检索结果-内蒙古大学图书馆

Attar: RRAM-based in-memory attention accelerator with software-hardware co-optimization

Science China(Information Sciences) 2025年第3期68卷 371-387页

作者： Bing LI Ying QI Ying WANG Yinhe HAN Information Engineering College Capital Normal University Research Center for Intelligent Computing Systems Institute of Computing TechnologyChinese Academy of Sciences

The attention mechanism has become a pivotal component in artificial intelligence, significantly enhancing the performance of deep learning applications. However, its quadratic computational complexity and intricate computations lead to substantial inefficiencies when processing long sequences. To address these challenges, we introduce Attar, a resistive random access memory(RRAM)-based in-memory accelerator designed to optimize attention mechanisms through software-hardware co-optimization. Attar leverages efficient Top-k pruning and quantization strategies to exploit the sparsity and redundancy of attention matrices, and incorporates an RRAM-based in-memory softmax engine by harnessing the versatility of the RRAM crossbar. Comprehensive evaluations demonstrate that Attar achieves a performance improvement of up to 4.88× and energy saving of 55.38% over previous computing-in-memory(CIM)-based accelerators across various models and datasets while maintaining comparable accuracy. This work underscores the potential of in-memory computing to enhance the efficiency of attention-based models without compromising their effectiveness.

关键词： RRAM computing-in-memory attention pruning quantization

来源：评论

学校读者我要写书评

暂无评论

Pyramid: Accelerating LLM Inference With Cross-Level Processing-in-Memory

IEEE Computer Architecture Letters

引用

IEEE Computer Architecture Letters 2025年第1期24卷 121-124页

作者： Yan, Liang Lu, Xiaoyang Chen, Xiaoming Han, Yinhe Sun, Xian-He Chinese Academy of Sciences Intelligent Computing Systems Institute of Computing Technology Beijing100190 China University of Chinese Academy of Sciences Beijing100190 China Illinois Institute of Technology Department of Compute Science ChicagoIL60616 United States

Integrating processing-in-memory (PIM) with GPUs accelerates large language model (LLM) inference, but existing GPU-PIM systems encounter several challenges. While GPUs excel in large general matrix-matrix multiplications (GEMM), they struggle with small-scale operations better suited for PIM, which currently cannot handle them independently. Additionally, the computational demands of activation operations exceed the capabilities of current PIM technologies, leading to excessive data movement between the GPU and memory. PIM's potential for general matrix-vector multiplications (GEMV) is also limited by insufficient support for fine-grained parallelism. To address these issues, we propose Pyramid, a novel GPU-PIM system that optimizes PIM for LLM inference by strategically allocating cross-level computational resources within PIM to meet diverse needs and leveraging the strengths of both technologies. Evaluation results demonstrate that Pyramid outperforms existing systems like NeuPIM, AiM, and AttAcc by factors of 2.31×, 1.91×, and 1.72×, respectively. © 2002-2011 IEEE.

关键词： Graphics processing unit

来源：评论

学校读者我要写书评

暂无评论

Evolution Analysis of Network Attack and Defense Situation Based on Game Theory

引用

Computers, Materials & Continua 2025年第4期83卷 1475-1494页

作者： Haiyan Sun Chenglong Shao Jianwei Zhang Kun Wang Wanwei Huang College of Software Engineering Zhengzhou University of Light IndustryZhengzhou450007China Intelligent Computing Power Research Department Zhengzhou Xinda Advanced Technology Research InstituteZhengzhou450007China

To address the problem that existing studies lack analysis of the relationship between attack-defense game behaviors and situation evolution from the game perspective after constructing an attack-defense model,this paper proposes a network attack-defense game model(ADGM).Firstly,based on the assumption of incomplete information between the two sides of the game,the ADGM model is established,and methods of payoff quantification,equilibrium solution,and determination of strategy confrontation results are ***,drawing on infectious disease dynamics,the network attack-defense situation is defined based on the density of nodes in various security states,and the transition paths of network node security states are ***,the network zero-day virus attack-defense behaviors are analyzed,and comparative experiments on the attack-defense evolution trends under the scenarios of different strategy combinations,interference methods,and initial numbers are conducted using the NetLogo simulation *** experimental results indicate that this model can effectively analyze the evolution of the macro-level network attack-defense situation from the micro-level attack-defense *** instance,in the strategy selection experiment,when the attack success rate decreases from 0.49 to 0.29,the network destruction rate drops by 11.3%,in the active defense experiment,when the interference coefficient is reduced from 1 to 0.7,the network destruction rate decreases by 7%,and in the initial node number experiment,when the number of initially infected nodes increases from 10 to 30,the network destruction rate rises by 3%.

关键词： Network attack-defense situation evolution zero-day virus NetLogo

来源：评论

学校读者我要写书评

暂无评论

A Multi-Hyperparameter Prediction Framework for Distributed Energy Trading on Photovoltaic Network

引用

Tsinghua Science and technology 2025年第2期30卷 864-874页

作者： Chun Chen Yong Zhang Boon Han Lim Li Ning Shengzhong Feng Peng Xie School of Intelligent Manufacturing and Equipment Shenzhen Institute of Information TechnologyShenzhen 518172China Shenzhen Institute of Advanced Technology Chinese Academy of SciencesShenzhen518055China Lee Kong Chian Faculty of Engineering and Science Universiti Tunku Abdul RahmanKajang 43000Malaysia Shenzhen Institute for Advanced Study Shenzhen 518028China Shenzhen Institute of Advanced Technology Chinese Academy of SciencesBeijing 100864China High-performance Intelligent Computing Research Group Guangdong Institute of Intelligent Science and TechnologyZhuhai 519031China

The rapid evolution of distributed energy resources,particularly photovoltaic systems,poses a formidable challenge in maintaining a delicate balance between energy supply and demand while minimizing *** integrated nature of distributed markets,blending centralized and decentralized elements,holds the promise of maximizing social welfare and significantly reducing overall costs,including computational and communication ***,achieving this balance requires careful consideration of various hyperparameter sets,encompassing factors such as the number of communities,community detection methods,and trading mechanisms employed among *** address this challenge,we introduce a groundbreaking neural network-based framework,the Energy Trading-based Artificial Neural Network(ET-ANN),which excels in performance compared to existing *** experiments underscore the superiority of ET-ANN in minimizing total energy transaction costs while maximizing social welfare within the realm of photovoltaic networks.

关键词： photovoltaic energy trading hyperparameter prediction

来源：评论

学校读者我要写书评

暂无评论

LOGIN: A Large Language Model Consulted Graph Neural Network Training Framework 25

引用

18th ACM International Conference on Web Search and Data Mining, WSDM 2025

作者： Qiao, Yiran Ao, Xiang Liu, Yang Xu, Jiarong Sun, Xiaoqian He, Qing Institute of Computing Technology CAS Beijing China Key Lab of AI Safety Chinese Academy of Sciences Institute of Computing Technology CAS Beijing China University of Chinese Academy of Sciences CAS Beijing China School of Management Fudan University Shanghai China Institute of Intelligent Computing Technology Suzhou China

ISBN: (纸本)9798400713293

Recent prevailing works on graph machine learning typically follow a similar methodology that involves designing advanced variants of graph neural networks (GNNs) to maintain the superior performance of GNNs on different graphs. In this paper, we aim to streamline the GNN design process and leverage the advantages of Large Language Models (LLMs) to improve the performance of GNNs on downstream tasks. We formulate a new paradigm, coined "LLMs-as-Consultants", which integrates LLMs with GNNs in an interactive manner. A framework named LOGIN (LLM cOnsulted GNN traINing) is instantiated, empowering the interactive utilization of LLMs within the GNN training process. First, we attentively craft concise prompts for spotted nodes, carrying comprehensive semantic and topological information, and serving as input to LLMs. Second, we refine GNNs by devising a complementary coping mechanism that utilizes the responses from LLMs, depending on their correctness. We empirically evaluate the effectiveness of LOGIN on node classification tasks across both homophilic and heterophilic graphs. The results illustrate that even basic GNN architectures, when employed within the proposed LLMs-as-Consultants paradigm, can achieve comparable performance to advanced GNNs with intricate designs. Our code is available at https://***/QiaoYRan/LOGIN. © 2025 Copyright held by the owner/author(s).

关键词： Graph neural networks

来源：评论

学校读者我要写书评

暂无评论

Machine learning empowered UAV-based beamforming design in ISAC systems

引用

Science China(Information Sciences) 2025年第5期68卷 285-286页

作者： Xiangyang DUAN Xiaoqi ZHANG Shuqiang XIA Zhongbin WANG Yihua MA Weijie YUAN State Key Laboratory of Mobile Network and Mobile Multimedia Technology Wireless and Computing Research InstituteZTE Corporation School of Electrical and Data Engineering University of Technology Sydney School of Automation and Intelligent Manufacturing Southern University of Science and Technology

Integrated sensing and communication (ISAC) has been proposed as an enabling technology for the realization of the next-generation wireless system,which focuses on performing wireless communication and sensing *** the various potential ISAC-based applications,unmanned aerial vehicle (UAV)-based ISAC plays a significant part in unlocking the potential of future next-generation wireless communication,facilitating low-latency data transmission in high-mobility *** by recent advancements,a variety of effective techniques have been investigated to optimize beamforming design in ISAC *** instance,the authors in [1] introduced an extended Kalman filtering (EKF)-based method tailored for millimeter wave (mmWave) ISAC ***,Ref.[2]proposed an extended interacting multiple model (IMM)-EKF framework designed for vehicular networks with intricate roadway *** these advancements,the aforementioned methods typically employ a separate scheme for channel prediction and beam alignment,which introduces additional signaling overhead in real-world ***,there is a demand for an end-to-end beamforming design approach specifically for UAV-based ISAC systems.

关键词：

来源：评论

学校读者我要写书评

暂无评论

RIS Enabled SCLAM: An Approach for Simultaneous Radio Communication, Localization, and Mapping

引用

IEEE Internet of Things Journal 2025年第10期12卷 14663-14676页

作者： Zhao, Jinqiu Bai, Zhiquan Guo, Shuaishuai Ma, Dejie Li, Na Kwak, Kyung Sup Shandong University Shandong Key Laboratory of Intelligent Communication and Sensing-Computing Integration School of Information Science and Engineering Qingdao266237 China Shandong University Shandong Key Laboratory of Intelligent Communication and Sensing-Computing Integration School of Control Science and Engineering Jinan250061 China Qilu Institute of Technology School of Intelligent Manufacturing and Control Engineering Jinan250200 China INHA University Graduate School of Information Technology and Telecommunications Incheon22212 Korea Republic of

Radio-based simultaneous localization and mapping (SLAM) facilitates unmanned systems to fulfill self-localization and navigation in complex environments. However, the existing studies overlooked communication between devices, which may lead to low efficiency and high costs when performing SLAM tasks. Reconfigurable intelligent surface (RIS) can satisfy the growing demands of users and improve SLAM accuracy in dynamic and complex environments. This paper proposes a RIS enabled simultaneous radio communication, localization, and mapping (SCLAM) system, where an unmanned aerial vehicle (UAV) can perform concurrent communication and SLAM with the help of RIS by utilizing communication signals from base station (BS). The proposed method enables the UAV to simultaneously accomplish SLAM and ensure communication with the BS, while also enhancing the spectrum utilization efficiency. We derive the Bayesian Fisher information matrix (BFIM) for joint SLAM and symbol detection of the proposed system, in which the trade-off between the BFIM of SLAM and the upper bound of the ergodic mutual information is illustrated. Then, a weighted factor based BFIM is presented to further achieve a performance trade-off between SLAM and communication. We formulate an optimization problem of joint BS active beamforming and RIS passive beamforming to maximize the log determinant of weighted BFIM. Numerical results verify the superiority of the proposed SCLAM system on position error bound (PEB), mapping error bound (MEB), and spectral efficiency (SE). The performance trade-off between communication and SLAM is also discussed and explored. © 2014 IEEE.

关键词： Unmanned aerial vehicles (UAV)

来源：评论

学校读者我要写书评

暂无评论

Extract features from lithium-ion battery electrochemical impedance spectra and estimate state of health based on improved convolutional autoencoder-temporal convolutional network

引用

Ionics 2025年第5期31卷 4261-4279页

作者： Lou, Cheng Wang, Shi Li, Zhaoting Wang, Kai School of Electrical Engineering Weihai Innovation Research Institute Qingdao University Qingdao266000 China Key Laboratory of Intelligent Information Processing Institute of Computing Technology Chinese Academy of Sciences Beijing China School of Engineering Brown University ProvidenceRI02912 United States Shandong Suoxiang Intelligent Technology Co. Ltd Weifang261101 China

Predicting the state of health (SOH) of batteries is crucial for understanding their remaining lifespan and formulating more effective maintenance and management strategies. Utilizing electrochemical impedance spectroscopy (EIS) to assess the health status of batteries offers advantages such as high precision, rapid response, non-invasiveness, and reliability for accurately forecasting the remaining lifespan of batteries. In this paper, we propose an innovative method combining electrochemical impedance spectroscopy data to efficiently recognize and process complex patterns using deep neural networks by converting EIS into a two-dimensional image format. We have developed an improved convolutional autoencoder (ICAE) optimized to extract key features directly related to battery capacity in 2D EIS images and significantly improve feature characterization. The optimized features are further fed into the temporal convolutional network (TCN) to perform SOH prediction tasks. TCN utilizes its powerful time-dependent capture capability and long sequence memory mechanism to demonstrate superior performance in the field of SOH estimation. Compared with traditional methods, the proposed strategy not only significantly increases the prediction accuracy, but also opens up a new way to understand and analyze the internal relationship between complex time series and image data. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2025.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

Optimizing Maize Cultivation: A Vision-Based AI-Driven Methodology for Automated Seedling Thinning

IEEE Transactions on AgriFood Electronics

引用

IEEE Transactions on AgriFood Electronics 2025年第1期3卷 224-232页

作者： Wang, Zijian An, Xiaofei Wang, Ling Tang, Jinshan Harbin Institute of Technology Department of Mathematics Harbin150001 China Beijing Academy of Agriculture and Forestry Sciences Intelligent Equipment Technology Research Center Beijing100097 China Harbin Institute of Technology Faculty of Computing Harbin150001 China George Maon University FairfaxVA22031 United States

To address the challenges of traditional manual maize seedling thinning, this article proposes an innovative approach that utilizes computer vision and deep learning for automated thinning. A zero-shot keypoint annotation algorithm, leveraging segment for anything model, is designed to label large datasets of maize seedling centers without requiring training samples. We also propose an improved hourglass network that significantly enhances seedling center positioning accuracy, enabling precise thinning decisions. Furthermore, a novel automatic thinning decision algorithm is devised to determine optimal removal strategies, ensuring ideal plant-to-plant spacing. The system's performance was evaluated against manually annotated data from 1020 images encompassing 2756 individual maize seedlings collected from farms. Impressively, the algorithm achieved a precision rate of 98.84%, confirming its ability to identify seedlings for removal while preserving healthy plants accurately. Evaluations of the keypoint detection network at a threshold of 0.2 yielded a percentage of correct keypoints of 97.66% and an object keypoint similarity of 0.87, surpassing existing methods and demonstrating the model's superior performance. © 2023 IEEE.

关键词： Zero-shot learning

来源：评论

学校读者我要写书评

暂无评论

Parallel Attention Based Network for Human Activity Recognition Using Wearable Devices 27th

Parallel Attention Based Network for Human Activity Recogni...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Xu, Chenyang Fan, Feiyi Ke, Guanzhou Guo, Changru Wu, Qingyu Shen, Jianfei Institute of Computing Technology Chinese Academy of Sciences Beijing100090 China School of Electrical and Information Engineering Tianjin University Tianjin300072 China Beijing Jiaotong University Beijing100044 China Jinan Zhongke Ubiquitous-Intelligent Institute of Computing Technology Shandong Jinan250102 China

ISBN: (纸本)9783031782008

In recent years, significant progress has been made in improving the efficacy of wearable human activity recognition (HAR) tasks using deep learning technology. Existing research indicates that stacked convolutional layers effectively extract high-semantic signal features from multi-sensor channel time-series data. However, these approaches disregard the fact that sensor signals are low-semantic and sensor-based deep networks lead to overfitting and gradient vanishing. In this paper, we present the parallel attention-based HAR (PA-HAR) method. Our method employs multiple small-scale receptive fields to extract low-semantic signals in parallel and a skip-squeeze excitation block to establish correlations among multi-feature maps based on the feature channel dimension. We also introduce smooth and non-monotonic sigmoid linear units (SiLU) to integrate multi-scale and cross-channel features in order to prevent the loss of non-linear information due to small-scale receptive fields and reduce representational ability loss. Extensive experiments on seven public datasets show that our proposed PA-HAR model outperforms state-of-the-art approaches in HAR tasks. In addition, we develop a wearable real-time activity recognition system based on the embedded device with our model. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Deep learning

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：