检索结果-内蒙古大学图书馆

Proceedings of the 51st International Conference on Parallel processing

作者： Refael Cohen Ido Hakimi Assaf Schuster Department of Computer Science Technion - Israel Institute of Technology Israel

ISBN: (纸本)9781450397339

As the field of deep learning progresses, and neural networks become larger, training them has become a demanding and time consuming task. To tackle this problem, distributed deep learning must be used to scale the training of deep neural networks to many workers. Synchronous algorithms, commonly used for distributing the training, are susceptible to faulty or straggling workers. Asynchronous algorithms do not suffer from the problems of synchronization, but introduce a new problem known as staleness. Staleness is caused by applying out-of-date gradients, and it can greatly hinder the convergence process. Furthermore, asynchronous algorithms that incorporate momentum often require keeping a separate momentum buffer for each worker, which cost additional memory proportional to the number of workers. We introduce a new asynchronous method, SMEGA2, which requires a single momentum buffer regardless of the number of workers. Our method works in a way that lets us estimate the future position of the parameters, thereby minimizing the staleness effect. We evaluate our method on the CIFAR and ImageNet datasets, and show that SMEGA2 outperforms existing methods in terms of final test accuracy while scaling up to as much as 64 asynchronous workers. Open-Source Code: https://***/rafi-cohen/SMEGA2

关键词： neural networks asynchronous gradient descent distributed deep learning

来源：评论

学校读者我要写书评

暂无评论

Attention, Distillation, and Tabularization: Towards Practical neural network-Based Prefetching 38

Attention, Distillation, and Tabularization: Towards Practic...

引用

International Parallel and distributed processing Symposium (IPDPS)

作者： Zhang, Pengmiao Gupta, Neelesh Kannan, Rajgopal Prasanna, Viktor K. Univ Southern Calif Los Angeles CA 90007 USA DEVCOM Army Res Lab Los Angeles CA USA

ISBN: (纸本)9798350387117;9798350387124

Attention-based neural networks (NN) have demonstrated their effectiveness in accurate memory access prediction, an essential step in data prefetching. However, the substantial computational overheads associated with these models result in high inference latency, limiting their feasibility as practical prefetchers. To close the gap, we propose a new approach based on tabularization that significantly reduces model complexity and inference latency without sacrificing prediction accuracy. Our novel tabularization methodology takes input as a distilled, yet highly accurate attention-based model for memory access prediction and efficiently converts its expensive matrix multiplications into a hierarchy of fast table lookups. As an exemplar of the above approach, we develop DART, a prefetcher comprised of a simple hierarchy of tables. With a modest 0.09 drop in F1-score, DART reduces 99.99% of arithmetic operations from the original attention-based model and 91.83% from the distilled model. DART accelerates the large model inference by 170x and the distilled model by 9.4x. DART has comparable latency and storage costs as state-of-the-art rule-based prefetcher BO but surpasses it by 6.1% in IPC improvement. DART outperforms state-of-the-art NN-based prefetchers TransFetch by 33.1% and Voyager by 37.2% in terms of IPC improvement, primarily due to its low prefetching latency.

关键词： memory access prediction attention neural network knowledge distillation tabularization prefetching

来源：评论

学校读者我要写书评

暂无评论

Multipath Exploitation for Human Activity Recognition Using a Radar network

引用

IEEE TRANSACTIONS ON GEOSCIENCE AND REMOTE SENSING 2024年 62卷 1页

作者： Guendel, Ronny Gerhard Kruse, Nicolas C. Fioranelli, Francesco Yarovoy, Alexander Delft Univ Technol Dept Microelect MS3 Grp NL-2628 CD Delft Netherlands

In this study, the problem of multipath in radar sensor networks for human activity recognition (HAR) has been examined. Traditionally considered as a source of additional clutter, the multipath is being investigated for its potential to be exploited through the creation of virtual radar nodes. These virtual nodes are conceptualized to observe targets from aspect angles that differ from those of physically existing radars. To realize this idea, an innovative processing pipeline is proposed that extracts information from multipath signals to improve HAR. The pipeline isolates and tracks the line-of-sight (LOS) and multipath components of a moving human target performing continuous sequences of activities observed by a network of three radar sensors. Furthermore, the method has been verified with experimental data consisting of six activities and 14 volunteers by comparing classification metrics with the use of a single radar as well as only the LOS components of the three radars in the network. A 12-layer convolutional neural network (CNN) classifier has been designed to operate on range-Doppler (RD) images derived from the LOS and multipath components, extracted by the proposed method. A substantial performance improvement using the leave-one-person-out (L1Po) test set is demonstrated in the order of +11% by exploiting a multiradar network with its LOS and multipath components.

关键词： Clustering distributed radar hierarchical clustering human activity recognition (HAR) multilateration multipath radar multipath radar signal processing trilateration

来源：评论

学校读者我要写书评

暂无评论

NetPU-M: a Generic Reconfigurable neural network Accelerator Architecture for MLPs

NetPU-M: a Generic Reconfigurable Neural Network Accelerator...

引用

37th IEEE International Parallel and distributed processing Symposium (IPDPS)

作者： Liu, Yuhao Rai, Shubham Ullah, Salim Kumar, Akash Tech Univ Dresden Ctr Adv Elect Dresden CfAED Processor Design Dresden Germany ScaDS AI Dresden Ctr Scalable Data Analyt & Artificial Intelligenc Leipzig Germany

ISBN: (纸本)9798350311990

Recent research widely deployed neural networks (NNs) in various scenarios, such as IoT systems, wearable devices, or smart sensors. However, the complex application scenarios cause the rapid extension of network model size and the requirement for higher-performance hardware platforms. Related works apply Heterogeneous Streaming Dataflow (HSD) and processing Element Matrix (PEM) architectures as the most popular schemes for FPGA-based implementation of NN accelerator: 1) HSD architecture implements a complete network for given trained models on FPGA with simplified control but more hardware consumption;2) PEM architecture implements reusable neuron structures controlled by runtime environments/drivers providing the generic acceleration supports for different network models. Our work explores a new hybrid architecture based on HSD and PEM to implement a reusable partial network structure on FPGA and achieve generic acceleration supports for different network models with simplified runtime control. This architecture supports scalable, mixable, quantized precision, and selectable activation functions, including ReLU, Sigmoid, Tanh, Sign, and Multi-Thresholds. Data stream transmission can reset the accelerator configuration in runtime without hardware implementation changes for different networks. Our design fully supports the generic inference acceleration for different Multi-Layer Perceptron (MLP) models.

关键词： FPGA Generic Hardware Accelerator neural network Quantization

来源：评论

学校读者我要写书评

暂无评论

An ANN-Aided Parameter Design Method for CLLC-Type DAB Converters Considering Parameter Perturbation

引用

IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS 2025年第4期72卷 3735-3745页

作者： Wang, Ning Jiang, Yongbin Hu, Weihao Wang, Yanbo Chen, Zhe Aalborg Univ Dept Energy Technol DK-9220 Aalborg Denmark Nanyang Technol Univ Sch Elect & Elect Engn Singapore 639798 Singapore Univ Elect Sci & Technol China Inst Elect & Informat Engn Guangzhou 523808 Guangdong Peoples R China Univ Elect Sci & Technol China Sch Mech & Elect Engn Chengdu 610056 Peoples R China

The distributed nature of power electronic components parameters can affect the desired output voltage of the CLLC-type dual active bridge (DAB) converters, especially in mass production with limited budgets. To minimize inconsistency for CLLC-type DAB converters against manufacturing tolerance in large-scale applications, this article proposes a novel resonant component parameter design method based on artificial neural network (ANN). Moreover, an ANN-based data-driven model of the probability density function is first developed to portray the distribution of component parameters within the allowable tolerance range. Furthermore, to enhance data processing efficiency in the parametric design process, a batch-normalization method is proposed to convert the original dataset to the normalized one in batches automatically. The co-simulation method is implemented with Monte Carlo analysis by combining MATLAB with LTspice. To ensure the accuracy of the co-simulation method, experimental results for the limited parameter combinations are provided as the verification for the co-simulation method. Finally, Monte Carlo analysis is adopted to optimize the resonant components parameter with three quantitative evaluation indexes. The verification results show that the failure rate of the output voltage can be reduced to less than 5%.

关键词： Design methodology Probability density function Data processing Manufacturing Voltage Mass production Magnetic resonance Artificial neural network (ANN) batch-normalization CLLC-type dual active bridge (DAB) data-driven model Monte Carlo

来源：评论

学校读者我要写书评

暂无评论

Fast fiber nonlinearity compensation method for PDM coherent optical transmission systems based on the Fourier neural operator

引用

OPTICS EXPRESS 2024年第2期32卷 2245-2256页

作者： Huang, Junling Yi, Anlin Yan, Lianshan He, Xingchen Jiang, Lin Yang, Hui Luo, Bin Pan, Wei Southwest Jiaotong Univ Ctr Informat Photon & Commun Sch Informat Sci & Technol Chengdu 610031 Sichuan Peoples R China

Fiber nonlinearity compensation (NLC) is likely to become an indispensable component of coherent optical transmission systems for extending the transmission reach and increasing capacity per fiber. In this work, we introduce what we believe to be a novel fast black -box neural network model based on the Fourier neural operator (FNO) to compensate for the chromatic dispersion (CD) and nonlinearity simultaneously. The feasibility of the proposed approach is demonstrated in uniformly distributed as well as probabilistically -shaped 32GBaud 16/32/64-ary quadrature amplitude modulation (16/32/64QAM) polarization -division -multiplexed (PDM) coherent optical communication systems. The experimental results demonstrate that about 0.31 dB Q -factor improvement is achieved compared to traditional digital back -propagation (DBP) with 5 steps per span for PDM-16QAM signals after 1600 km standard single -mode fiber (SSMF) transmission at the optimal launched power of 4 dBm. While, the time consumption is reduced from 6.04 seconds to 1.69 seconds using a central processing unit (CPU), and from 1.54 seconds to only 0.03 seconds using a graphic processing unit (GPU), respectively. This scheme also reveals noticeable generalization ability in terms of launched power and modulation format.

关键词： Fiber nonlinearity Fiber optic communications neural networks Nonlinear impairments Nonlinearity compensation Optical signal to noise ratio

来源：评论

学校读者我要写书评

暂无评论

Automated Detection of Seafloor Gas Seeps in Multibeam Echosounder Data With an Attention-Guided Convolutional neural network

引用

IEEE JOURNAL OF SELECTED TOPICS IN APPLIED EARTH OBSERVATIONS AND REMOTE SENSING 2025年 18卷 5633-5645页

作者： Manjur, Sultan Mohammad Senyurek, Volkan Kalski, Ramon Gupta, Surabhi Skarke, Adam Gurbuz, Ali C. Mississippi State Univ Dept Elect & Comp Engn Mississippi State MS 39672 USA Mississippi State Univ Informat Proc & Sensing Lab Mississippi State MS 39672 USA Mississippi State Univ Geosyst Res Inst Mississippi State MS 39672 USA Mississippi State Univ Dept Geosci Mississippi State MS 39672 USA

Seafloor gas seeps are a globally distributed feature that hold substantial implications for ocean carbon cycling, chemosynthetic ecology, energy production, and geohazards. Detection of seafloor gas seeps currently requires human visual interpretation of sonar water column imagery by a trained individual, which is time consuming, costly, and inconsistent. In this study, we introduce a novel dataset composed of 428 396 high-resolution multibeam echosounder (MBES) sonar water column images annotated for the presence and position of gas bubble plume targets generated by seafloor seeps. Leveraging this dataset for training and validation, we present a machine learning model based on an attention-guided convolutional neural network framework for automated gas seep detection in seafloor environments. In addition, we present an analysis of a range of processing approaches for MBES sonar water column data, evaluating their impact on the seep detection performance of the machine learning model. Finally, our proposed ML framework underwent a rigorous evaluation using a location-independent validation strategy, achieving an overall accuracy of 87%, with 78% precision and 72% recall rate ensuring robustness and reliability. This research represents a substantial advancement in automated gas seep detection methodology.

关键词： Convolutional block attention module (CBAM) convolutional neural network gas seep gas seep machine learning (ML) machine learning (ML) multibeam echosounder (MBES) multibeam echosounder (MBES) sonar sonar transfer learning transfer learning transfer learning

来源：评论

学校读者我要写书评

暂无评论

neural network Compiler for Parallel High-Throughput Simulation of Digital Circuits 37

Neural Network Compiler for Parallel High-Throughput Simulat...

引用

37th IEEE International Parallel and distributed processing Symposium (IPDPS)

作者： Gavier, Ignacio Russell, Joshua Patel, Devdhar Rietman, Edward Siegelmann, Hava Univ Massachusetts Manning Coll Informat & Comp Sci Amherst MA 01003 USA

ISBN: (纸本)9798350337662

Register Transfer Level (RTL) simulation and verification of Digital Circuits are extremely important and costly tasks in the Integrated Circuits industry. While some simulators have incorporated the exploitation of parallelism in the structure of Digital Circuits to run on multi-core CPUs, the maximum throughput they achieve quickly reaches a plateau, as described by Amdahl's Law. Recent research from Nvidia has obtained much higher throughput in simulations using GPUs, highlighting the potential of these devices for Digital Circuit simulation. However, they were required to incorporate sophisticated algorithms to support GPU simulation. In addition, the unbalanced structure of real-life Digital Circuits provides difficulties for processing on multi-threaded devices. In this paper, we present a Digital Circuit compiler that utilizes neural networks to exploit the various parallelisms in RTL simulation, making use of PyTorch, a widely-used neural network framework that facilitate their simulation on GPUs. By using properties of Boolean Functions, we developed a novel algorithm that converts any Digital Circuit into a neural network, and optimization techniques that help in pushing the thread computational capability to the limit. The results show three orders of magnitude higher throughput than Verilator RTL simulator, an improvement of one order of magnitude compared to the state-of-the-art GPU techniques from Nvidia. We believe that the use of neural networks not only provides a significant improvement in simulation and verification tasks in the Integrated Circuits industry, but also opens a line of research for simulators at the logic and physical gate level.

关键词： Integrated Circuits Design and Verification of Digital Circuits RTL Synthesis and Simulation neural network Representation Parallel Computation GPU

来源：评论

学校读者我要写书评

暂无评论

Adaptability Analysis of distributed Motor Measurement and Control System Based on Electronic Information Technology

引用

IEEE ACCESS 2024年 12卷 91851-91862页

作者： Deng, Yuan Xi An Jiao Tong Univ Sch Elect Engn Xian 710049 Peoples R China Pinggao Grp Co Ltd Pingdingshan 467001 Peoples R China

In this paper, the adaptability of distributed motor control system in window motor is investigated, aiming at improving the accuracy and efficiency of the system. The accuracy and efficiency of traditional methods are insufficient. Then, intelligent automation systems combining high-precision instruments and advanced algorithms are used to better simulate real-world situations and enhance noise analysis. The research method combines random position control technology and noise analysis mechanism. The experiment is carried out by simulating various working conditions of window motor to measure the rotation angle error, signal acquisition difference, and sound quality change. The results showed that the rotation angle error was less than 2 degrees, the signal acquisition error was less than 0.8%, and the amplitude error was kept below 2%, indicating that the system had good performance and significantly improved the reliability and accuracy of the test. The results showed that the integration of modern electronic information technology and intelligent algorithm significantly improved the adaptability, accuracy, and noise evaluation ability of distributed motor measurement and control system. The paper lays a foundation for the future development of motor test system and suggests further exploration of machine learning model and real-time data processing technology.

关键词： Motors Motor drives Accuracy Control systems Noise Windows Stability criteria BP neural network distributed motor control system random control sound quality

来源：评论

学校读者我要写书评

暂无评论

Deep photonic reservoir computing based on a distributed feedback laser array

引用

APL PHOTONICS 2025年第2期10卷

作者： Zhou, Changdi Mu, Penghua Huang, Yu Yang, Yigong Zhou, Pei Lau, Kuenyao Li, Nianqiang Soochow Univ Sch Optoelect Sci & Engn Suzhou 215006 Peoples R China Soochow Univ Collaborat Innovat Ctr Suzhou Nano Sci & Technol Suzhou 215006 Peoples R China Soochow Univ Key Lab Adv Opt Mfg Technol Jiangsu Prov Suzhou 215006 Peoples R China Soochow Univ Key Lab Modern Opt Technol Educ Minist China Suzhou 215006 Peoples R China Yantai Univ Inst Sci & Technol Optoelect Informat Yantai 264005 Peoples R China

Photonic reservoir computing (RC) is emerging as a competitive candidate for ultra-fast and energy-efficient neuromorphic computing, noted for its physical compatibility and straightforward training process. However, this widely appreciated form of machine learning typically employs only a single hidden layer with a feedback loop (FL), which essentially constrains the capability for complex task processing and poses challenges to integration. Here, we present an integrated deep photonic RC setup based on a distributed feedback laser array. This setup eliminates the dependence of the FL through the pre-processing termed quasi-convolution coding (QC), where the QC-based RC (QRC) significantly reduces the complexity of the network under the ensured performance. In particular, we extend this QRC into an on-chip deep structure, where this deep QRC (DQRC) exhibits remarkable superiority to its shallow counterparts. We demonstrate the augmented capabilities of DQRC through both simulations and experimental comparisons with QRC, extreme learning machine, and time-delay RC in time-dependent tasks. Moreover, the proposed deep configuration also excels in static image processing. We confirm the potential for simplifying the hardware implementations of deep neural networks, revealing a promising solution to satisfy the urgent demand for high-integration brain-inspired systems.

关键词： distributed feedback lasers

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：