检索结果-内蒙古大学图书馆

SPViT: Accelerate Vision Transformer Inference on Mobile Devices via Adaptive Splitting and Offloading

IEEE Transactions on Mobile computing 2025年

作者： Zhao, Sifan Liu, Tongtong Jin, Hai Yao, Dezhong Huazhong University of Science and Technology National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Wuhan430074 China

The Vision Transformer (ViT), which benefits from utilizing self-attention mechanisms, has demonstrated superior accuracy compared to CNNs. However, due to the expensive computational costs, deploying and inferring ViTs on resource-constrained mobile devices has become a challenge. To resolve this challenge, we conducted an empirical analysis to identify performance bottlenecks in deploying ViTs on mobile devices and explored viable solutions. In this paper, we propose SPViT, an adaptive split and offloading method that accelerates ViT inference on mobile devices. SPViT executes collaborative inference of ViT across available edge devices. We introduce a fine-grained splitting technique for the vision transformer structure. Furthermore, we propose an algorithm based on the Auto Regression model to predict partition latency and adaptive offload partitions. Finally, we design offline and online optimization methods to minimize the computational and communication overhead on each device. Based on real-world prototype experiments, SPViT effectively reduces inference latency by 2.2x to 3.3x across four state-of-the-art models. © 2002-2012 IEEE.

关键词： Regression analysis

来源：评论

学校读者我要写书评

暂无评论

FFCI: A Camera and IMU Sensors Based Multi-modal Neural Network for Activity Recognition in Smart Factory

引用

IEEE Transactions on Consumer Electronics 2025年

作者： Wang, Yujue Niu, Xin Lv, Xianwei Yu, Chen Huazhong University of Science and Technology National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Wuhan430074 China

Worker activity recognition is an important aspect of the construction of smart factory. The development of deep neural networks and the widespread distribution of sensors in the smart factory have brought opportunities for the recognition of workers' activities. The existing methods based on camera and IMU sensors respectively have problems of visual occlusion and difficulty in recognition of similar activities, which result in the reduction of recognition accuracy. Therefore, we propose a feature-level fusion based on camera and IMU sensor (FFCI) for activity recognition in smart factory. The FFCI employs an optimized multi-modal fusion strategy to integrate the visual information and the action information of workers' activities. In order to verify the effectiveness of FFCI, we define a simple yet fine-grained assembly task and collect six common operations. Furthermore, we add visual occlusion and similar activity data to get closer to the smart factory. Finally, we evaluate the FFCI on the collected dataset and achieve a recognition accuracy of 98.17%, demonstrating its effectiveness in accurately classifying workers' activities. © 1975-2011 IEEE.

关键词： Factory automation

来源：评论

学校读者我要写书评

暂无评论

X-Stream: A Flexible, Adaptive Video Transformer for Privacy-Preserving Video Stream Analytics

X-Stream: A Flexible, Adaptive Video Transformer for Privacy...

引用

IEEE Annual Joint Conference: INFOCOM, IEEE Computer and Communications Societies

作者： Dou Feng Lin Wang Shutong Chen Lingching Tung Fangming Liu National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Huazhong University of Science and Technology China National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Paderborn University Germany National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Guangxi University China National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Peng Cheng Laboratory China

ISBN: (数字)9798350383508

ISBN: (纸本)9798350383515

Video stream analytics (VSA) systems fuel many exciting applications that facilitate people’s lives, but also raise critical concerns about exposing too much individuals’ privacy. To alleviate these concerns, various frameworks have been presented to enhance the privacy of VSA systems. Yet, existing solutions suffer two limitations: (1) being scenario-customized, thus limiting the generality of adapting to multifarious scenarios, (2) requiring complex, imperative programming, and tedious process, thus largely reducing the usability of such systems. In this paper, we present X-Stream, a privacy-preserving video transformer that achieves flexibility and efficiency for a large variety of VSA tasks. X-Stream features three major novel designs: (1) a declarative query interface that provides a simple yet expressive interface for users to describe both their privacy protection and content exposure requirements, (2) an adaptation mechanism that dynamically selects the most suitable privacy-preserving techniques and their parameters based on the current video context, and (3) an efficient execution engine that incorporates optimizations for multi-task deduplication and inter-frame inference. We implement X-Stream and evaluate it with representative VSA tasks and public video datasets. The results show that X-Stream achieves significantly improved privacy protection quality and performance over the state-of-the-art, while being simple to use.

关键词： Privacy Transforms Streaming media Transformers Multitasking Real-time systems Task analysis

来源：评论

学校读者我要写书评

暂无评论

Fusion of Natural Language and Knowledge Graph for Multi-hop Reasoning 19th

Fusion of Natural Language and Knowledge Graph for Multi-hop...

引用

19th International Conference on Web Information systems and Applications, WISA 2022

作者： Lu, Xun Zhao, Feng Jin, Hai National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Huazhong University of Science and Technology Wuhan China

ISBN: (纸本)9783031203084

Multi-hop reasoning has been widely studied for its important application values in the domain of intelligent search and question answering. Real-world applications are often dominated by natural language input, and it is difficult to directly model the relation semantics in natural language to fit the multi-hop reasoning model. In addition, the extremely large scale and complex structure of knowledge graphs increase the challenge. We propose a natural language and knowledge graph fusion model (NLKGF) for multi-hop reasoning. NLKGF embeds knowledge graph by fusing natural language semantics during the graph propagation process and adds a relation attention mechanism to enable entity representations to perceive the contribution of different relations. A relation path encoder is designed to encode the relation path by an improved recurrent neural network, the reasoning entity is obtained by calculating the correlation score with the natural language. We tested the performance of the NLKGF model on two datasets requiring multi-hop reasoning. The experimental results show that NLKGF beats advanced benchmark models in multi-hop reasoning tasks, which proves superiority of our model. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Knowledge graph

来源：评论

学校读者我要写书评

暂无评论

Semantic Reasoning technology on Temporal Knowledge Graph 19th

Semantic Reasoning Technology on Temporal Knowledge Graph

引用

19th International Conference on Web Information systems and Applications, WISA 2022

作者： Li, Jianuo Zhao, Feng Jin, Hai National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Huazhong University of Science and Technology Wuhan China

ISBN: (纸本)9783031203084

Semantic reasoning techniques based on knowledge graphs have been widely studied since they were proposed. Previous studies are mostly based on closed-world assumptions, which cannot reason about unknown facts. To this end, we propose the Two-Stage Temporal Reasoning Model (TSTR) for reasoning about future facts. In the first stage, probability of future facts occurring is reasoned using repeated information in history. In the second stage, the semantics of the neighborhood nodes are aggregated using the structural encoder and the temporal information is captured using the temporal encoder. The predicted probabilities are obtained by the decoder. Finally, the candidate entity probabilities of the two-stage reasoning are weighted to achieve the prediction of the two-stage fusion. We tested the performance of the TSTR on public datasets and the results demonstrated the effectiveness of the TSTR. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Recurrent neural networks

来源：评论

学校读者我要写书评

暂无评论

λGrapher: A Resource-Efficient Serverless system for GNN Serving through Graph Sharing 24

λGrapher: A Resource-Efficient Serverless System for GNN Se...

引用

33rd ACM Web Conference, WWW 2024

作者： Hu, Haichuan Liu, Fangming Pei, Qiangyu Yuan, Yongjie Xu, Zichen Wang, Lin National Engineering Research Center for Big Data Technology and System The Services Computing Technology and System Lab Cluster and Grid Computing Lab in the School of Computer Science and Technology Huazhong University of Science and Technology 1037 Luoyu Road Wuhan China Peng Cheng Laboratory Huazhong University of Science and Technology China School of Mathematics and Computer Science Nanchang University China Paderborn University Paderborn Germany

ISBN: (纸本)9798400701719

Graph Neural Networks (GNNs) have been increasingly adopted for graph analysis in web applications such as social networks. Yet, efficient GNN serving remains a critical challenge due to high workload fluctuations and intricate GNN operations. Serverless computing, thanks to its flexibility and agility, offers on-demand serving of GNN inference requests. Alas, the request-centric serverless model is still too coarse-grained to avoid resource waste. Observing the significant data locality in computation graphs of requests, we propose λGrapher, a serverless system for GNN serving that achieves resource efficiency through graph sharing and fine-grained resource allocation. "Grapher features the following designs: (1) adaptive timeout for request buffering to balance resource efficiency and inference latency, (2) graph-centric scheduling to minimize computation and memory redundancy, and (3) resource-centric function management with fine-grained resource allocation catered to the resource sensitivities of GNN operations and function orchestration optimized to hide communication latency. We implement a prototype of λGrapher based on the representative open-source serverless platform Knative and evaluate it with real-world traces from various web applications. Our results show that λGrapher can achieve an average savings of 61.5% in memory resource and 47.2% in computing resource compared with the state of the arts while ensuring GNN inference latency. © 2024 ACM.

关键词： Efficiency

来源：评论

学校读者我要写书评

暂无评论

Towards Stricter Black-box Integrity Verification of Deep Neural Network Models 24

Towards Stricter Black-box Integrity Verification of Deep Ne...

引用

32nd ACM International Conference on Multimedia, MM 2024

作者： He, Chaoxiang Bai, Xiaofan Ma, Xiaojing Zhu, Bin B. Hu, Pingyi Fu, Jiayun Jin, Hai Zhang, Dongmei Huazhong University of Science and Technology Hubei Wuhan China Shanghai Jiao Tong University Shanghai China Microsoft Corporation Beijing China National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Hubei Engineering Research Center on Big Data Security Hubei Key Laboratory of Distributed System Security School of Cyber Science and Engineering Huazhong University of Science and Technology China National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Huazhong University of Science and Technology China

ISBN: (纸本)9798400706868

Cloud-based machine learning services offer significant advantages but also introduce the risk of tampering with cloud-deployed deep neural network (DNN) models. Black-box integrity verification (BIV) allows model owners and end-users to determine if a cloud-deployed DNN model has been tampered with by examining only the top-1 label responses. Fingerprinting generates fingerprint samples to query the model, achieving BIV with no impact on the model's accuracy. In this paper, we present BIVBench, the first comprehensive benchmark for BIV of DNN models. BIVBench covers 16 types of model modifications, providing extensive coverage of practical modification scenarios. Our analysis reveals that existing fingerprinting methods, which are typically focused on significant tampering, lack the sensitivity needed to effectively detect subtle yet common and potentially severe modifications. To address this limitation, we propose MiSentry (Model Integrity Sentry), a novel fingerprinting method that leverages meta-learning. MiSentry strategically incorporates a few subtly modified models into the meta-learning model zoo and maximizes the divergence of output predictions between the target model and the modified models in the model zoo to generate highly sensitive, generalizable, and effective fingerprint samples. Extensive evaluations using BIVBench demonstrate that MiSentry outperforms existing state-of-the-art methods overall and significantly surpasses them in detecting subtle modifications. The BIVBench and supplementary materials are available at: https://***/CGCL-codes/BIVBench. © 2024 ACM.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

Commonsense Knowledge Construction with Concept and Pretrained Model 19th

Commonsense Knowledge Construction with Concept and Pretrain...

引用

19th International Conference on Web Information systems and Applications, WISA 2022

作者： Cai, Hanjun Zhao, Feng Jin, Hai National Engineering Research Center for Big Data Technology and System Services Computing Technology and System Lab Cluster and Grid Computing Lab School of Computer Science and Technology Huazhong University of Science and Technology Wuhan China

ISBN: (纸本)9783031203084

Commonsense knowledge (CSK) is the information that people use in daily life but do not often mention. It summarizes the practical knowledge about how the world works. Existing machines have knowledge but lack commonsense because they do not understand and master commonsense knowledge in the same way that humans do. In the latest works, crowdsourcing-based method is costly and has low coverage, knowledge base completion method can highly fit samples, and methods extracted from unstructured data have the defects of low quality. CG &BF is commonsense knowledge construction with a concept-based generator and a BERT-based filter. We utilize semantic search for node matching and entropy encoder for filtering triples with high abstraction. Two algorithms based on concept aggregation and path credibility are proposed to obtain high-quality CSK triples. We subsequently finetuning a BERT to filter incorrect triples. We obtain 500,000 CSK triples based on ConceptNet, which is superior to other construction methods in novelty and quality. In the reading comprehension task, the three-way attention network is selected as the basic model and the knowledge we generate enables the base model to perform better, which proves that the output of CG &BF has higher quality and ease of use. © 2022, The Author(s), under exclusive license to Springer Nature Switzerland AG.

关键词： Knowledge graph

来源：评论

学校读者我要写书评

暂无评论

RGraph: Asynchronous graph processing based on asymmetry of remote direct memory access

RGraph: Asynchronous graph processing based on asymmetry of ...

引用

作者： Chen, Hanhua Yuan, Jie Jin, Hai Wang, Yonghui Wu, Sijie Jiang, Zhihao National Engineering Research Center for Big Data Technology and System Cluster and Grid Computing Lab Services Computing Technology and System Lab School of Computer Science and Technology Huazhong University of Science and Technology Wuhan China

The scale of real-world graphs is constantly growing. To deal with large-scale graphs, distributed graph processing has attracted much research efforts. Existing distributed graph processing systems are commonly built on traditional TCP/IP communication stack, which leads to network bottleneck because of low bandwidth and heavy kernel stack operations. Meanwhile, in real power-law graphs, the average number of mirror vertices after graph partitioning is very large, resulting in significant communication overhead among nodes. The emerging high-performance Remote Direct Memory Access (RDMA) network has the features of low latency, high bandwidth, and low CPU overhead, which brings new opportunities for distributed graph processing systems. Existing RDMA-assisted graph processing systems focus on synchronous execution, which imposes barriers between consecutive iterations. Synchronous execution transfers bulk data among nodes and thus only needs a small number of network transfers. However, synchronous execution is usually less efficient than asynchronous execution because of bulk synchronization. Asynchronous execution accelerates graph processing by eliminating barriers, which in turn requires to transfer a large amount of small size data. In this paper, we propose RGraph, an RDMA-assisted asynchronous distributed graph processing system. RGraph distributes edges into two parts to isolate master and mirror vertices. RGraph exploits the asymmetry of RDMA to accelerate the one-to-many communication between master and mirror vertices. We implement RGraph on top of PowerGraph and conduct comprehensive experiments with large-scale real graphs to evaluate its performance. Results show that compared to existing designs, RGraph reduces the execution time by up to 81%. © 2021 John Wiley & Sons Ltd.

关键词： Mirrors

来源：评论

学校读者我要写书评

暂无评论

MeG2: In-Memory Acceleration for Genome Graphs Analysis 23

MeG2: In-Memory Acceleration for Genome Graphs Analysis

引用

Proceedings of the 60th Annual ACM/IEEE Design Automation Conference

作者： Yu Huang Long Zheng Haifeng Liu Zhuoran Zhou Dan Chen Pengcheng Yao Qinggang Wang Xiaofei Liao Hai Jin National Engineering Research Center for Big Data Technology and System/Services Computing Technology and System Lab/Cluster and Grid Computing Laboratory Huazhong University of Science and Technology Wuhan China and Zhejiang Lab Hangzhou China National Engineering Research Center for Big Data Technology and System/Services Computing Technology and System Lab/Cluster and Grid Computing Laboratory Huazhong University of Science and Technology Wuhan China

ISBN: (纸本)9798350323481

Genome graphs analysis has emerged as an effective means to enable mapping DNA fragments (known as reads) to the reference genome. It replaces the traditional linear reference with a graph-based representation to augment the genetic variations and diversity information, significantly improving the quality of genotyping. The in-depth characterization of genome graphs analysis uncovers that it is bottlenecked by the irregular seed index access and the intensive alignment operation, stressing both the memory system and computing *** on these observations, we propose MeG2, a lightweight, commodity DRAM-compliant, processing-in-memory architecture to accelerate genome graphs analysis. MeG2 is specifically integrated with the capabilities of both near-memory processing and bitwise in-situ computation. Specifically, MeG2 leverages the low access latency of near-memory processing with the index-centric offload mechanism to alleviate the irregular memory access in the seeding procedure, and harnesses the row-parallel capacity of in-situ computation with the distance-aware technique to exploit the intensive computational parallelism in the alignment process. Results show that MeG2 outperforms the CPU-, GPU-, and ASIC-based genome graphs analysis solutions by 502× (30.2×), 272× (15.1×), and 5.5× (8.3×) for short (long) reads, while reducing energy consumption by 1628× (85.6×), 1443× (77.1×), and 7.8× (11.7×), respectively. We also demonstrate that MeG2 offers significant improvements over existing PIM-based genome sequence analysis accelerators.

关键词：

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：