检索结果-内蒙古大学图书馆

The Journal of Machine Learning Research 2024年第1期25卷 1132-1187页

作者： Ali Ramezani-Kebrya Kimon Antonakopoulos Volkan Cevher Ashish Khisti Ben Liang Department of Informatics University of Oslo and Visual Intelligence Centre Integreat Norwegian Centre for Knowledge-driven Machine Learning Oslo Norway Laboratory for Information and Inference Systems (LIONS) EPFL EPFL STI IEL LIONS Lausanne Switzerland Department of Electrical and Computer Engineering University of Toronto Toronto ON Canada

While momentum-based accelerated variants of stochastic gradient descent (SGD) are widely used when training machine learning models, there is little theoretical understanding on the generalization error of such methods. In this work, we first show that there exists a convex loss function for which the stability gap for multiple epochs of SGD with standard heavy-ball momentum (SGDM) becomes unbounded. Then, for smooth Lipschitz loss functions, we analyze a profiles momentum-based update rule, i.e., SGD with early momentum (SGDEM) under a broad range of step-sizes, and show that it can train machine learning models for multiple epochs with a guarantee for generalization. Finally, for the special case of strongly convex loss functions, we find a range of momentum such that multiple epochs of standard SGDM, as a special form of SGDEM, also generalizes. Extending our results on generalization, we also develop an upper bound on the expected true risk, in terms of the number of training steps, sample size, and momentum. Our experimental evaluations verify the consistency between the numerical results and our theoretical bounds. SGDEM improves the generalization error of SGDM when training ResNet-18 on ImageNet in practical distributed settings.

关键词： uniform stability generalization error heavy-ball momentum stochastic gradient descent non-convex

来源：评论

学校读者我要写书评

暂无评论

Multi-behavior Learning for Socially Compatible Autonomous Driving

Multi-behavior Learning for Socially Compatible Autonomous D...

引用

International Conference on Intelligent Transportation

作者： Sanjula Jayawardana Vindula Jayawardana Kaneeka Vidanage Cathy Wu University of Westminster UK Informatics Institute of Technology Colombo Sri Lanka Department of Electrical Engineering and Computer Science MIT Laboratory for Information & Decision Systems Cambridge MA USA Department of Civil and Environmental Engineering MIT Laboratory for Information Decision Systems Institute of Data Systems and Society Cambridge MA USA

Autonomous vehicles (AVs) have the potential to revolutionize transportation, but their effective integration into the real world requires addressing the challenge of interacting with human drivers. Real-world driving involves negotiating and cooperating with fellow drivers through social cues, necessitating AVs to also demonstrate such social compatibilities. However, despite the popularity, current learning-based control methods for AV policy synthesis often overlook this crucial aspect. In this work, we look at the problem of enabling socially compatible driving when AV control policies are learned. We leverage human driving data to learn a social preference model of human driving and then integrate it with reinforcement learning-based AV policy synthesis using Social Value Orientation theory. In particular, we propose to use multi-task reinforcement learning to learn diverse social compatibility levels in driving (ex: altruistic, prosocial, individualistic, and competitive), focusing on the requirement of having diverse behaviors in real-world driving. Using highway driving scenarios, we demonstrate through experiments that socially compatible AV driving not only enables naturalistic driving behaviors but also reduces collision rates from the baseline. Our findings reveal that without social compatibility, AV policies tend to adopt dangerously competitive driving behaviors, while the incorporation of social compatibility fosters smoother vehicle maneuvers.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Geospatial Analysis of Extreme Temperature Impacts in Agricultural systems Using Machine Learning

Geospatial Analysis of Extreme Temperature Impacts in Agricu...

引用

IEEE International Symposium on systems engineering (ISSE)

作者： Rayshaun L. Wheeler John S. Miller Gia N. Smith James H. Lambert Cody A. Pennetti Garrick E. Louis Deandre A. Johnson Department of Civil & Environmental Engineering University of Virginia Charlottesville VA USA School of Data Science University of Virginia Charlottesville VA USA Department of Systems & Information Engineering University of Virginia Charlottesville VA USA Department of Engineering & Society University of Virginia Charlottesville VA USA

ISBN: (数字)9798350353723

ISBN: (纸本)9798350353730

Recent higher temperatures, observed compared to previous years, have led to more frequent and intense heat waves, adversely affecting agriculture and human health. This paper explores using Inverse Distance Weighting (IDW) and Kriging to develop GIS layers for a heat-based risk index. Recent increases in global mean temperatures have led to more frequent and intense heat waves, impacting food supply, vegetation, and human health. Optimal plant growth temperatures in temperate climates range from 20°C to 30°C, with nighttime temperatures also being critical. Exceeding these temperature ranges can reduce plant immunity, cause disease, and result in poor crop yields. For example, mango cultivars need a "safe" temperature range of 10°C to 12°C to avoid damage. The intensification of higher temperatures established July 2023 record the hottest days on Earth, highlighting the need for effective monitoring and management tools. IDW, a deterministic method, offers a straightforward and efficient approach by assuming closer points have more influence. Kriging, a probabilistic geostatistical technique, accounts for spatial autocorrelation and prediction uncertainty, making it suitable for complex datasets. The study compares these methods' performance in creating continuous surfaces representing heat-based risk. IDW is simple and efficient but less accurate in sparsely populated areas, while Kriging is more accurate and detailed but requires more computational resources and expertise. An automated workflow for data preprocessing, model training, and cross-validation ensures consistent results. The study recommends adopting IDW for initial assessments and transitioning to Kriging as expertise grows. Enhanced data collection on temperature and population density is also advised to improve GIS layer accuracy, aiding better decision-making and resource allocation to mitigate extreme heat effects on agriculture and public health.

关键词： Heating systems Support vector machines Training Temperature distribution Accuracy Computational modeling Decision making Linear regression Resource management Random forests Climate change

来源：评论

学校读者我要写书评

暂无评论

Representing Additive Gaussian Processes by Sparse Matrices

arXiv

引用

arXiv 2023年

作者： Zou, Lu Chen, Haoyuan Ding, Liang Information Hub The Hong Kong University of Science and Technology Guangzhou China Department of Industrial and Systems Engineering School of Data Science Fudan University China

Among generalized additive models, additive Matérn Gaussian Processes (GPs) are one of the most popular for scalable high-dimensional problems. Thanks to their additive structure and stochastic differential equation representation, back-fitting-based algorithms can reduce the time complexity of computing the posterior mean from O(n3) to O(nlogn) time where n is the data size. However, generalizing these algorithms to efficiently compute the posterior variance and maximum log-likelihood remains an open problem. In this study, we demonstrate that for Additive Matérn GPs, not only the posterior mean, but also the posterior variance, log-likelihood, and gradient of these three functions can be represented by formulas involving only sparse matrices and sparse vectors. We show how to use these sparse formulas to generalize back-fitting-based algorithms to efficiently compute the posterior mean, posterior variance, log-likelihood, and gradient of these three functions for additive GPs, all in O(nlogn) time. We apply our algorithms to Bayesian optimization and propose efficient algorithms for posterior updates, hyperparameters learning, and computations of the acquisition function and its gradient in Bayesian optimization. Given the posterior, our algorithms significantly reduce the time complexity of computing the acquisition function and its gradient from O(n2) to O(logn) for general learning rate, and even to O(1) for small learning rate. © 2023, CC BY.

关键词： Gaussian distribution

来源：评论

学校读者我要写书评

暂无评论

Intra-datacenter Load Balancing in a Federated Cloud with Throttled Algorithm

Intra-Datacenter Load Balancing in a Federated Cloud with Th...

引用

2024 International Conference on Science, engineering and Business for Driving Sustainable Development Goals, SEB4SDG 2024

作者： Akinola, Damola Gideon Adetiba, Emmanuel Abayomi, Abdultaofeek Nnaji, Uche Thakur, Surendra Moyo, Sibusiso Covenant University Covenant Applied Informatics and Communications African Center of Excellence Department of Electrical and Information Engineering Ota Nigeria Durban University of Technology HRA Institute for Systems Science Durban South Africa Mangosuthu University of Technology Department of Information and Communication Technology Durban South Africa Covenant University Dept. of Computer and Information Science Covenant Applied Informatics and Communications African Center of Excellence Ota Nigeria Durban University of Technology KZN E-Skills CoLab Durban South Africa Stellenbosch University School for Data Science and Computational Thinking Department of Mathematical Sciences Stellenbosch7602 South Africa

ISBN: (纸本)9798350358155

The emergence of cloud computing technology has motivated a very high number of users in different organizations to access its services in running and delivering their various operations and services. However, this surge of cloud users has led to the problem of uneven load sharing among the cloud computing infrastructures. As a result, federated cloud infrastructure was initiated to provide more cloud computing resources to accommodate more cloud users' requests and also to provide equal requests mapping with the available cloud resources. Nevertheless, the issue of unequal allocation of requests within the datacenter(s) that makes up the federated cloud environment still persists. This research work presents an intra-datacenter load balancing algorithm in a federated cloud environment named;the intra-balancer adapted throttled algorithm to evenly distribute requests in each datacenter that formulate the federated cloud infrastructures. The simulation of the intra-balancer was carried out in CloudAnalyst, the results of the implementation showed that the intra-balancer outperformed the Round-Robin and ESCE with 98.93ms and 2.26ms for the response and processing time respectively. The results assert that the intra-balancer algorithm provides a better load balancing solution in a federated cloud environment. © 2024 IEEE.

关键词： Cloud platforms

来源：评论

学校读者我要写书评

暂无评论

Design and Implementation of an active safety system for Vehicular Ad-Hoc Networks(VANETs)

Design and Implementation of an active safety system for Veh...

引用

International Conference on information Networking

作者： Halimjon Hujamatov Debasis Das Amir Lazarev Ernazar Reypnazarov Doston Khasaniv Ankur Nahar Data Communication Networks and Systems Department Tashkent University of Information Technologies Named After Muhammad al-Khwarizmi Tashkent Uzbekistan Department of Computer Science and Engineering Indian Institite of Technology Jodhpur Rajasthan India

ISBN: (数字)9798350330946

ISBN: (纸本)9798350330953

Vehicular communication, underpinned by IEEE 802.11p/WAVE-based Vehicle Ad-hoc Networks (VANETs), is instrumental in the seamless functioning of intra-vehicle exchanges. However, a comprehensive assessment of these systems reveals suboptimal efficiencies at the data layer, specifically regarding default broadcast intervals. Such inefficiencies lead to escalated packet collisions and subpar utilization of the delay time counter—factors that undermine the synergistic interplay between Active Safety systems (ASS), such as Adaptive Cruise Control (ACC), and their passive safety counterparts. To address these intricacies, this research proposes an innovative mathematical framework tailored for the IEEE 802.11p MAC layer. We propose a model that elucidates the intricate dynamics of the delay time counter and offers refined broadcast intervals buttressed by robust algorithmic strategies. Empirical evaluations, conducted in meticulously simulated vehicular environments, validate the prowess of the proposed paradigm, highlighting a decline in packet collision instances. Quantitative findings from this research evince a notable decrease in packet collision rates and a commensurate enhancement in communication reliability, pivotal for advanced vehicular systems. Such technical augmentations directly elevate the operational reliability of cutting-edge safety mechanisms, exemplified by systems like the Toyota Pre-Crash Safety System.

关键词： Heuristic algorithms Instruments Vehicular ad hoc networks Physical layer Road safety Mathematical models Safety

来源：评论

学校读者我要写书评

暂无评论

Real-Time IoT Cybersecurity using Machine Learning-based AI Threat Detection System to Train Generative Robots

Real-Time IoT Cybersecurity using Machine Learning-based AI ...

引用

Trends in Material Science and Inventive Materials (ICTMIM), International Conference on

作者： P Srilakshmi Koushik Reddy Chaganti Talachendri Suryam Sweety Julia I D. Chaithanya Jami Kavitha Department of CSE G Pulla Reddy Engineering College Dept. of Computer and Information Systems University of the Cumberlands Department of CSE (DATA SCIENCE) Vignan Institute of Technology and Science Department of AI&ML Malla Reddy University Department of CSE CVR College of Engineering Department of CSE Sanketika Vidya Parishad Engineering College

ISBN: (数字)9798331501488

ISBN: (纸本)9798331501495

Existing security protocols encounter major cybersecurity difficulties because the online devices' growing popularity continues to spread across networks. These security protocols are ineffective because they do not properly handle current cyber threats. The main goal of this study involves developing enhanced IoT cybersecurity through the development of a threat detection system which brings together adversarial training and deep learning models (CNN-LSTM) and Federated Learning (FL). The system enables distributed Internet of Things devices to work on security model development through Federated Learning while maintaining total privacy of their information. Security procedures controlled by generative artificial intelligence robots alongside real-time attack protection functions decrease security response durations. Through its Federated CNN-LSTM model the system upholds a 1.2% false positive rate alongside a 98.3% accuracy evaluation and 160 milliseconds of exact threat tracking time. The designed system sustains a minimal occurrence of incorrect alarm activations. The developed system provides real-time security for the Internet of Things framework because it enables adaptive protection systems while preserving user privacy in current IoT settings.

关键词： Training Privacy Protocols Federated learning Real-time systems Threat assessment Internet of Things Computer security Protection Robots

来源：评论

学校读者我要写书评

暂无评论

Aspect category and sentiment polarity detection using a permutation language-based transformer model

引用

Procedia Computer Science 2025年 258卷 4179-4189页

作者： Ashok Kumar Jayaraman Tina Ester Trueman Gayathri Ananthakrishnan Abirami Murugappan Erik Cambria Manikandan G Satanik Mitra HCLTech Engineering and R&D Services Chennai Tamil Nadu India Department of Computer Science University of the People California USA School of Computer Science Engineering and Information Systems Vellore Institute of Technology Vellore India Department of Information Science and Technology Anna University Chennai India College of Computing and Data Science Nanyang Technological University Nanyang Singapore School of Computer Science and Engineering Vellore Institute of Technology Vellore India Department of Management BITS Pilani India

Aspect-based sentiment analysis (ABSA) performs fine-grained analysis on text to determine a specific aspect category and a sentiment polarity. Recently, machine learning models have played a key role in ABSA tasks. In particular, transformer-based pre-trained models have achieved promising results in natural language processing tasks. Therefore, we propose a permutation based XLNet fine-tuning model for aspect category detection and sentiment polarity detection. Our model learns bidirectional contexts via positional encoding and factorization order. We evaluate the proposed permutation language model on three ABSA datasets, namely, SentiHood, SemEval 2015, and SemEval 2016. Specifically, we studied the ABSA tasks in a constrained system with a multi-class environment. Our result indicates that the proposed permutation language model achieves a better result.

关键词： Aspect-based sentiment analysis permutation language model deep learning transformers multiclass classification

来源：评论

学校读者我要写书评

暂无评论

Transformer Embeddings for the Implementation of a Semantic MOOC Recommender System

Transformer Embeddings for the Implementation of a Semantic ...

引用

International Conference on Intelligent Computing in data Sciences (ICDS)

作者： Sara Assami Najima Daoudi Rachida Ajhoun Department of computer science Ecole Nationale Supérieure d'Informatique et d'Analyse des Systèmes of Mohammed V University Rabat Morocco Department of data and knowledge engineering Ecole des Sciences de l'Information Rabat Morocco

ISBN: (数字)9798350351200

ISBN: (纸本)9798350351217

Semantic web technologies aim to interconnect objects using descriptors that identify their characteristics and improve information retrieval. In Massive Open Online Courses (MOOCs), the learning contents semantic links and their pertinence for learner needs are essential to motivate learners and reduce dropout rates. This work aims to develop a semantic recommender system prototype that uses external data sources on learners to identify their knowledge, skills, and interests to recommend MOOCs. First, this paper traces the knowledge-based recommendation of MOOCs and its data representation in the literature. Afterward, it explores a dataset of a MOOC platform to extract MOOCs content-related data, and a dataset of a professional social media to extract knowledge-based characteristics of a learner from his/her social profile. For the implementation of the semantic recommender, three different techniques: TF-IDF, GloVe, and USE models are used to compute the semantic similarity of learner needs and MOOCs content. The semantic MOOC recommender system implementation showed the advantages of using different data sources and the increased performance of a MOOC recommender system based on transformer models like the USE model.

关键词： Computer aided instruction Electronic learning Social networking (online) Computational modeling Semantics knowledge based systems data visualization Transformers data models Recommender systems

来源：评论

学校读者我要写书评

暂无评论

TREC: transient redundancy elimination-based convolution 22

TREC: transient redundancy elimination-based convolution

引用

Proceedings of the 36th International Conference on Neural information Processing systems

作者： Jiawei Guan Feng Zhang Jiesong Liu Hsin-Hsuan Sung Ruofan Wu Xiaoyong Du Xipeng Shen Key Laboratory of Data Engineering and Knowledge Engineering (MOE) and School of Information Renmin University of China Computer Science Department North Carolina State University

ISBN: (纸本)9781713871088

The intensive computations in convolutional neural networks (CNNs) pose challenges for resource-constrained devices; eliminating redundant computations from convolution is essential. This paper gives a principled method to detect and avoid transient redundancy, a type of redundancy existing in input data or activation maps and hence changing across inferences. By introducing a new form of convolution (TREC), this new method makes transient redundancy detection and avoidance an inherent part of the CNN architecture, and the determination of the best configurations for redundancy elimination part of CNN backward propagation. We provide a rigorous proof of the robustness and convergence of TREC-equipped CNNs. TREC removes over 96% computations and achieves 3.51× average speedups on microcontrollers with minimal (about 0.7%) accuracy loss.

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：