel Paradigms in Remote Monitoring

引用

IEEE INTERNET OF THINGS JOURNAL 2024年第6期11卷 9587-9598页

作者： Leal-Junior, Arnaldo Avellar, Leticia Blanc, Wilfried Frizera, Anselmo Marques, Carlos Univ Fed Espirito Santo Grad Program Elect Engn BR-29075910 Vitoria Brazil Univ Cote Azur Inst Phys Nice CNRS F-06108 Nice 2 France Univ Aveiro I3N P-3810193 Aveiro Portugal Univ Aveiro Dept Phys P-3810193 Aveiro Portugal

This article presents the development and implementation of an-optical fiber integrated smart environment with heterogeneous opto-electronic approaches. In this case, the so-called opto-electronic smart home is composed of three different optical fiber sensor system, which are also based on different optical fibers, resulting in more than 50 integrated sensors. The proposed smart environment is capable of detecting the location of the patient inside the home environment, recognize patient's activities and provide the gait analysis through kinematics and spatio-temporal parameters of the gait. The heterogeneity of the system is verified by the use of the transmission-reflection analysis (TRA) using nanoparticle (NP)-doped optical fibers for the patient localization in Layer 1. Then, in Layer 2 a polymer optical fiber (POF) integrated pants is used by the patient, where the activity detection, especially walking, sitting and lying down is performed by the multiplexed intensity variation-based sensor integrated in the pants (with 30 sensors at each leg of the pants). Layer 3 comprises a fiber Bragg grating (FBG)-embedded smart carpet, where ten FBGs are inscribed in a single mode silica optical fiber. In addition, a graphical interface is developed for the sensors integration and cloud connectivity, where the signal processing is performed using the feedforward neural network (FFNN) approach for the location of mechanical perturbation along the optical fiber (for patient localization), activity classification and footsteps location along the FBG-embedded smart carpet. The implementation results show the feasibility of the proposed system, where the location of the patients, their activities and gait analysis.

关键词： Optical fiber sensors Optical fibers Monitoring Optical reflection Optical signal processing Fiber gratings Instruments distributed optical fiber sensors Internet of Things (IoT) photonics digitalization remote monitoring

来源：评论

学校读者我要写书评

暂无评论

POST-processing FLOWS USING PHYSICS-INFORMED neural networkS 10

POST-PROCESSING FLOWS USING PHYSICS-INFORMED NEURAL NETWORKS

引用

10th Convention of the European Acoustics Association, EAA 2023

作者： Schoder, Stefan Kraxberger, Florian Mušeljić, Eniz Wurzinger, Andreas Graz University of Technology Graz Austria

ISBN: (纸本)9788888942674

In this contribution, the Helmholtz decomposition of a compressible flow velocity field into vortical and compressible structures is implemented using a finite element framework and physics-informed neural networks. These two implementations of Helmholtz's decomposition are compared for a verification example and a 2D mixing layer flow. The work shows how neural networks can leverage physical knowledge to perform the inverse task of post-processing a compressible flow field into subparts. Furthermore, different input variables, network setups, network parameters, network types, and formulations of the objective function for the optimizer are investigated and compared to each other. The physics-informed neural network formulation results on the verification example outline promising directions for further applications to post-process compressible flow fields. © 2023 Stefan Schoder et al. This is an open-access article distributed under the terms of the Creative Commons Attribution 3.0 Unported License, which permits unrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.

关键词： Compressible flow

来源：评论

学校读者我要写书评

暂无评论

Exploring Load Forecasting: Bridging Statistical Methods to Deep Learning Techniques in Real-World Environment 20

Exploring Load Forecasting: Bridging Statistical Methods to ...

引用

20th Annual International Conference on distributed Computing in Smart Sensor Systems and the Internet of Things (DCOSS-IoT)

作者： Makris, Ioannis Moschos, Nikolaos Radoglou-Grammatikis, PaPanagiotis Andriopoulos, Nikolaos Tzanis, Nikolaos Malamaki, Kyriaki-Nefeli Fotopoulou, Maria Kaousias, Konstantinos Tompros, George Lampsidis Sarigiannidis, Panagiotis MetaMind Innovat Kozani Greece Ctr Res & Technol Hellas CERTH Athens Greece Hellen Elect Distribut Network Operator HEDNO Athens Greece Univ Western Macedonia Dept Elect & Comp Engn Kozani Greece K3Y Ltd Sofia Bulgaria

ISBN: (纸本)9798350369458;9798350369441

Load forecasting has a significant impact on energy management and planning, facilitating efficient allocation of resources and grid operations. In this study, a comparative analysis of traditional statistical methods and deep learning techniques is conducted utilizing a real-world dataset from the Ikaria islanded grid. This paper focuses on four different forecasting approaches: Autoregressive Integrated Moving Average (ARIMA), Seasonal Autoregressive Integrated Moving Average with Exogenous Variables (SARIMAX), Long Short-Term Memory (LSTM) networks, and Deep neural networks (DNN). Through the appropriate processing of the data, extensive experimentation took place, aiming to capture the complex and nonlinear patterns of the dataset. The results indicated that LSTM and DNN outperformed both ARIMA and SARIMAX in all three evaluation metrics, achieving 0.13, 0.09, and 2.11%, RMSE, MAE, and MAPE, respectively. As a result, this study validates the superiority of deep learning techniques in real-world islanded grid environments being capable of accurately predicting future load values based on historical data.

关键词： Deep neural networks Electricity Demand Prediction Load Forecasting Long Short-term Memory Time Series Analysis

来源：评论

学校读者我要写书评

暂无评论

Efficient Implementation of 2-D Convolution on DRRA and DiMArch Architectures 23

Efficient Implementation of 2-D Convolution on DRRA and DiMA...

引用

13th International Symposium on Highly Efficient Accelerators and Reconfigurable Technologies (HEART)

作者： Pudi, Dhilleswararao Ryansh, Rajeev Boppu, Srinivas Yang, Yu Hemani, Ahmed Indian Inst Technol Bhubaneswar Odisha India Royal Inst Technol KTH Kista Sweden

ISBN: (纸本)9798400700439

Convolution has been widely employed in image processing and computer vision applications such as picture augmentation, smoothing, and structure extraction. In addition, convolution operations are the most prevalent computing patterns in machine learning domains. Convolutions, for example, are used in a substantial chunk of state-of-the-art convolutional neural network operations. Therefore, effectively mapping convolution operations onto hardware architectures is crucial for achieving superior performance while accelerating convolutional neural networks. In this paper, we proposed various algorithms to efficiently map the 2-D convolution operation onto a dynamically reconfigurable resource array and distributed memory architecture. Furthermore, we have discussed the mapping of 2-D convolution on the target architecture for an input matrix of arbitrary size, as well as the generalization of the proposed approaches for multi-column DRRA architectures.

关键词： Convolution Coarse Grain Reconfigurable Architectures Dynamically Reconfigurable Resource Array distributed Memory Architecture Machine Learning Convolutional neural networks FPGA

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：