检索结果-内蒙古大学图书馆

FlexPDA:A Flexible Programming Framework for Deep Learning Accelerators

Journal of computer science & technology 2022年第5期37卷 1200-1220页

作者： Lei Liu Xiu Ma Hua-xiao Liu Guang-li Li Lei Liu College of Computer Science and Technology Jilin UniversityChangchunChina Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin UniversityChangchunChina State Key Laboratory of Computer Architecture Institute of Computing TechnologyChinese Academy of SciencesBeijingChina University of Chinese Academy of Sciences BeijingChina

There are a wide variety of intelligence accelerators with promising performance and energy efficiency,deployed in a broad range of applications such as computer vision and speech ***,programming productivity hinders the deployment of deep learning *** low-level library invoked in the high-level deep learning framework which supports the end-to-end execution with a given model,is designed to reduce the programming burden on the intelligence ***,it is inflexible for developers to build a network model for every deep learning application,which probably brings unnecessary repetitive *** this paper,a flexible and efficient programming framework for deep learning accelerators,FlexPDA,is proposed,which provides more optimization opportunities than the low-level library and realizes quick transplantation of applications to intelligence accelerators for fast *** evaluate FlexPDA by using 10 representative operators selected from deep learning algorithms and an end-to-end *** experimental results validate the effectiveness of FlexPDA,which achieves an end-to-end performance improvement of 1.620x over the low-level library.

关键词： deep learning accelerator programming framework domain-specific language

来源：评论

学校读者我要写书评

暂无评论

Scalable parallel photonic processing unit for various neural network accelerations

引用

Photonics Research 2024年第11期12卷 2691-2702页

作者： SHIYIN DU JUN ZHANG HAO OUYANG ZILONG TAO QIUQUAN YAN HAO HAO JUNHU ZHOU JIE YOU YUHUA TANG TIAN JIANG State Key Laboratory of High-Performance Computing College of ComputerNational University of Defense TechnologyChangsha 410073China National Innovation Institute of Defense Technology Academy of Military Sciences PLA ChinaBeijing 100071China College of Advanced Interdisciplinary Studies National University of Defense TechnologyChangsha 410073China Institute for Quantum Science and Technology College of ScienceNational University of Defense TechnologyChangsha 410073China

In recent years,integrated optical processing units(IOPUs)have demonstrated advantages in energy efficiency and computational speed for neural network inference ***,limited by optical integration technology,the practicality and versatility of IOPU face serious *** this work,a scalable parallel photonic processing unit(SPPU)for various neural network accelerations based on high-speed phase modulation is proposed and implemented on a silicon-on-insulator platform,which supports parallel processing and can switch between multiple computational paradigms simply and without latency to infer different neural network structures,enabling to maximize the utility of on-chip *** SPPU adopts a scalable and process-friendly architecture design,with a preeminent photonic-core energy efficiency of 0.83 TOPS/W,two to ten times higher than existing integrated *** the proof-of-concept experiment,a convolutional neural network(CNN),a residual CNN,and a recurrent neural network(RNN)are all implemented on our photonic processor to handle multiple tasks of handwritten digit classification,signal modulation format recognition,and review emotion *** SPPU achieves multi-task parallel processing capability,serving as a promising and attractive research route to maximize the utility of on-chip components under the constraints of integrated technology,which helps to make IOPU more practical and universal.

关键词： network acceleration attractive

来源：评论

学校读者我要写书评

暂无评论

DNS Service Optimization Through In-Network Caching on Programmable Switches

DNS Service Optimization Through In-Network Caching on Progr...

引用

International Conference on computer and Communications (ICCC)

作者： Fan Yang Cenman Wang Bowen Zheng Ye Tian Anhui Key Laboratory on High Performance Computing School of Computer Science and Technology University of Science and Technology of China Hefei China

ISBN: (数字)9798331507077

ISBN: (纸本)9798331507084

The Domain Name System (DNS) plays a crucial role in contemporary internet infrastructure. Despite the geographical distance of public DNS servers posing a challenge in minimizing domain name resolution latency, there is an increasing reliance on these servers for resolving domain names. One effective approach to mitigate this challenge involves caching DNS query results at network nodes. The storage of cached results commonly employs hash tables due to the variable lengths of domain names. However, within network nodes, hash tables encounter limitations in capacity and susceptibility to hash collisions, potentially leading to inaccuracies in cached results. Current conventional caching strategies have not produced satisfactory storage outcomes. This research introduces a hash-based DNS caching mechanism to tackle these issues. The proposed mechanism presents a unique DNS storage structure on routers and suggests two implementation strategies that incorporate fingerprint verification and Bloom filters. Through analysis and extensive simulations utilizing real-world network topology, the results demonstrate that the hash-based DNS caching mechanism introduced in this study outperforms existing methods in cache hit rate and diminishes query latency.

关键词： Analytical models Filters Network topology Computational modeling Fingerprint recognition Cache storage Topology Domain Name System Servers Optimization

来源：评论

学校读者我要写书评

暂无评论

HSToR: Enhancing Scalability for Optical Circuit Switched Networks with Recirculation

HSToR: Enhancing Scalability for Optical Circuit Switched Ne...

引用

International Conference on computer and Communications (ICCC)

作者： Bowen Zheng Fan Yang Cenman Wang Ye Tian Anhui Key Laboratory on High Performance Computing School of Computer Science and Technology University of Science and Technology of China Hefei China

ISBN: (数字)9798331507077

ISBN: (纸本)9798331507084

Combining optical circuit switches (OCSes) and electrical packet switches (EPSes) has emerged as an appealing architecture for data centers. However, in this hybrid data center network architecture with widespread application, challenges persist in data packet scheduling and buffer management for OCS-based matching patterns due to spatial and temporal traffic pattern constraints. Given the demand-agnostic way for OCS configurations, where each switch can only provide matching between input and output ports at any given time. In other words, the availability of optical circuits is intermittent, resulting in potential blockage of flows between them when optical circuits are unavailable. Traditionally, in the event of flow blockage, multiple queue buffers are maintained at Top-of- Rack (ToR) switches, with the number of queues corresponding to the number of destinations. Unfortunately, the “per-destination queue” approach suffers from poor scalability. To address this issue, we propose HSToR, a novel design for ToR switches, aimed at enhancing scalability and data transfer efficiency in optical circuit switched networks. Experimental results demonstrate that HSToR achieves high scalability, maintains lower latency as the network scales, better addresses packet loss during buffering, and sustains high throughput and link utilization.

关键词： Integrated optics Scheduling algorithms Optical switches Scalability Optical buffering Packet loss Optical fiber networks Throughput Optical packet switching Switching circuits

来源：评论

学校读者我要写书评

暂无评论

Accurate Detection of Superspreaders on Multi-Pipeline Switches

Accurate Detection of Superspreaders on Multi-Pipeline Switc...

引用

International Conference on computer and Communications (ICCC)

作者： Cenman Wang Bowen Zheng Fan Yang Ye Tian Anhui Key Laboratory on High Performance Computing School of Computer Science and Technology University of Science and Technology of China Hefei China

ISBN: (数字)9798331507077

ISBN: (纸本)9798331507084

In various network monitoring tasks, the detection of superspreaders has become increasingly crucial. However, latest superspreader detection algorithms and data structures assume the target switch with a single pipeline, while state-of-the-art programmable switches actually comprise multiple independent pipelines. In current method for deploying super-spreader detection applications on multi-pipeline switches, a separate measurement data structure instance is employed for each pipeline. However, under multi-path routing, this simplistic approach leads to lower detection accuracy due to hash collisions. To address this issue, in this paper, we present PipelineAlloc, a HyperLogLog-based superspreader detection system designed for multi-pipeline switches. In order to reduce hash collisions in measurement data structures, PipelineAlloc assigns an egress pipeline as the monitor pipeline for each source IP address. PipelineAlloc introduces novel measurement data structures in the ingress pipelines for temporarily storing statistics of the source and destination IP addresses of packets, and uses packets to carry these measurement data to egress pipelines, which serve as the monitor pipelines. The HyperLogLog-based measurement data structures in egress pipelines are updated using data carried by packets. Experiments driven by real-world traffic traces show that PipelineAlloc outperforms existing solutions in terms of detection accuracy.

关键词： Accuracy Scalability Current measurement Pipelines Switches Data structures Routing IP networks Detection algorithms Monitoring

来源：评论

学校读者我要写书评

暂无评论

Multi-Grained Feature Pruning for Video-Based Human Pose Estimation

Multi-Grained Feature Pruning for Video-Based Human Pose Est...

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Wang, Zhigang Fan, Shaojing Liu, Zhenguang Wu, Zheqi Wu, Sifan Jiao, Yingying College of Computer Science and Technology Zhejiang Gongshang University Hangzhou China School of Computing National University of Singapore Singapore The State Key Laboratory of Blockchain and Data Security Zhejiang University Hangzhou China Institute of Blockchain and Data Security Hangzhou China College of Computer Science and Technology Jilin University Changchun China College of Computer Science and Technology Zhejiang University of Technology Hangzhou China

Human pose estimation, with its broad applications in action recognition and motion capture, has experienced significant advancements. However, current Transformer-based methods for video pose estimation often face challenges in managing redundant temporal information and achieving fine-grained perception because they only focus on processing low-resolution features. To address these challenges, we propose a novel multi-scale resolution framework that encodes spatio-temporal representations at varying granularities and executes fine-grained perception compensation. Furthermore, we employ a density peaks clustering method to dynamically identify and prioritize tokens that offer important semantic information. This strategy effectively prunes redundant feature tokens, especially those arising from multi-frame features, thereby optimizing computational efficiency without sacrificing semantic richness. Empirically, it sets new benchmarks for both performance and efficiency on three large-scale datasets. Our method achieves a 93.8% improvement in inference speed compared to the baseline, while also enhancing pose estimation accuracy, reaching 87.4 mAP on the PoseTrack2017 dataset. © 2025 IEEE.

关键词： Large datasets

来源：评论

学校读者我要写书评

暂无评论

Online Demand Peak Shaving with Machine-Learned Advice in Cyber-Physical Energy Systems

Online Demand Peak Shaving with Machine-Learned Advice in Cy...

引用

2023 IEEE International Conference on Dependable, Autonomic and Secure computing, 2023 International Conference on Pervasive Intelligence and computing, 2023 International Conference on Cloud and Big Data computing, 2023 International Conference on Cyber science and technology Congress, DASC/PiCom/CBDCom/CyberSciTech 2023

作者： Fengl, Minxi Li, Wei Qin, Boyu Zomaya, Albert Y. Centre for Distributed and High Performance Computing School of Computer Science The University of Sydney Australia Xi'an Jiaotong University State Key Laboratory of Electrical Insulation and Power Equipment Department of Electrical Engineering China

ISBN: (纸本)9798350304602

With the deep integration of cyber tools, control algorithms are increasingly employed in cyber-physical energy systems to enhance management, cost efficiency, and robustness. Effective demand load management is crucial in cyber-physical energy systems as it directly impacts operational costs. Failure to adequately manage spiky or seasonal demand loads can lead to significant expenses on monthly utility bills. In this study, we propose AMPAMOD, a randomized online algorithm with machine-learned advice, to optimize battery operations and mitigate highly dynamic peak loads. AMPAMOD utilizes limited advice from machine learning algorithms to guide our online algorithm and ensure cost-effective peak load management. The theoretical analysis shows that our solution has minimal advice complexity, a linear computational cost, and an improved competitive ratio. Finally, we conduct extensive trace-driven experiments on real-world datasets. AMPAMOD achieves a peak shaving success rate of over 90%, outperforming baselines by at least 50%. These experimental results confirm theoretical findings and demonstrate the potential of AMPAMOD. © 2023 IEEE.

关键词： Cyber Physical System

来源：评论

学校读者我要写书评

暂无评论

Thermally tunable high Q-factor and low-power silicon resonators using Graphene transparent electrodes

Thermally tunable high Q-factor and low-power silicon resona...

引用

2023 Applied Optics and Photonics China: Laser technology and Applications;and Optoelectronic Devices and Integration, AOPC 2023

作者： Hong, Qilin Xu, Ping Xu, Wei Zhu, Zhihong College of Advanced Interdisciplinary Studies Hunan Provincial Key Laboratory of Novel Nano Optoelectronic Information Materials and Devices National University of Defense Technology Changsha410073 China Nanhu Laser Laboratory National University of Defense Technology Changsha410073 China Institute for Quantum Information State Key Laboratory of High-Performance Computing College of Computer Science and Technology National University of Defense Technology Changsha410073 China

ISBN: (纸本)9781510672246

Tunable and low-power microcavities play a vital role in facilitating the development of large-scale photonic integrated circuits. Among various tuning methods, thermal tuning has gained significant popularity due to its convenience and stability, especially in the fields of optical neural networks and quantum information processing. In recent years, graphene thermal tuning has emerged as a promising technique, offering both tunability and power efficiency by eliminating the need for thick spacers to prevent light absorption. In this study, we propose and fabricate a silicon-based on-chip Fano resonator with graphene nano-heaters. This innovative Fano structure incorporates a scattering block and can be easily manufactured in large quantities. Experimental results demonstrate that the resonator exhibits desirable characteristics, including a high quality factor of approximately 31000 and a low state-switching power consumption of around 1 mW. The temporal responses of the microcavity exhibit satisfactory modulation speed, with a rise time of 9.8 µs and a fall time of 16.6 µs. The findings of this research offer an alternative solution for the future development of large-scale tunable and low-power-consumption optical networks, with potential applications in optical filters and switches. © 2023 SPIE.

关键词： Graphene

来源：评论

学校读者我要写书评

暂无评论

Towards robust neural networks via a global and monotonically decreasing robustness training strategy

引用

Frontiers of Information technology & Electronic Engineering 2023年第10期24卷 1375-1389页

作者： Zhen LIANG Taoran WU Wanwei LIU Bai XUE Wenjing YANG Ji WANG Zhengbin PANG Institute for Quantum Information&State Key Laboratory of High Performance Computing National University of Defense TechnologyChangsha 410073China State Key Laboratory of Computer Science Institute of Software Chinese Academy of SciencesBeijing 100190China School of Computer Science and Technology University of Chinese Academy of SciencesBeijing 100190China College of Computer Science and Technology National University of Defense TechnologyChangsha 410073China Key Laboratory of Software Engineering for Complex Systems National University of Defense TechnologyChangsha 410073China

Robustness of deep neural networks(DNNs)has caused great concerns in the academic and industrial communities,especially in safety-critical *** of verifying whether the robustness property holds or not in certain neural networks,this paper focuses on training robust neural networks with respect to given ***-of-the-art training methods,interval bound propagation(IBP)and CROWN-IBP,perform well with respect to small perturbations,but their performance declines significantly in large perturbation cases,which is termed“drawdown risk”in this ***,drawdown risk refers to the phenomenon that IBPfamily training methods cannot provide expected robust neural networks in larger perturbation cases,as in smaller perturbation *** alleviate the unexpected drawdown risk,we propose a global and monotonically decreasing robustness training strategy that takes multiple perturbations into account during each training epoch(global robustness training),and the corresponding robustness losses are combined with monotonically decreasing weights(monotonically decreasing robustness training).With experimental demonstrations,our presented strategy maintains performance on small perturbations and the drawdown risk on large perturbations is alleviated to a great *** is also noteworthy that our training method achieves higher model accuracy than the original training methods,which means that our presented training strategy gives more balanced consideration to robustness and accuracy.

关键词： Robust neural networks Training method Drawdown risk Global robustness training Monotonically decreasing robustness

来源：评论

学校读者我要写书评

暂无评论

DISCRETE ENERGY ANALYSIS OF THE THIRD-ORDER VARIABLE-STEP BDF TIME-STEPPING FOR DIFFUSION EQUATIONS

引用

Journal of Computational Mathematics 2023年第2期41卷 325-344页

作者： Hong-lin Liao Tao Tang Tao Zhou Department of Mathematics Nanjing University of Aeronautics and AstronauticsNanjing 211106China Key Laboratory of Mathematical Modelling and High Performance Computing of Air Vehicles(NUAA) MIITNanjing 211106China Division of Science and Technology BNU-HKBU United International CollegeZhuhai 519087China Institute of Computational Mathematics and Scientific/Engineering Computing Academy of Mathematics and Systems ScienceChinese Academy of SciencesBeijing100190China

This is one of our series works on discrete energy analysis of the variable-step BDF *** this part,we present stability and convergence analysis of the third-order BDF(BDF3)schemes with variable steps for linear diffusion equations,see,e.g.,[SIAM ***.,58:2294-2314]and[***.,90:1207-1226]for our previous works on the BDF2 *** this aim,we first build up a discrete gradient structure of the variable-step BDF3 formula under the condition that the adjacent step ratios are less than 1.4877,by which we can establish a discrete energy dissipation ***-robust stability and convergence analysis in the L^(2) norm are then *** the mesh robustness means that the solution errors are well controlled by the maximum time-step size but independent of the adjacent time-step *** also present numerical tests to support our theoretical results.

关键词： Diffusion equations Variable-step third-order BDF scheme Discrete gradient structure Discrete orthogonal convolution kernels Stability and convergence

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：