检索结果-内蒙古大学图书馆

TSCompiler: efficient compilation framework for dynamic-shape models

science China(Information sciences) 2024年第10期67卷 67-84页

作者： Xiang LUO Chen ZHANG Chenbo GENG Yanzhi YI Jiahui HU Renwei ZHANG Zhen ZHANG Gianpietro CONSOLARO Fan YANG Tun LU Ning GU Li SHANG School of Computer Science Fudan University School of Electronic Information and Electrical Engineering Shanghai Jiao Tong University School of Computer Science and Technology Harbin Institute of Technology Huawei Technologies Co. Ltd. Huawei Paris Research Center School of Microelectronics Fudan University

Today's deep learning models face an increasing demand to handle dynamic shape tensors and computation whose shape information remains unknown at compile time and varies in a nearly infinite range at runtime. This shape dynamism brings tremendous challenges for existing compilation pipelines designed for static models which optimize tensor programs relying on exact shape values. This paper presents TSCompiler, an end-to-end compilation framework for dynamic shape models. TSCompiler first proposes a symbolic shape propagation algorithm to recover symbolic shape information at compile time to enable subsequent optimizations. TSCompiler then partitions the shape-annotated computation graph into multiple subgraphs and fine-tunes the backbone operators from the subgraph within a hardware-aligned search space to find a collection of high-performance schedules. TSCompiler can propagate the explored backbone schedule to other fusion groups within the same subgraph to generate a set of parameterized tensor programs for fused cases based on dependence analysis. At runtime, TSCompiler utilizes an occupancy-targeted cost model to select from pre-compiled tensor programs for varied tensor shapes. Extensive evaluations show that TSCompiler can achieve state-of-the-art speedups for dynamic shape models. For example, we can improve kernel efficiency by up to 3.97× on NVIDIA RTX3090, and 10.30× on NVIDIA A100 and achieve up to five orders of magnitude speedups on end-to-end latency.

关键词： machine learning tensor compilers dynamic shape operator fusion code generation auto-tuning

来源：评论

学校读者我要写书评

暂无评论

Full-model-free Adaptive Graph Deep Deterministic Policy Gradient Model for Multi-terminal Soft Open Point Voltage Control in Distribution Systems

引用

Journal of Modern Power Systems and Clean Energy 2024年第6期12卷 1893-1904页

作者： Huayi Wu Zhao Xu Minghao Wang Youwei Jia Department of Electrical and Electronic Engineering The Hong Kong Polytechnic UniversityHong KongChina Shenzhen Research Institute The Hong Kong Polytechnic UniversityShenzhenChina Department of Electrical and Computer Engineering University of MacaoMacaoChina Department of Electrical and Electronic Engineering Southern University of Science and TechnologyShenzhenChina

High penetration of renewable energy sources(RESs)induces sharply-fluctuating feeder power,leading to volt-age deviation in active distribution *** prevent voltage violations,multi-terminal soft open points(M-sOPs)have been integrated into the distribution systems to enhance voltage con-trol ***,the M-SOP voltage control recalculated in real time cannot adapt to the rapid fluctuations of photovol-taic(PV)power,fundamentally limiting the voltage controllabili-ty of *** address this issue,a full-model-free adaptive graph deep deterministic policy gradient(FAG-DDPG)model is proposed for M-SOP voltage ***,the attention-based adaptive graph convolutional network(AGCN)is lever-aged to extract the complex correlation features of nodal infor-mation to improve the policy learning ***,the AGCN-based surrogate model is trained to replace the power flow cal-culation to achieve model-free ***,the deep deterministic policy gradient(DDPG)algorithm allows FAG-DDPG model to learn an optimal control strategy of M-SOP by continuous interactions with the AGCN-based surrogate *** tests have been performed on modified IEEE 33-node,123-node,and a real 76-node distribution systems,which demonstrate the effectiveness and generalization ability of the proposed FAG-DDPGmodel.

关键词： Soft open point graph attention graph convolutional network reinforcement learning voltage control distribution system deep deterministic policy gradient

来源：评论

学校读者我要写书评

暂无评论

A ShuffleNet Architecture-Based Multi-Layer Feature Scaling and Fusion Approach for Underwater Net Cage Damage Detection in Aquaculture

A ShuffleNet Architecture-Based Multi-Layer Feature Scaling ...

引用

2024 IEEE International Conference on Consumer Electronics-Asia, ICCE-Asia 2024

作者： Huang, Xiang-Rui Chen, Yi-Cheng Chen, Liang-Bi Hu, Wu-Chih National Penghu University of Science and Technology Master Program in Electrical Engineering and Computer Science Department of Electrical Engineering Penghu Taiwan National Penghu University of Science and Technology Department of Computer Science and Information Engineering Penghu Taiwan

ISBN: (纸本)9798331530839

In offshore aquaculture operations, personnel equipped with diving gear are often necessary to inspect the underwater net cages for damage, particularly on the sea floor. This manual inspection process is time-consuming and complex. To overcome this problem, this paper proposes a computer vision solution for identifying damage in underwater net cages to address the inefficiencies and challenges of traditional manual inspections. The proposed scheme utilizes a high-performance multi-branch computational architecture designed based on ShuffleNet architecture to detect net cage damage more efficiently. Experimental results demonstrate that this work performs well on the ImageNet ILSVRC-2010 dataset and achieves an accuracy of 88.54% in underwater net damage detection. © 2024 IEEE.

关键词： Aquaculture

来源：评论

学校读者我要写书评

暂无评论

A Game Theoretic Link Scheduler for Wireless Networks with Heterogeneous Energy Sources

引用

China Communications 2024年第12期21卷 108-125页

作者： He Tengjiao Kwanwu Chin Wang Yishun Sieteng Soh College of Information Science and Technology Jinan UniversityGuangzhouChina School of Electrical Computer and Telecommunications EngineeringUniversity of WollongongNSWAustralia School of Electrical Engineering Computing and Mathematical SciencesCurtin UniversityWAAustralia

This paper considers link scheduling in a wireless network comprising of two types of nodes:(i)hybrid access points(HAPs)that harvest solar en-ergy,and(ii)devices that harvest radio frequency(RF)energy whenever HAPs *** aim is to de-rive the shortest possible link schedule that determines the transmission time of inter-HAPs links,and uplinks from devices to *** first outline a mixed in-teger linear program(MILP),which can be run by a central node to determine the optimal schedule and transmit power of HAPs and *** then out-line a game theory based protocol called Distributed Schedule Minimization Protocol(DSMP)that is run by HAPs and ***,it does not require causal energy arrivals and channel gains *** results show that DSMP produces schedule lengths that are at most 1.99x longer than the schedule computed by MILP.

关键词： channel access finite battery mathemati-cal program renewable energy wireless power trans-fer

来源：评论

学校读者我要写书评

暂无评论

Uncertainty quantification in large language models through convex hull analysis

引用

Discover Artificial Intelligence 2024年第1期4卷 1-14页

作者： Catak, Ferhat Ozgur Kuzlu, Murat Department of Electrical Engineering and Computer Science University of Stavanger Rogaland Norway Department of Engineering Technology Old Dominion University NorfolkVA United States

Uncertainty quantification approaches have been more critical in large language models (LLMs), particularly high-risk applications requiring reliable outputs. However, traditional methods for uncertainty quantification, such as probabilistic models and ensemble techniques, face challenges when applied to the complex and high-dimensional nature of LLM-generated outputs. This study proposes a novel geometric approach to uncertainty quantification using convex hull analysis. The proposed method leverages the spatial properties of response embeddings to measure the dispersion and variability of model outputs. The prompts are categorized into three types, i.e., ’easy’, ’moderate’, and ’confusing’, to generate multiple responses using different LLMs at varying temperature settings. The responses are transformed into high-dimensional embeddings via a BERT model and subsequently projected into a two-dimensional space using Principal Component Analysis (PCA), Isomap, Multidimensional Scaling (MDS). The Density-Based Spatial Clustering of Applications with Noise (DBSCAN) algorithm is utilized to cluster the embeddings and compute the convex hull for each selected cluster. The experimental results indicate that the uncertainty of the model for LLMs depends on the prompt complexity, the model, and the temperature setting. © The Author(s) 2024.

关键词： Embeddings

来源：评论

学校读者我要写书评

暂无评论

Room-temperature waveguide-integrated photodetector using bolometric effect for mid-infrared spectroscopy applications

引用

光：科学与应用（英文版） 2025年第5期14卷 1300-1311页

作者： Joonsup Shim Jinha Lim Inki Kim Jaeyong Jeong Bong Ho Kim Seong Kwang Kim Dae-Myeong Geum SangHyeon Kim School of Electrical Engineering Korea Advanced Institute of Science and Technology(KAIST)291 Daehak-RoYuseong-GuDaejeon 34141Republic of Korea Department of Electrical & Computer Engineering Inha University100Inha-roMichuhol-guIncheon 22212Republic of Korea

Waveguide-integrated mid-infrared(MIR)photodetectors are pivotal components for the development of molecular spectroscopy applications,leveraging mature photonic integrated circuit(PIC)*** various strategies,critical challenges still remain in achieving broadband photoresponse,cooling-free operation,and large-scale complementary-metal-oxide-semiconductor(CMOS)-compatible *** leap beyond these limitations,the bolometric effect-a thermal detection mechanism-is introduced into the waveguide *** importantly,we pursue a free-carrier absorption(FCA)process in germanium(Ge)to create an efficient light-absorbing medium,providing a pragmatic solution for full coverage of the MIR spectrum without incorporating exotic materials into ***,we present an uncooled waveguide-integrated photodetector based on a Ge-on-insulator(Ge-OI)PIC architecture,which exploits the bolometric effect combined with ***,our device exhibits a broadband responsivity of 28.35％/mW across 4030-4360 nm(and potentially beyond),challenging the state of the art,while achieving a noise-equivalent power of 4.03×10-7W/Hz0.5 at 4180 *** further demonstrate label-free sensing of gaseous carbon dioxide(CO2)using our integrated photodetector and sensing waveguide on a single *** approach to room-temperature waveguide-integrated MIR photodetection,harnessing bolometry with FCA in Ge,not only facilitates the realization of fully integrated lab-on-a-chip systems with wavelength flexibility but also provides a blueprint for MIR PICs with CMOS-foundry-compatibility.

关键词： applications temperature integrated room bolometric effect mid-infrared photodetector spectroscopy using

来源：评论

学校读者我要写书评

暂无评论

A hybrid feature weighting and selection-based strategy to classify the high-dimensional and imbalanced medical data

引用

Neural Computing and Applications 2024年第20期36卷 12299-12316页

作者： Singh, Harpreet Kaur, Manpreet Singh, Birmohan Department of Computer Science and Engineering Sant Longowal Institute of Engineering and Technology Longowal Sangrur148106 India Department of Electrical and Instrumentation Engineering Sant Longowal Institute of Engineering and Technology Longowal Sangrur148106 India

Machine learning algorithms generally assume that the data are balanced in nature. However, medical datasets suffer from the curse of dimensionality and class imbalance problems. The medical datasets are obtained from the patient information which creates an imbalance in class distribution as the number of normal persons is more than the number of patients and contains a large number of features to represent a sample. It tends to the machine learning algorithms biased toward the majority class which degrades their classification performance for minority class samples and increases the computation overhead. Therefore, oversampling, feature selection and feature weighting-based four strategies are proposed to deal with the problems of class imbalance and high dimensionality. The key idea behind the proposed strategies is to generate a balanced sample space along with the optimal weighted feature space of the most relevant and discriminative features. The Synthetic Minority Oversampling Technique is utilized to generate the synthetic minority class samples and reduce the bias toward the majority class. An Improved Elephant Herding Optimization algorithm is applied to select the optimal features and weights for reducing the computation overhead and improving the interpretation ability of the learning algorithms by providing weights to relevant features. In addition, thirteen methods are developed from the proposed strategies to deal with the problems of high-dimensionality and imbalanced data. The optimized k-Nearest Neighbor (k-NN) learning algorithm is utilized to perform classification. The performance of the proposed methods is evaluated and compared for sixteen high-dimensional imbalanced medical datasets. Further, Freidman’s mean rank test is applied to show the statistical difference between the proposed methods. Experimental and statistical results show that the proposed Feature Weighting followed by the Feature Selection (FW–FS) method performed significantly b

关键词： Feature Selection

来源：评论

学校读者我要写书评

暂无评论

Class-conditional domain adaptation for semantic segmentation

引用

Computational Visual Media 2024年第5期10卷 1013-1030页

作者： Yue Wang Yuke Li James H.Elder Runmin Wu Huchuan Lu School of Information and Communication Engineering Dalian University of TechnologyDalian 116024China School of Computer Science Wuhan UniversityWuhan 430072China Department of Electrical Engineering and Computer Science York UniversityToronto M3J 1P3Canada Department of Computer Science the University of Hong KongHong Kong 999077China

Semantic segmentation is an important sub-task for many ***,pixel-level ground-truth labeling is costly,and there is a tendency to overfit to training data,thereby limiting the generalization *** domain adaptation can potentially address these problems by allowing systems trained on labelled datasets from the source domain(including less expensive synthetic domain)to be adapted to a novel target *** conventional approach involves automatic extraction and alignment of the representations of source and target domains *** limitation of this approach is that it tends to neglect the differences between classes:representations of certain classes can be more easily extracted and aligned between the source and target domains than others,limiting the adaptation over all ***,we address:this problem by introducing a Class-Conditional Domain Adaptation(CCDA)*** incorporates a class-conditional multi-scale discriminator and class-conditional losses for both segmentation and ***,they measure the segmentation,shift the domain in a classconditional manner,and equalize the loss over *** results demonstrate that the performance of our CCDA method matches,and in some cases,surpasses that of state-of-the-art methods.

关键词： domain adaptation generative adversarial networks semantic segmentation cityscapes

来源：评论

学校读者我要写书评

暂无评论

iLIAC: An approach of identifying dissimilar groups on unstructured numerical image dataset using improved agglomerative clustering technique

引用

Multimedia Tools and Applications 2024年第39期83卷 86359-86381页

作者： S, Sreedhar Kumar Ahmed, Syed Thouheed Fathima, Afifa Salsabil Mathivanan, Sandeep Kumar Jayagopal, Prabhu Saif, Abdu Gupta, Sachin Kumar Sinha, Garima Department of Computer Science and Engineering School of Engineering and Technology CMR University Karnataka Bengaluru India Department of Electrical Engineering Indian Institute of Technology Hyderabad Telangana Kandi India School of Computer Science and Engineering REVA University Bangalore India School of Computer Science and Engineering Galgotias University Uttar Pradesh Greater Noida203201 India School of Computer Science Engineering and Information Systems Vellore Institute of Technology Tamil Nadu Vellore632014 India Department of Communication and Computer Engineering Faculty of Engineering and IT Taiz University Taiz Yemen Katra India Department of Computer Science Engineering Jain University Bangalore India

Unstructured Numerical Image Dataset Separation (UNIDS) method employing an enhanced unsupervised clustering technique. The objective is to delineate an optimal number of distinct groups within the input grayscale (G-S) image content, utilizing the improved limited iteration agglomerative clustering (iLIAC) clustering technique for the separation and enhancement of picture elements. The UNIDS method is structured into two primary stages: partitioning and validation. In the partitioning stage, the UNIDS method identifies an appropriate number of discrete clusters within the grayscale image using the iLIAC technique, eliminating the need for predetermined procedures. Subsequently, the method evaluates the similarity and deviation among data elements within the same group in the resultant image dataset. Additionally, it assesses the proximity and inters severance among clusters in the outcome of the image dataset through the partitioning process. Empirical results indicate that the UNIDS system excels in the spontaneous identification of an optimal number of discrete clusters within the input G-S image. The system demonstrates superior thickness, reduced deviation among data elements within the same cluster, increased inter-separation, and diminished inter-closeness between cluster elements. Furthermore, empirical analysis establishes the superior performance of the UNIDS approach compared to existing clustering techniques. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Iterative methods

来源：评论

学校读者我要写书评

暂无评论

Hybrid Water Quality Prediction with Multimodal Low-rank Fusion and Localized Attention

引用

IEEE Internet of Things Journal 2025年第12期12卷 21158-21169页

作者： Bi, Jing Li, Yibo Yuan, Haitao Wang, Mengyuan Wang, Ziqi Zhang, Jia Zhou, MengChu Beijing University of Technology College of Computer Science Beijing100124 China Beihang University School of Automation Science and Electrical Engineering Beijing100191 China Beihang University School of Energy and Power Engineering Beijing100191 China Southern Methodist University Department of Computer Science DallasTX75206 United States New Jersey Institute of Technology Department of Electrical and Computer Engineering NewarkNJ07102 United States

Water quality prediction methods forecast the short-or long-term trends of its changes, providing proactive advice for preventing and controlling water pollution. Existing water quality prediction methods typically fail to capture water quality's nonlinear characteristics accurately and only consider historical time series data. However, meteorology and other factors also significantly impact water quality indicators. Therefore, considering only historical data of water quality time series is not feasible. To solve this problem, this work proposes a hybrid water quality prediction model called CMLIP, which integrates ConvNeXt V2, Multimodal bottleneck transformer, Low-rank multimodal fusion, ITransformer, and PatchTST. CMLIP inputs water quality time series and meteorological remotely sensed rainfall images into a multimodal fusion module before prediction. Specifically, CMLIP integrates the model of ConvNeXt V2 to extract image features. Its multimodal fusion module combines a multimodal bottleneck transformer and the low-rank multimodal fusion to fuse the time series and images. Furthermore, CMLIP combines iTransformer and PatchTST to form an improved prediction module that realizes the prediction of fused features. Experimental results with real-life water quality time series and remotely sensed rainfall images demonstrate that CMLIP when fusing meteorological data, achieves an average improvement of 17% in water quality forecasting accuracy compared to forecasts using only water quality time series. Moreover, CMLIP outperforms other state-of-the-art algorithms in both data fusion and prediction, with an average enhancement of 6% in fusion effectiveness and an average improvement of 22% in prediction accuracy. © 2025 IEEE.

关键词： Rain

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：