检索结果-内蒙古大学图书馆

41st International Conference on Machine Learning, ICML 2024

作者： Yue, Sheng Hua, Xingyuan Ren, Ju Lin, Sen Zhang, Junshan Zhang, Yaoxue Department of Computer Science and Technology Tsinghua University Beijing China Zhongguancun Laboratory Beijing China Department of Computer Science University of Houston Texas United States Department of Electrical and Computer Engineering University of California Davis United States

In this paper, we study offline-to-online Imitation Learning (IL) that pretrains an imitation policy from static demonstration data, followed by fast finetuning with minimal environmental interaction. We find the naïve combination of existing offline IL and online IL methods tends to behave poorly in this context, because the initial discriminator (often used in online IL) operates randomly and discordantly against the policy initialization, leading to misguided policy optimization and unlearning of pretraining knowledge. To overcome this challenge, we propose a principled offline-to-online IL method, named OLLIE, that simultaneously learns a near-expert policy initialization along with an aligned discriminator initialization, which can be seamlessly integrated into online IL, achieving smooth and fast finetuning. Empirically, OLLIE consistently and significantly outperforms the baseline methods in 20 challenging tasks, from continuous control to vision-based domains, in terms of performance, demonstration efficiency, and convergence speed. This work may serve as a foundation for further exploration of pretraining and finetuning in the context of IL. Copyright 2024 by the author(s)

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Cross-Layer Optimized HARQ Protocol for Underwater Acoustic Wireless Sensor Networks Using Stochastic Network Calculus 2

Cross-Layer Optimized HARQ Protocol for Underwater Acoustic ...

引用

2nd IEEE International Conference on Emerging Research in Computational science, ICERCS 2024

作者： Sivajayaprakash, A. Sukumaran, Rajeev Kumar, C. SRM Institute of Science and Technology Department of Computer Science and Engineering Kattankulathur India Karpagam College of Engineering Department of Electrical and Electronics Engineering Coimbatore641032 India

ISBN: (纸本)9798331534950

UAWSNs face challenges such as long propagation delays, limited bandwidth, and varying channel conditions. To solve these problems, we developed a new protocol called Multi- Hop Cross-Layer Optimized Hybrid Automatic Repeat Request (CLO-HARQ) suitable for UAWSNs. This protocol automatically changes the retransmission strategy and error correction redundancy based on real-time feedback from the physical and network layers. A unique feature of this protocol is the inclusion of environmental factors such as water salinity and temperature in the decision-making process. Also, using Stochastic Network Calculus (SNC), stochastic traffic characteristics predict single-hop and multi-hop communication delay (delay), energy usage (energy usage), and throughput. In the performance evaluation of CLO-HARQ, it has been shown to provide significant improvement in efficiency, throughput, and delay over existing HARQ methods. It is well-suited for long-term underwater monitoring and is a sustainable solution for energy-constrained sensor networks. © 2024 IEEE.

关键词： Network layers

来源：评论

学校读者我要写书评

暂无评论

Optimizing Monocular 3D Object Detection on KITTI: Harnessing Power of Right Images 20

Optimizing Monocular 3D Object Detection on KITTI: Harnessin...

引用

20th CSI International Symposium on Artificial Intelligence and Signal Processing, AISP 2024

作者： Bakhtiarian, Ali Karimi, Nader Samavi, Shadrokh Isfahan University of Technology Department of Electrical and Computer Engineering Isfahan84156-83111 Iran Seattle University Computer Science Department Seattle98122 United States

ISBN: (纸本)9798350383942

Monocular 3D object detection is an important yet challenging problem in computer vision, with applications such as autonomous driving. A key limitation in advancing this field is the scarcity of annotated training data, an issue exacerbated in benchmarks like KITTI, which provide only around 7,000 labeled images. Prior arts have developed techniques to improve monocular 3D detection but often rely on external sources of data like LiDAR to supplement the limited training images. In this work, we propose a pre-training strategy that addresses the limited data issue by leveraging unlabeled right-camera images available within the KITTI dataset itself. We pre-train a model initialized for 3D detection by using 'right' views before fine-tuning on just 'left' images. Our experiments validate that, through this strategic pre-training with readily available 'right' images, significant improvements can be achieved over models trained from scratch on only 'left' images. We observe consistent gains in 3D detection performance when leveraging 'right' image pre-training, without requiring any external LiDAR data. Our method provides evidence that mining unlabeled or weakly labeled in-domain data can effectively remedy the pervasive challenge of limited training data for monocular 3D object detection. This offers a plug-and-play practical strategy to use available datasets like KITTI better and reduce overfitting. © 2024 IEEE.

关键词： Object detection

来源：评论

学校读者我要写书评

暂无评论

Design of electric powertrains to achieve NVH performance using autoencoders and a physical meaningful latent space

引用

Neural Computing and Applications 2025年第16期37卷 10287-10308页

作者： Ricardo Souza, Marcos Offner, Guenter Soltoggio, Andrea Mohammadpour, Mahdi Theodossiades, Stephanos Wolfson School of Mechanical Electrical and Manufacturing Engineering Loughborough University Loughborough United Kingdom AVL List GmbH Graz Austria Computer Science Department School of Science Loughborough University Loughborough United Kingdom

One of the fundamental differences in the perception of electric (e-) vehicles is how their radiated noise is perceived with respect to classic internal combustion engines. Even though e-vehicles are usually quieter, the tonal content of the radiated noise can be more annoying. This paper proposes a novel approach that starts from the assumed radiated noise spectrum profile as input to a neural network that can return powertrain design parameters that would lead to generation of that specific noise profile. The proposed network acts as an autoencoder where the latent space is forced to have a physical meaning. As diverse combinations of powertrain parameters can result in similar noise profiles, a variational autoencoder is used to learn a structured latent representation, ensuring continuity and smooth transitions between possible solutions. The network predictions are validated against results of a three-dimensional CAE e-powertrain model. Overall, the mean absolute error is around 5 dBA for this feasibility study, which aims to demonstrate the concept. This work takes an inverse approach to the optimisation problem by starting from the user-perceived noise to predict the parameters required to achieve that. Although this study focuses solely on gear teeth microgeometry changes and bearing preloads, additional powertrain parameters could be incorporated as needed. © The Author(s) 2025.

关键词： Internal combustion engines

来源：评论

学校读者我要写书评

暂无评论

Towards Zero Trust Security for Prosumer-Driven Verifiable Green Energy Certificates 7

Towards Zero Trust Security for Prosumer-Driven Verifiable G...

引用

7th International Conference on Energy Conservation and Efficiency, ICECE 2024

作者： Rasool, Saqib Saleem, Afshan Ikram Ul Haq, Muhammad Jacobsen, Rune Hylsberg Aarhus University Department of Electrical and Computer Engineering Aarhus Denmark Nexthon Technologies Gujrat Pakistan Superior University Lahore Department of Computer Science Lahore Pakistan

ISBN: (纸本)9798350370898

In response to the critical need for environmentally sustainable practices, green energy certificates have emerged as a key mechanism to prevent greenwashing and ensure the authenticity of renewable energy sources. While incorporating blockchain technology has significantly enhanced the trans-parency and verifiability of these certificates, it has also intro-duced a trade-off in terms of increased operational costs due to the added control mechanisms of the blockchain. This challenge led to recent advancements in the state-of-the-art blockchain-based framework for verifiable green energy certificates, integrating trustworthy registries to balance trust and operational efficiency to enhance feasibility for real-world deployments. Our paper extended the same framework to elevate the trust factor without incurring additional operational costs of blockchain. At the heart of our proposed solution is integrating Self-Sovereign Identity (SSI), a technology that bolsters the system's reliability and empowers users/devices with control over their digital identities. By harmonizing the strengths of the Distributed Ledger Technologies (DLT) of IOTA with the public blockchain of Concordium, our framework maintains the integrity and transparency of the green energy certification process. It propels the system towards zero-trust security by distributing trust from a few centralized registries to many decentralized prosumer meters. © 2024 IEEE.

关键词： Blockchain

来源：评论

学校读者我要写书评

暂无评论

On the Online Weighted Non-Crossing Matching Problem 19

On the Online Weighted Non-Crossing Matching Problem

引用

19th Scandinavian Symposium on Algorithm Theory, SWAT 2024

作者： Boyar, Joan Kamali, Shahin Larsen, Kim S. Lavasani, Ali Mohammad Li, Yaqiao Pankratov, Denis Department of Mathematics and Computer Science University of Southern Denmark Odense Denmark Department of Electrical Engineering and Computer Science York University Toronto Canada Department of CSSE Concordia University Montreal Canada

ISBN: (纸本)9783959773188

We introduce and study the weighted version of an online matching problem in the Euclidean plane with non-crossing constraints: 2n points with non-negative weights arrive online, and an algorithm can match an arriving point to one of the unmatched previously arrived points. In the vanilla model, the decision on how to match (if at all) a newly arriving point is irrevocable. The goal is to maximize the total weight of matched points under the constraint that straight-line segments corresponding to the edges of the matching do not intersect. The unweighted version of the problem was introduced in the offline setting by Atallah in 1985, and this problem became a subject of study in the online setting with and without advice in several recent papers. We observe that deterministic online algorithms cannot guarantee a non-trivial competitive ratio for the weighted problem. We study various regimes of the problem which permit non-trivial online algorithms. In particular, when weights are restricted to the interval [1, U] we give a deterministic algorithm achieving competitive ratio Ω ( 2−2√logU). We also prove that deterministic online algorithms cannot achieve competitive ratio better than O ( 2−√logU) . Interestingly, we establish that randomization alone suffices to achieve competitive ratio 1/3 even when there are no restrictions on the weights. Additionally, if one allows an online algorithm to revoke acceptances, then one can achieve a competitive ratio ≈ 0.2862 deterministically for arbitrary weights. We also establish a lower bound on the competitive ratio of randomized algorithms in the unweighted setting, and improve the best-known bound on advice complexity to achieve a perfect matching. © Joan Boyar, Shahin Kamali, Kim S. Larsen, Ali Mohammad Lavasani, Yaqiao Li, and Denis Pankratov.

关键词： Geometry

来源：评论

学校读者我要写书评

暂无评论

End-to-End Secure Video Streaming Using Homomorphic Encryption Techniques 2

End-to-End Secure Video Streaming Using Homomorphic Encrypti...

引用

2nd International Conference on Artificial Intelligence, Blockchain, and Internet of Things, AIBThings 2024

作者： Darwich, Mahmoud Khalil, Kasem Bayoumi, Magdy University of Mount Union Department of Mathematics and Computer Science AllianceOH44601 United States University of Mississippi University Department of Electrical and Computer Engineering MS38677 United States University of Louisiana Department of Electrical and Computer Engineering LafayetteLA70504 United States

ISBN: (纸本)9798331529529

In the rapidly evolving landscape of digital media, video streaming has become a ubiquitous method of content delivery. However, the increasing sophistication of cyber threats poses significant risks to the security and privacy of streamed content. Traditional encryption methods, while effective in many scenarios, often fall short in providing comprehensive protection without compromising performance and user experience. This paper introduces a novel approach to achieving end-to-end secure video streaming by leveraging homomorphic encryption techniques. Homomorphic encryption allows for computations to be performed on encrypted data without needing to decrypt it first, thereby maintaining data privacy and security throughout the streaming process. Our proposed method integrates homomorphic encryption into the video streaming pipeline, ensuring that video data remains secure from the point of origin to the end user. We discuss the theoretical foundations of homomorphic encryption, detail the implementation of our approach, and evaluate its performance against conventional encryption methods. Experimental results demonstrate that our technique provides robust security with minimal impact on streaming quality and latency, making it a viable solution for secure video streaming in various applications. © 2024 IEEE.

关键词： Video streaming

来源：评论

学校读者我要写书评

暂无评论

hChain 2.0: Leveraging Blockchain and Distributed File System for EHR Management in Smart Healthcare

引用

SN computer science 2025年第2期6卷 1-19页

作者： Alruwaill, Musharraf N. Mohanty, Saraju P. Kougianos, Elias Department of Computer Science and Engineering University of North Texas Denton TX United States Department of Electrical Engineering University of North Texas Denton TX United States

Within the context of the digital revolution, the domain of smart healthcare has emerged as a promising area with the aim of enhancing patient care, streamlining clinical operations, and facilitating prompt medical interventions. However, the main priority in smart healthcare revolves around the challenges of security and privacy. In addition to concerns over security and privacy, there are also other problems associated with the sharing of electronic healthcare records EHR through a robust mechanism that is both controlled and authenticated. The paper presents a proposed framework that aims to enhance security, privacy, and automation through the utilization of blockchain and IPFS technologies. Furthermore, the integration of wearable devices and deep learning techniques can be utilized to develop a risk evaluation system that facilitates fall detection and automated response, which helps with immediate medical assessment requests through a secured automated system. Integrating blockchain and IPFS provides a secure system that is cost-effective. Smart contracts provide secure and robust business logic and access control management. The integration of blockchain with IPFS offers a secure and cost-effective approach. The utilization of smart contracts enables reliable access control management and robust business logic. It provides a more restricted method for EHR sharing such as the quantity of accesses, duration of access, and specific permissions granted. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2025.

关键词： Blockchain Deep learning Healthcare cyber-physical systems (H-CPS) Long short-term memory (LSTM) Smart contract Smart healthcare

来源：评论

学校读者我要写书评

暂无评论

Unlocking MicroLED Potential: Damage-free Anisotropic Etching for Enhanced Pixel Density

Unlocking MicroLED Potential: Damage-free Anisotropic Etchin...

引用

CLEO: Applications and Technology in CLEO 2024, CLEO: A and T 2024 - Part of Conference on Lasers and Electro-Optics

作者： Chan, Clarence Y. Roberts, Henry C. Xiao, Yixin Mi, Zetian Li, Xiuling Department of Electrical and Computer Engineering University of Illinois at Urbana-Champaign UrbanaIL61801 United States Department of Electrical and Computer Engineering Microelectronics Research Center The University of Texas AustinTX78758 United States Department of Electrical Engineering and Computer Science University of Michigan Ann ArborMI48109 United States

One of the biggest hurdles in microLED-technology is the efficiency degradation with shrinking pixel-size, due to etching damage. We present the scaling of µLED from 45 to 5 µm by MacEtch, with near size-ind... 详细信息

关键词： Anisotropic etching

来源：评论

学校读者我要写书评

暂无评论

A Comprehensive Review on Object Detection in the Context of Autonomous Driving 4

A Comprehensive Review on Object Detection in the Context of...

引用

4th International Conference on Ubiquitous Computing and Intelligent Information Systems, ICUIS 2024

作者： Islam, Md Moynul Chowdhury, Ismatul Jannat Mahboob, Tanvir Zobair Mazumder, Md Shah Jalal Hossain, Mohammad Jakaria Biswas, Md Siam Rone, Poncanon Datta Maharishi International University Department of Computer Science IA United States University of North Carolina Department of Electrical Engineering CharlotteNC United States Chittagong University of Engineering and Technology Department of Computer Science and Engineering Chittagong Bangladesh

ISBN: (纸本)9798331529635

In autonomous driving, safety assessment is becoming an essential component, particularly in the perception and analysis of the environment around them. To make safe driving judgments, autonomous cars mostly rely on their capacity to observe and comprehend their environment. Detecting objects is a frequent and challenging problem in autonomous driving. Over the course of the last ten years, as deep learning has rapidly advanced, researchers have conducted numerous experiments and made significant contributions to the performance improvement of object recognition and related tasks including object categorization, localization, and segmentation. In this paper, an overview of state-of-the-art object detection algorithms is presented with in depth analysis. Region proposal-based approaches and single-shot approaches are two main frameworks of object detection which are discussed here. R-CNN, SPP-net, Fast R-CNN, and Faster R-CNN refer to region proposal-based methods while single-shot methods consist of YOLOv2, YOLOv3, YOLOv4 and SSD. The authors truly believe that this study would have a meaningful impact on the current advances in object detection algorithms. © 2024 IEEE.

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：