检索结果-内蒙古大学图书馆

Proceedings of the 41st International Conference on Machine Learning

作者： Pedram Akbarian Tongzheng Ren Jiacheng Zhuo Sujay Sanghavi Nhat Ho Department of Electrical and Computer Engineering University of Texas at Austin Texas Computer Science University of Texas at Austin Texas Department of Statistics and Data Sciences University of Texas at Austin Texas

It is known that when the statistical models are singular, i.e., the Fisher information matrix at the true parameter is degenerate, the fixed step-size gradient descent algorithm takes polynomial number of steps in terms of the sample size n to converge to a final statistical radius around the true parameter, which can be unsatisfactory for the practical application. To further improve that computational complexity, we consider utilizing the local curvature information for parameter estimation. Even though there is a rich literature in using the local curvature information for optimization, the statistical rate of these methods in statistical models, to the best of our knowledge, has not been studied rigorously. The major challenge of this problem is due to the non-convex nature of sample loss function. To shed light on these problems, we specifically study the normalized gradient descent (NormGD) algorithm, a variant of gradient descent algorithm whose step size is scaled by the maximum eigenvalue of the Hessian matrix of the empirical loss function, and deal with the aforementioned issue with a population-to-sample analysis. When the population loss function is homogeneous, the NormGD iterates reach a final statistical radius around the true parameter after a logarithmic number of iterations in terms of n. Therefore, for fixed dimension d, the NormGD algorithm achieves the optimal computational complexity O(n) to reach the final statistical radius, which is cheaper than the complexity O(nτ) of the fixed step-size gradient descent algorithm for some τ > 1.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Characterizing the Performance of Deep Learning Inference for Edge Video Analytics 9th

Characterizing the Performance of Deep Learning Inference fo...

引用

9th CCF Conference on Big data, Bigdata 2021

作者： Luo, Zhenxiao Wang, Xinglong Zhang, Yuxiao Wang, Zelong Hu, Miao Wu, Di School of Computer Science and Engineering Sun Yat-Sen University Guangzhou510006 China Guangdong Key Laboratory of Big Data Analysis and Processing Guangzhou510006 China

ISBN: (纸本)9789811697081

Real-time video analytics is a killer application for edge computing, however, it has not been fully understood how much edge resource is required to support compute-intensive video analytics tasks at the edge. In this paper, we conduct an in-depth measurement study to unveil the resource requirement of deep learning (DL) inference for edge video analytics (EVA). We measure the DL inference time under various resource combinations, with different physical resources (e.g., processor types, core numbers, memory), video resolutions, and DL inference models. It is observed that the relationship between DL inference time and resource configuration is complicated, which cannot be captured by a simple model. Considering the coupling effects, we model the resource requirements of the DL inference by utilizing tensor factorization (TF) method. Our TF-based DL inference model can match well with the observations from field measurements. Different from previous models, our model is completely explainable and all the components have their physical meanings. Especially, it is possible to match the top three components extracted from the TF-based model with the practical DL functional layers, namely, convolutional layer, pooling layer, and fully-connected layer. The measurement results show that the top three components contribute approximately 50%, 25%, and 10% time for DL inference execution, which provides clear instruction on fine-grained DL task scheduling for edge video analytics applications. © 2022, Springer Nature Singapore Pte Ltd.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

Class Incremental Learning for Knowledge Acquisition Based on Prototype Network Gating Mechanism

Class Incremental Learning for Knowledge Acquisition Based o...

引用

Medical Artificial Intelligence (MedAI), IEEE International Conference on

作者： Yuming Feng Xiaolong Jin Haozhe Liang Jiafeng Guo School of Computer Science and Technology University of Chinese Academy of Sciences CAS Key Laboratory of Network Data Science and Technology Institute of Computing Technology Chinese Academy of Sciences Institute of Systems Engineering Academy of Military Science

ISBN: (数字)9798350377613

ISBN: (纸本)9798350377620

With the development of information technology, the medical field has accumulated a large amount of medical data, which continues to grow. Efficient extraction and effective utilization of knowledge from this vast amount of medical data are crucial for realizing its immense value. Traditional knowledge acquisition algorithms face challenges in effectively identifying and acquiring new types of knowledge in practical applications. To address this, this paper proposes a class-incremental learning method based on a prototype network gating mechanism, called PGMoE. This method integrates multiple experts through a prototype network-based gating mechanism, dynamically adjusting the model's response to new and old knowledge, thereby effectively mitigating the issue of catastrophic forgetting in class-incremental learning. PGMoE not only improves the stability and accuracy of the model in continuous learning processes but also ensures the model's ability to remember historical knowledge. Comparative experiments on multiple datasets with existing class-incremental learning methods demonstrate that PGMoE has significant performance advantages in handling continuous learning tasks.

关键词： Learning systems Accuracy Incremental learning Knowledge acquisition Scalability Heuristic algorithms Prototypes Logic gates Stability analysis Information technology

来源：评论

学校读者我要写书评

暂无评论

An Energy-Aware UAVs Path Coverage for Critical Infrastructure Inspections 12

An Energy-Aware UAVs Path Coverage for Critical Infrastructu...

引用

12th International Conference on Advanced Cloud and Big data, CBD 2024

作者： Wang, Yifan Zeng, Wei Li, Guanyu Xiong, Chenglong Wang, Zicheng Mao, Yingchi College of Computer Science and Software Engineering Hohai University Nanjing China Hohai University Key Laboratory of Water Big Data Technology of Ministry of Water Resources Nanjing China Huaneng Lancang River Hydropower Co. CHINA HUANENG Group Kunming China Power China Kunming Engineering Co. Yunnan China

ISBN: (纸本)9798331511074

Critical infrastructures increasingly rely on unmanned aerial vehicles (UAVs) for inspection tasks. The significance of different components within these infrastructures is subject to variability and can be influenced by external factors. The principal aim of routine UAV inspections is to ensure differentiated coverage of pivotal sections with minimal energy consumption. However, extant research on UAV path coverage fails to fully account for the variability and dynamic shifts in regional significance and their impact on coverage efficacy. This paper presents an energy-aware collaborative coverage policy for UAVs, designated as ***-MATD3 employs a dynamic weight region partitioning method tailored to real-world environments and addresses the action selection challenge for UAVs using a discrete Partially Observable Markov Decision Process (Dec-POMDP). By amalgamating MATD3 with stacked LSTM, this approach mitigates redundant path overlaps and unnecessary action replication across multiple agents, thus optimizing coverage and diminishing energy usage. Simulation outcomes demonstrate that EA-MATD3 reduces energy consumption by an average of 9.65% relative to the Greedy, MADDPG, and MATD3 algorithms while sustaining a superior coverage rate. © 2024 IEEE.

关键词： Markov processes

来源：评论

学校读者我要写书评

暂无评论

Empirical Evaluation of Internet of Things Assisted Fire Fighting Robotic Design Based on Intelligent Sensors Association

Empirical Evaluation of Internet of Things Assisted Fire Fig...

引用

Electronics and Sustainable Communication Systems (ICESC), 2020 International Conference on

作者： T Mahitha Sajja Suneel M. Amina Begum Er. S. John Pimo B. Swapna Revathi. R Department of Computer Science Saveetha College of Liberal Arts and Sciences SIMATS Department of CSE(Data Science) Institute of Aeronautical Engineering Hyderabad Telangana India Department of ECE Oxford College of Engineering Venmani Tamilnadu India Computer Science and Engineering St. Xavier's Catholic College of Engineering Nagercoil ECE Dr MGR Educational and Research Institute Chennai Department of Computer Science Karpagam Academy Of Higher Education Coimbatore

ISBN: (数字)9798350379945

ISBN: (纸本)9798350379952

This research work presents the design and implementation of an Internet of Things (IoT) assisted fire fighting robotic system based on intelligent sensors association. The system integrates various components including intelligent sensors for detecting fire, smoke, temperature changes, and gas emissions, Arduino microcontrollers for data processing and control logic execution, Wi-Fi communication modules for remote monitoring and data exchange, and actuators for physical motion and manipulation tasks. Through the integration and coordination of these components, the firefighting robot achieves autonomous capabilities in detecting, localizing, and responding to fire incidents effectively. The intelligent sensors, including the MQ-2 Gas Sensor, TMP36 Temperature Sensor, MQ-7 Carbon Monoxide Sensor, and MLX90614 Non-Contact Infrared Temperature Sensor, provide real-time environmental data inputs to the Arduino Uno microcontroller, which interprets sensor readings, executes decision-making algorithms, and orchestrates responses. Leveraging bidirectional Wi-Fi connectivity facilitated by the ESP8266 Wi-Fi Module, the system enables remote monitoring and data exchange, empowering operators with situational awareness for informed decision-making. Actuators such as DC Gear Motors and SG90 Micro Servo Motors translate digital commands into mechanical motion and manipulation, enabling the robot to navigate through fire-affected environments, manipulate fire fighting tools, and execute precision fire fighting maneuvers. This abstract presents a comprehensive overview of the IoT-assisted fire fighting robotic system, highlighting its autonomous capabilities, integration of components, and potential for enhancing fire fighting efficiency and safety.

关键词： Temperature sensors Actuators Microcontrollers Robot kinematics Decision making Robot sensing systems Internet of Things Intelligent sensors Wireless fidelity Remote monitoring

来源：评论

学校读者我要写书评

暂无评论

Multi-scale Variable Window Attention Network for Continuous Sign Language Recognition

Multi-scale Variable Window Attention Network for Continuous...

引用

2024 International Conference on Cyber-Physical Social Intelligence, ICCSI 2024

作者： Zhang, Nan Wei, Chao Guo, Leming Cheng, Zeyu Zhao, Zhao Wu, Yangcan Xue, Wanli Tianjin University of Technology School of Computer Science and Engineering Tianjin China Tianjin Emergency Centre Tianjin China Tianjin University of Technology School of Mechanical Engineering Tianjin China Shaanxi Big Data Group Co. Ltd Shaanxi China Industry Research Institute Co Shaanxi Logistics Group Shaanxi China

ISBN: (纸本)9798350376739

Sign language is primarily conveyed through hands and face movements. Therefore, a pivotal challenge in continuous sign language recognition (CSLR) is the meticulous modeling and interpreting of visual cues emanating from hands and facial movements. However, contemporary CSLR methods have exhibited limitations in adequately capturing information from both the hands and facial regions, often overlooking the continuity, and correlations across multiple consecutive frames. To mitigate these shortcomings, we propose the Multi-scale Variable Window Attention Network (MVAN). This framework fuses the strengths of an improved Efficient Multi-scale Attention (iEMA) with Variable Window Adaptation (VWA) attention. Specifically, iEMA first scans information spatially at multiple scales to simultaneously capture information from areas such as hands and faces, and models spatial continuity interactions between them, enhancing spatial features. Secondly, VWA dynamically changes its window size at the temporal level to adaptively capture complex temporal changes in hands and face movements in successive frames. Thus, VWA can effectively model the continuity and correlation between hands and face movements to have the ability to capture sign language movements of different lengths. Comprehensive experimental results demonstrate that our proposed MVAN strengthens the visual feature extraction of hands and face, improving the recognition performance. © 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

Edit and Alphabet-Ordering Sensitivity of Lex-parse

arXiv

引用

arXiv 2024年

作者： Nakashima, Yuto Köppl, Dominik Funakoshi, Mitsuru Inenaga, Shunsuke Bannai, Hideo Department of Informatics Kyushu University Japan Department of Computer Science and Engineering University of Yamanashi Japan NTT Communication Science Laboratories M&D Data Science Center Tokyo Medical and Dental University Japan

We investigate the compression sensitivity [Akagi et al., 2023] of lex-parse [Navarro et al., 2021] for two operations: (1) single character edit and (2) modification of the alphabet ordering, and give tight upper and lower bounds for both operations. For both lower bounds, we use the family of Fibonacci words. For the bounds on edit operations, our analysis makes heavy use of properties of the Lyndon factorization of Fibonacci words to characterize the structure of lex-parse. © 2024, CC BY.

关键词：

来源：评论

学校读者我要写书评

暂无评论

EdgeAuth: An intelligent token-based collaborative authentication scheme

EdgeAuth: An intelligent token-based collaborative authentic...

引用

作者： Jiang, Xutong Dou, Ruihan He, Qiang Zhang, Xuyun Dou, Wanchun State Key Laboratory for Novel Software Technology Nanjing University Nanjing China Faculty of Mathematics University of Waterloo Waterloo Canada Department of Computer Science and Software Engineering Swinburne University of Technology Melbourne Australia Department of Computing Macquarie University Sydney Australia College of Big Data and Intelligent Engineering Southwest Forestry University China

Edge computing is regarded as an extension of cloud computing that brings computing and storage resources to the network edge. For some Industrial Internet of Things (IIoT) applications such as supply-chain supervision and collaboration, Internet of Vehicles, real-time video analysis and so forth, users should be authenticated before visiting the geographically distributed edge servers. Limited by the considerable latency between the cloud and edge servers, and the limited capacity of edge servers, it is infeasible to copy the authentication method from cloud servers when users are authenticated in edge servers. In view of this challenge, this paper proposes a novel token-based authentication scheme, named EdgeAuth, that enables fast edge user authentication through collaboration among cloud servers and edge servers. Under the EdgeAuth scheme, edge servers can rapidly verify the credentials of users who have been authenticated by the cloud server. EdgeAuth can also protect users from a series of authentication attacks, for example, the replay attack, DoS attack and man-in-the-middle attack. The results of experiments conducted on a simulated edge computing environment validate the usefulness of EdgeAuth through a comparison in latency and throughput against two baseline schemes. © 2023 John Wiley & Sons Ltd.

关键词： Authentication

来源：评论

学校读者我要写书评

暂无评论

A Survey on Large Language Model Acceleration based on KV Cache Management

arXiv

引用

arXiv 2024年

作者： Li, Haoyang Li, Yiming Tian, Anxin Tang, Tianhao Xu, Zhanchao Chen, Xuejia Hu, Nicole Dong, Wei Li, Qing Chen, Lei Department of Computing The Hong Kong Polytechnic University China Department of Computer Science and Engineering Hong Kong University of Science and Technology China Department of Computer Science and Technology Huazhong University of Science and Technology China The Chinese University of Hong Kong Hong Kong Department of Computing and Data Science Nanyang Technological University Singapore

Large Language Models (LLMs) have revolutionized a wide range of domains such as natural language processing, computer vision, and multi-modal tasks due to their ability to comprehend context and perform logical reasoning. However, the computational and memory demands of LLMs, particularly during inference, pose significant challenges when scaling them to real-world, long-context, and real-time applications. Key-Value (KV) cache management has emerged as a critical optimization technique for accelerating LLM inference by reducing redundant computations and improving memory utilization. This survey provides a comprehensive overview of KV cache management strategies for LLM acceleration, categorizing them into token-level, model-level, and system-level optimizations. Token-level strategies include KV cache selection, budget allocation, merging, quantization, and low-rank decomposition, while model-level optimizations focus on architectural innovations and attention mechanisms to enhance KV reuse. System-level approaches address memory management, scheduling, and hardware-aware designs to improve efficiency across diverse computing environments. Additionally, the survey provides an overview of both text and multimodal datasets and benchmarks used to evaluate these strategies. By presenting detailed taxonomies and comparative analyses, this work aims to offer useful insights for researchers and practitioners to support the development of efficient and scalable KV cache management techniques, contributing to the practical deployment of LLMs in real-world applications. The curated paper list for KV cache management is in: https://***/TreeAI-Lab/Awesome-KV-Cache-Management. Copyright © 2024, The Authors. All rights reserved.

关键词： Budget control

来源：评论

学校读者我要写书评

暂无评论

Genre Classification: A Machine Learning Based Comparative Study of Classical Bengali Literature

Genre Classification: A Machine Learning Based Comparative S...

引用

International Conference on computer and Information Technology (ICCIT)

作者： Asadullah Al Galib Maisha Mostofa Prima Satabdi Rani Debi Md Muntasir Mahadi Nayema Ahmed Ehsanur Rahman Rhythm Adib Muhammad Amit Annajiat Alim Rasel Department of Computer Science and Engineering (CSE) School of Data and Sciences (SDS) Brac University Dhaka Bangladesh

Bengali literature, specifically classical Bengali literature has been a source of inspiration, a spark for paradigm-shifting revolutions, and the sole sustaining source of cultural thirst for hundreds of millions of people over many generations. Unfortunately, very few attempts have been made to analyze this never-ending collection of literary works from the luminary figures of Bengali literature. The availability of high-quality research-ready datasets comprising all the authenticated literary works has been a key obstacle in conducting NLP research, utilizing the most recent advancements in deep learning and large language models. Identifying the genre of a given text snippet is a key step in analyzing a vast collection of works comprising different styles, themes, and motivations from classical authors. From classifying previously unexplored archival documents to identifying and suggesting similar literary works for modern recommender engines, genre classification opens the door for many downstream and specialized use cases. In this project, we initiate an ambitious goal of compiling a comprehensive dataset of literary works from classical authors and eventually extending the collection to contemporary writers as well. We explore both classical methods such as Naive Bayes as well as LSTM and recent transformer-based models to classify genre from short text snippets. We concluded that fine-tuning pre-trained BERT models produced much higher accuracy than both classical and LSTM models.

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：