检索结果-内蒙古大学图书馆

An Adaptive and Scalable Framework for Resource-Efficient Deployment of Mixture of Experts in LLM-Based Intelligent IoT Networks

引用

IEEE Internet of Things Journal 2025年第13期12卷 22746-22757页

作者： Liu, Chengxu Li, Yangfan Chen, Cen Zou, Xiaofeng Kuang, HaiLan Ma, Xiaolin Lu, Zeyu Zhang, Zhaoyuan Liu, Jing Liu, Xinhua Central South University School of Computer Science and Engineering Changsha410083 China Wuhan University of Technology School of Information Engineering Wuhan430000 China South China University of Technology School of Future Technology Guangzhou510641 China Pazhou Laboratory Guangzhou510335 China School of Information Engineering of Wuhan University of Technology Wuhan430000 China Guilin University of Electronic Technology Computer and Information Security Department Guilin China Hunan Post and Telecommunication College China

The exponential growth of the Internet of Things (IoT) necessitates the deployment of large-scale models capable of processing the complex and diverse data generated by IoT devices. However, the substantial memory requirements of these models pose significant challenges, especially in scenarios where rapid decision-making and low-latency responses are critical. To address these challenges, we propose three innovative strategies for optimizing large model usage in IoT environments. The first strategy is an adaptive loading scheme, which enables dynamic loading of individual model experts. The second strategy involves an expert-by-expert loading approach, further enhancing the ability to load experts as needed, which optimizes memory usage and accelerates computations. The third strategy employs an inter-layer expert reuse mechanism, facilitating the efficient reuse of experts across different layers, thus enhancing response rates without compromising model accuracy. Importantly, these strategies can be directly applied to Mixture of Experts (MoE) large language models without requiring additional training, thereby providing a seamless and efficient solution for leveraging these models in memory-constrained, high-performance IoT environments. © 2014 IEEE.

关键词： Computational modeling Load modeling Adaptation models Memory management Loading Internet of Things Transformers Resource management Graphics processing units Decoding

来源：评论

学校读者我要写书评

暂无评论

HUSS:A Heuristic Method for Understanding the Semantic Structure of Spreadsheets

引用

Data Intelligence 2023年第3期5卷 537-559页

作者： Xindong Wu Hao Chen Chenyang Bu Shengwei Ji Zan Zhang Victor S.Sheng Key Laboratory of Knowledge Engineering with Big Data(the Ministry of Education of China) Hefei University of TechnologyChinaSchool of Computer Science and Information EngineeringHefei University of TechnologyHefeiChina Research Institute of Artificial Intelligence Zhejiang LabHangzhouChina Department of Computer Science Texas Tech UniversityLubbockTX 79409USA

Spreadsheets contain a lot of valuable data and have many practical *** key technology of these practical applications is how to make machines understand the semantic structure of spreadsheets,e.g.,identifying cell function types and discovering relationships between cell *** existing methods for understanding the semantic structure of spreadsheets do not make use of the semantic information of cells.A few studies do,but they ignore the layout structure information of spreadsheets,which affects the performance of cell function classification and the discovery of different relationship types of cell *** this paper,we propose a Heuristic algorithm for Understanding the Semantic Structure of spreadsheets(HUSS).Specifically,for improving the cell function classification,we propose an error correction mechanism(ECM)based on an existing cell function classification model[11]and the layout features of *** improving the table structure analysis,we propose five types of heuristic rules to extract four different types of cell pairs,based on the cell style and spatial location *** experimental results on five real-world datasets demonstrate that HUSS can effectively understand the semantic structure of spreadsheets and outperforms corresponding baselines.

关键词： Spreadsheet semantic structure Information extraction Heuristics Cell function analysis Table structure analysis

来源：评论

学校读者我要写书评

暂无评论

Comparative Analysis of SVM and CNN Classifiers for EEG Signal Classification in Response to Different Auditory Stimuli

Comparative Analysis of SVM and CNN Classifiers for EEG Sign...

引用

2024 International Conference on Telecommunications and Intelligent Systems, ICTIS 2024

作者： Tibermacine, Ahmed Akrour, Djouher Khamar, Rofeida Tibermacine, Imad Eddine Rabehi, Abdelaziz University of Biskra Department of Computer Science Biskra Algeria Sapienza University of Rome Department of Computer Automation and Management Engineering Rome Italy Zian Achour University of Djelfa Laboratory of Telecommunications and Smart Systems Faculty of Sciences and Technology Djelfa17000 Algeria

ISBN: (纸本)9798331527396

This study aims to classify brainwave patterns using electroencephalogram (EEG) signals in response to various auditory stimuli, specifically Quran recitation, participants' favorite music, and Interstellar's main theme. By analyzing the unique EEG responses elicited by these stimuli, the study explores cognitive and emotional processing differences. Two prominent classifiers, Support Vector Machine (SVM) and Convolutional Neural Networks (CNN), were utilized. SVM, combined with Power Spectral Density (PSD) features, achieved the highest accuracy of 95% when distinguishing Quran recitation from rest. CNN, particularly with Kernel Density Estimation (KDE) features, exhibited strong performance in distinguishing between different auditory stimuli, with an average accuracy exceeding 90%. These results highlight the critical role of feature extraction techniques, such as PSD and KDE, in improving classification performance and deepening our understanding of EEG-based emotional and cognitive responses to auditory stimuli. © 2024 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Grid Sweep: Active and Passive Signal Processing

Grid Sweep: Active and Passive Signal Processing

引用

2025 IEEE PES Grid Edge Technologies Conference and Exposition, Grid Edge 2025

作者： Yop, Philip Shehada, Halah Lawrence Livermore National Laboratory LivermoreCA94550 United States Virginia Tech Electrical and Computer Engineering Department BlacksburgVA24060 United States

ISBN: (纸本)9798350352528

The GridSweep device is an actively probing wave-form measurement unit developed by McEachern Laboratories in conjunction with Lawrence Livermore National Lab and Lawrence Berkeley National Lab. The active probing of the device combined with high sensitivity, and GPS time synchronization allow coordinated measurements of distribution system frequency response and other grid characteristics from common electrical outlets. This paper presents an initial review of the signal processing techniques applied to the collected data, a summary of interesting results from several different processing modalities including the active signal response, machine learning techniques, and ambient signal correlation. The early results indicate the measurements from the GridSweep device have potential for use to detect local grid properties like system impedance, and characteristics and identification of distributed energy resources connected on the system. © 2025 IEEE.

关键词： Signal analysis

来源：评论

学校读者我要写书评

暂无评论

mmAlphabet: Air Writing Alphabet Recognition System Based on mmWave FMCW Radar and Convolutional Neural Network 27th

mmAlphabet: Air Writing Alphabet Recognition System Based on...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Huang, Chao-Wang Wang, Chien-Yao Wang, Jia-Ching Department of Computer Science and Information Engineering National Central University Taoyuan Taiwan Multimedia Technologies Laboratory Institute of Information Science Academia Sinica Taipei Taiwan

ISBN: (纸本)9783031781032

An air writing alphabet recognition system based on the images of temporal spectrogram such as average range-time map, average Doppler-time map and average angle-time map derived from an mmWave FMCW radar is proposed. All 26 English lowercase letters written in midair right above the radar sensor can be recognized. Two valuable radar data sets are introduced in diverse environments such as meeting room, office cubicle and living room etc. The use case is the usage scenarios of laptop computers and mobile phones. In one data set, volunteers write freely in their own handwriting styles including different hand speeds and different stroke orders. In the other data set, volunteers are asked to write according to a prescribed sequence of strokes. Then, the gestures sensed by radar are processed into images of temporal spectrogram to represent the written letter. A convolutional neural network which achieves 98.6% test accuracy is exploited as the classifier to recognize the air written alphabet. In the Leave-One-Subject-Out (LOSO) cross validation, it achieves an average test accuracy of 87.74%. The effectiveness of the proposed alphabet recognition system is extensively verified on the two created data sets for different variants of temporal spectrograms. It can be used to implement natural, intelligent noncontact human machine interface. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

Non-planar dielectrics derived thermal and electrostatic field inhomogeneity for boosted weather-adaptive energy harvesting

引用

National Science Review 2023年第9期10卷 248-257页

作者： Yi Zhou Tianpeng Ding Yin Cheng Yi Huang Wu Wang Jianmin Yang Lin Xie Ghim Wei Ho Jiaqing He Shenzhen Key Laboratory of Thermoelectric Materials and Department of Physics Southern University of Science and Technology Department of Electrical and Computer Engineering National University of Singapore School of Electronic Science and Engineering State Key Laboratory of Electronic Thin Film and Integrated Devices University of Electronic Science and Technology of China Department of Materials Science and Engineering National University of Singapore Institute of Materials Research and Engineering Agency for ScienceTechnology and Research (A*STAR) GuangdongHong Kong-Macao Joint Laboratory for Photonic-Thermal-Electrical Energy Materials and Devices Southern University of Science and Technology

Weather-adaptive energy harvesting of omnipresent waste heat and rain droplets,though promising in the field of environmental energy sustainability,is still far from practice due to its low electrical output owing to diele ctric structure irrationality and *** we present atypical upcycling of ambient heat and raindrop energy via an all-in-one non-planar energy harvester,simultaneously increasing solar pyroelectricity and droplet-based triboelectricity by two-fold,in contrast to conventional *** delivered non-planar dielectric with high transmittance confines the solar irradiance onto a focal hotspot,offering transverse thermal field propagation towards boosted inhomogeneous polarization with a generated power density of 6.1 mW m-2at 0.2 ***,the enlarged lateral surface area of curved architecture promotes droplet spreading/separation,thus travelling the electrostatic field towards increased *** enhanced pyroelectric and triboelectric outputs,upgraded with advanced manufacturing,demonstrate applicability in adaptive sustainable energy harvesting on sunny,cloudy,night,and rainy *** findings highlight a facile yet efficient strategy,not only for weather-adaptive environmental energy recovery but also in providing key insights for spatial thermal/electrostatic field manipulation in thermo electrics and ferroelectrics.

关键词： non-planar dielectrics heat manipulation pyroelectrics weather-adaptive energy harvesting

来源：评论

学校读者我要写书评

暂无评论

Disentangling Spatio-Temporal Knowledge for Weakly Supervised Object Detection and Segmentation in Surgical Video

Disentangling Spatio-Temporal Knowledge for Weakly Supervise...

引用

2025 IEEE/CVF Winter Conference on Applications of computer vision, WACV 2025

作者： Liao, Guiqiu Jogan, Matjaz Koushik, Sai Eaton, Eric Hashimoto, Daniel A. University of Pennsylvania Pcaso Laboratory Department of Surgery United States University of Pennsylvania Department of Electrical and Systems Engineering United States University of Pennsylvania Department of Computer and Information Science United States

ISBN: (纸本)9798331510831

Weakly supervised video object segmentation (WSVOS) enables the identification of segmentation maps without requiring extensive annotations of object masks, relying instead on coarse video labels indicating object presence. WSVOS in surgical videos is, however, more challenging due to the complex interaction of multiple transient objects, such as surgical tools moving in and out of the surgical field. In this scenario, state-of-the-art WSVOS methods struggle to learn accurate segmentation maps. We address this problem by introducing ViDeo Spatio-Temporal disentanglement Networks (VDST-Net), a framework to disentangle complex spatio-temporal object interactions using semi-decoupled knowledge distillation to predict high-quality class activation maps (CAMs). A teacher network is designed to help a temporal-reasoning student network resolve activation conflicts, as the student leverages temporal dependencies when specifics about object location and timing in the video are not provided. We demonstrate the efficacy of our framework on a challenging surgical video dataset where objects are, on average, present in less than 60% of annotated frames, and compare our method to state-of-the-art methods on surgical data and on a public dataset commonly used to benchmark WSVOS. Our method outperforms state-of-the-art techniques and generates accurate segmentation masks under video-level weak supervision. Our code is available at: https://***/PCASOlab/VDST-net. © 2025 IEEE.

关键词： Image segmentation

来源：评论

学校读者我要写书评

暂无评论

Enhancing Low-Light Images: A Synthetic Data Perspective on Practical and Generalizable Solutions 39

Enhancing Low-Light Images: A Synthetic Data Perspective on ...

引用

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Long, Yu Lin, Qinghua Wang, Zhihua Zhang, Kai Zhang, Jianguo Fang, Yuming School of Computer Science Beijing Institute of Technology China School of Computer Science Guangdong University of Technology China Department of Engineering Shenzhen MSU-BIT University China School of Intelligence Science and Technology Nanjing University Suzhou China Department of Computer Science and Engineering Southern University of Science and Technology China Pengcheng Laboratory China School of Information Management Jiangxi University of Finance and Economics China

ISBN: (纸本)157735897X

Recently, deep neural networks (DNNs) have emerged as the leading approach for low-light image enhancement (LLIE). However, training these models requires large-scale paired datasets, which are challenging to obtain due to the labor-intensive nature of real-world data collection. To alleviate this issue, synthetic data are often combined with real-captured data for training. However, most existing low-light image synthesis methods are simply performed in the sRGB domain using Gamma correction or manual adjustments, which fail to incorporate the physical imaging prior through the image signal processing (ISP) pipeline and thus result in limited dataset size and degradation space. Consequently, LLIE methods trained on such data often exhibit some drawbacks in the results, such as inaccurate white balance and abnormal enhancement artifacts, which limit their practicality and generalizability. In this paper, we propose a practical low-light image synthesis pipeline capable of generating unlimited paired training data. Our pipeline starts with a reverse ISP model that converts sRGB images back to the unprocessed RAW domain, where we then simulate low-light degradation, noise degradation, and white balance adjustments. Finally, the degraded RAW images are processed through a forward ISP model to produce low-light sRGB images. The pipeline further employs multiple tone mapping curves and color correction matrices (CCMs) to expand the degradation space. Hence, trained with our proposed synthetic data, existing state-of-the-art (SOTA) LLIE deep models are expected to improve their performance. Extensive experiments across various datasets demonstrate that our synthetic data can indeed effectively enhance existing LLIE deep models, improving both their practicality and generalizability. Copyright © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词： Image enhancement

来源：评论

学校读者我要写书评

暂无评论

Exploring BERT and ELMo for Bangla Spam SMS Dataset Creation and Detection 26

Exploring BERT and ELMo for Bangla Spam SMS Dataset Creation...

引用

26th International Conference on computer and Information Technology, ICCIT 2023

作者： Khan, Fariba Mustafa, Rashed Tasnim, Farzana Mahmud, Tanjim Hossain, Mohammad Shahadat Andersson, Karl University of Chittagong Department of Computer Science and Engineering Chittagong4331 Bangladesh International Islamic University Chittagong Department of Computer Science and Engineering Chittagong4318 Bangladesh Rangamati Science and Technology University Department of Computer Science and Engineering Rangamati4500 Bangladesh Luleå University of Technology Pervasive and Mobile Computing Laboratory Luleå971 87 Sweden

ISBN: (纸本)9798350359015

The proliferation of digital communication and social media platforms has led to an upsurge in spam messages across various languages, including Bangla. These intrusive messages not only cause user annoyance but also elevate the risk of important messages going unnoticed. To address this concern, we harnessed the power of natural language processing (NLP) techniques in conjunction with machine-learning models. This study delves into the efficacy of diverse machine-learning and deep-learning models in detecting spam messages in the Bangla language. Particularly, deep-learning methodologies such as ELMo and BERT have exhibited promising results in spam detection across other languages. However, the realm of Bangla spam detection remains underexplored, with a dearth of deep-learning approaches. To bridge this gap, we constructed a unique Bangla SMS dataset, given the scarcity of available Bangla Spam SMS datasets, and proceeded to evaluate the performance of ELMo, BERT, and conventional machine learning models. The findings from our experiments underscore the superior performance of deep learning-based models over traditional methods, with Elmo surpassing BERT in Bangla spam detection, achieving an impressive accuracy of approximately 94%. This endeavor bears the potential to not only enhance SMS utilization efficiency but also bolster cybersecurity measures against phishing attacks. © 2023 IEEE.

关键词： BERT ELMo Machine Learning Spam SMS Word Embedding

来源：评论

学校读者我要写书评

暂无评论

VERO: Verification and Zero-Shot Feedback Acquisition for Few-Shot Multimodal Aspect-Level Sentiment Classification 39

VERO: Verification and Zero-Shot Feedback Acquisition for Fe...

引用

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Sun, Kai Wu, Hao Shi, Bin Mensah, Samuel Liu, Peng Dong, Bo Shaanxi Provincial Key Laboratory of Big Data Knowledge Engineering Xi’an Jiaotong University China School of Computer Science and Technology Xi’an Jiaotong University China School of computer science and engineering Guangxi Normal University Guilin China Department of Computer Science University of Sheffield Sheffield United Kingdom School of Continuing Education Xi’an Jiaotong University China

ISBN: (纸本)157735897X

Deep learning approaches for multimodal aspect-level sentiment classification (MALSC) often require extensive data, which is costly and time-consuming to obtain. To mitigate this, current methods typically fine-tune small-scale pretrained models like BERT and BART with few-shot examples. While these models have shown success, Large vision-Language Models (LVLMs) offer significant advantages due to their greater capacity and ability to understand nuanced language in both zero-shot and few-shot settings. However, there is limited work on fine-tuning LVLMs for MALSC. A major challenge lies in selecting few-shot examples that effectively capture the underlying patterns in data for these LVLMs. To bridge this research gap, we propose an acquisition function designed to select challenging samples for the few-shot learning of LVLMs for MALSC. We compare our approach, Verification and ZERO-shot feedback acquisition (VERO), with diverse acquisition functions for few-shot learning in MALSC. Our experiments show that VERO outperforms prior methods, achieving an F1 score improvement of up to 6.07% on MALSC benchmark datasets. © 2025, Association for the Advancement of Artificial.

关键词： Zero-shot 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：