检索结果-内蒙古大学图书馆

Balancing Feature Alignment and Uniformity for Few-Shot Classification

IEEE Transactions on Image Processing 2023年 PP卷 1-1页

作者： Yu, Yunlong Zhang, Dingyi Ji, Zhong Li, Xi Han, Jungong Zhang, Zhongfei College of Information Science and Electronic Engineering Zhejiang University China School of Electrical and Information Engineering Tianjin University Tianjin China College of Computer Science Zhejiang University Hangzhou China Department of Computer Science Aberystwyth University U.K Computer Science Department Watson School State University of New York Binghamton University Binghamton NY USA

In Few-Shot Learning (FSL), the objective is to correctly recognize new samples from novel classes with only a few available samples per class. Existing methods in FSL primarily focus on learning transferable knowledge from base classes by maximizing the information between feature representations and their corresponding labels. However, this approach may suffer from the “supervision collapse" issue, which arises due to a bias towards the base classes. In this paper, we propose a solution to address this issue by preserving the intrinsic structure of the data and enabling the learning of a generalized model for the novel classes. Following the InfoMax principle, our approach maximizes two types of mutual information (MI): between the samples and their feature representations, and between the feature representations and their class labels. This allows us to strike a balance between discrimination (capturing class-specific information) and generalization (capturing common characteristics across different classes) in the feature representations. To achieve this, we adopt a unified framework that perturbs the feature embedding space using two low-bias estimators. The first estimator maximizes the MI between a pair of intra-class samples, while the second estimator maximizes the MI between a sample and its augmented views. This framework effectively combines knowledge distillation between class-wise pairs and enlarges the diversity in feature representations. By conducting extensive experiments on popular FSL benchmarks, our proposed approach achieves comparable performances with state-of-the-art competitors. For example, we achieved an accuracy of 69.53% on the miniImageNet dataset and 77.06% on the CIFAR-FS dataset for the 5-way 1-shot task. IEEE

关键词： Distillation

来源：评论

学校读者我要写书评

暂无评论

On Pareto-frontier Approximate Computing for Many-core Systems

On Pareto-frontier Approximate Computing for Many-core Syste...

引用

2021 International Conference on Intelligent Technology and Embedded Systems, ICITES 2021

作者： Hou, Xinyue Wang, Xiaohang Palesi, Maurizio Singh, Amit Kumar Jiang, Yingtao Yang, Mei Huang, Letian Chen, Junying Ministry of Education University of Catania Department of Electrical Electronics and Computer Engineering Catania Italy University of Essex School of Computer Science and Electronic Engineering Colchester United Kingdom University of Nevada Department of Electrical and Computer Engineering Las Vegas United States University of Electronic Science and Technology of China School of Electronic Science and Engineering China

ISBN: (纸本)9781665427555

Approximate computing is an emerging paradigm that aggressively improves performance or reduces energy consumption by sacrificing computation quality for error forgiving applications. Various approximate techniques, including loop truncation, approximate communication, etc. have been proposed. Previous works focus on optimization using only one approximation knob. However, we have observed that simultaneously optimizing with multiple approximation knobs leads to a large search space and is more likely to find better solutions. Therefore, in this paper, we first develop application models for performance, error, and power, followed by formulation of an optimization problem to maximize system performance under error and power constraints, using three approximation knobs, which are loop truncation, data dropping, and computational precision scaling. In order to solve the problem efficiently, a lightweight algorithm based on interior point algorithm is proposed. Experimental results show that, compared to state-of-the-art approximate approaches, the proposed scheme can reduce the execution time by as much as 33.1%. The overhead of the proposed method is low, making it a suitable approximate scheme for future many-core systems. © 2021 IEEE.

关键词： Errors

来源：评论

学校读者我要写书评

暂无评论

Analysis and Design of Vector Holographic Metasurfaces

Analysis and Design of Vector Holographic Metasurfaces

引用

Progress in Electromagnetic Research Symposium (PIERS)

作者： A. Pitilakis O. Tsilipakos A. C. Tasolamprou A. Tsioliaridou N. V. Kantartzis S. Ioannidis C. Liaskos M. Kafesaki Foundation for Research and Technology Hellas Institute of Electronic Structure and Laser Greece School of Electrical and Computer Engineering Aristotle University of Thessaloniki Greece National Hellenic Research Foundation Theoretical and Physical Chemistry Institute Greece Department of Physics National and Kapodistrian University of Athens Greece Foundation for Research and Technology Hellas Institute of Computer Science Greece School of Electrical & Computer Engineering Technical University of Crete Greece Department of Computer Science and Engineering University of Ioannina Greece Department of Materials Science and Technology University of Crete Greece

ISBN: (数字)9798350375909

ISBN: (纸本)9798350375916

In this work we present a comprehensive framework for the analysis and design of vector holographic metasurfaces (VHMS). The term vector refers to the metasurface ability to manipulate the polarization of the scattered wave, while holographic refers to the configuration of the metasurface aperture, which can have any arbitrary complex profile in its equivalent surface impedance (phase and/or magnitude). We start from the presentation of the analysis of such VHMS, i.e., the calculation of its scattering under given equivalent surface impedance configuration and illumination; the analysis relies on physical optics and can, in theory, model the scattering and diffraction by a finite-aperture metasurface of arbitrary convex shape under arbitrary spatially coherent wavefront illumination. We then present the VHMS design and synthesis procedure, i.e., we compute its equivalent and physical configuration, when given the illuminating wavefront and the desired scattering response or functionality, e.g., beam steering or focusing. For the synthesis, we use a simple scalable unit cell topology incorporating square conductor patches interconnected with lumped complex loads; these can locally and continuously tune the VHMS phase and amplitude response, with independent control over both polarizations, thus allowing arbitrary incident-to-scattered wavefront transformations. These versatile devices, especially when fused with wavefront sensing capabilities, can be used in applications spanning wireless communications (e.g., beamforming, integrated/joint sensing and communication, or programmable wireless environments), radar scattering (e.g., stealth), and holography/imaging.

关键词： Wireless communication Surface impedance Surface waves Scattering Physical optics Lighting Metasurfaces

来源：评论

学校读者我要写书评

暂无评论

MetaCo-Distillation: A Robust Collaborative Distillation Framework for Multivariate Knowledge

MetaCo-Distillation: A Robust Collaborative Distillation Fra...

引用

2024 IEEE International Conference on Smart City (SmartCity)

作者： Jianing Wang Jiqiang Liu Tao Zhang Jian Wang Xiangyun Tang Geng Sun Zhaohui Yang Zhu Han Dusit Niyato School of Cyberspace Science and Technology Beijing Jiaotong University Beijing China School of Information Engineering Minzu University of China Beijing China College of Computer Science and Technology Jilin University Changchun China College of Information Science and Electronic Engineering Zhejiang University Hangzhou China Department of Electrical and Computer Engineering University of Houston Houston TX USA School of Computer Science and Engineering Nanyang Technological University Singapore

ISBN: (数字)9798331540487

ISBN: (纸本)9798331540494

Federated Learning (FL), as a rapidly evolving privacy-preserving collaborative machine learning paradigm, has shown great potential in the emerging Metaverse areas. Although numerous studies have theoretically demonstrated the feasibility of FL, practical applications in the Metaverse face challenges such as learning forgetting due to non-independent and identically distributed (non-i.i.d.) data, high communication overhead from model parameter transmission, and the inability to accommodate heterogeneous models. To address these challenges, this paper proposes a collaborative distillation framework called Co-distillation. In this distributed environment, each device independently trains its local model using private data, communicates predictions on an unlabeled public dataset, and generates surrogate labels for the public dataset based on received predictions. These surrogate labels, along with the received predictions, serve as teacher logits for local model distillation. This approach is more adaptable to heterogeneous streaming data continuously generated by industrial devices, offers higher communication efficiency than traditional methods, allow for heterogeneous models. Experiments have been conducted, demonstrating the superior performance of the proposed framework.

关键词： Performance evaluation Training Adaptation models Metaverse Smart cities Collaboration Distributed databases Predictive models Data models Security

来源：评论

学校读者我要写书评

暂无评论

Hashing-based Undersampling for Large Scale Histopathology Image Classification 21

Hashing-based Undersampling for Large Scale Histopathology I...

引用

21st IEEE International Conference on Cognitive Informatics and Cognitive Computing, ICCI*CC 2022

作者： Tian, Xing Qiu, Lin Li, Qihua Ng, Wing W. Y. Zhang, Jianjun Kwong, Sam Wang, Hui Dong, Xinran Liu, Baoyi Hu, Yijun Yu, Honghua School of Computer Science and Engineering South China University of Technology Guangzhou China Hong Kong City University Department of Computer Science Hong Kong Hong Kong School of Electronics Electrical Engineering and Computer Science Queen's University Belfast United Kingdom Guangdong Eye Institute Guangdong Provincial People's Hospital Department of Ophthalmology Guangdong Academy of Medical Sciences Guangzhou China Aier School of Ophthalmology Central South University Changsha China

ISBN: (纸本)9781665490849

The early diagnosis of cancer based on histopathology images plays an important role in medical science. Existing techniques generally partition the original histopathology image into small pieces for further classification. However, due to the fact that the number of benign (majority) samples is much larger than that of malignant (minority) samples, the classification is significantly imbalanced which adversely affects classification performance. Undersampling is commonly used to address the class-imbalance problem. However, existing methods are typically time consuming so they are not suitable to handle large-scale and high-dimensional data. In this paper we propose a fast and scalable undersampling method, hashing-based undersampling (HBU), to address class imbalance in large-scale medical image classification. Benign images are hashed and then placed into different buckets according to their locations in the input space. Undersampling is achieved by proportionally selecting benign images from the hash buckets. The HBU method is experimentally evaluated on two real histopathology image datasets, CAMELYON16 and ACDC@LUNGHP, by comparison with existing methods. Experimental results show that the HBU method outperforms six state-of-The-Art methods and is scalable and fast. © 2022 IEEE.

关键词： Image classification

来源：评论

学校读者我要写书评

暂无评论

Investigating Labeled Cyberbullying Incidents on the Weibo Social Network

Investigating Labeled Cyberbullying Incidents on the Weibo S...

引用

IEEE International Conference on Networking, Sensing and Control

作者： Jun Xu Songnan Yu Xiaoyu Sean Lu Di Wu Qi Kang School of Cyber Science and Engineering Nanjing University of Science and Technology Nanjing China Department of Electrical and Computer Engineering New Jersey Institute of Technology Newark New Jersey US

ISBN: (数字)9798350365221

ISBN: (纸本)9798350365238

This study aims to investigate cyberbullying incidents on Weibo by constructing a comprehensive dataset of labeled conversations. We collect 89K social media sessions from 10K user profiles and manually annotated the data to identify instances of cyberbullying. We analyze cyberbullying based on three fundamental characteristics, examining its distribution across different topics and the relationship between posting times and cyberbullying incidents. Additionally, we explore the influence of user popularity on the similarity of receiving bullying comments. Our study provides valuable insights into the characteristics of cyberbullying on Chinese social media and highlights the need for culturally sensitive detection models.

关键词： Blogs Cyberbullying Oral communication Sensors

来源：评论

学校读者我要写书评

暂无评论

Effectiveness of Data Augmentation for Parameter Efficient Tuning with Limited Data

arXiv

引用

arXiv 2023年

作者： Obadinma, Stephen Guo, Hongyu Zhu, Xiaodan Department of Electrical and Computer Engineering Queen’s University Canada Digital Technologies Research Centre National Research Council Canada School of Electrical Engineering and Computer Science University of Ottawa Canada

Recent work has demonstrated that using parameter efficient tuning techniques such as prefix tuning (or P-tuning) on pretrained language models can yield performance that is comparable or superior to fine-tuning while dramatically reducing trainable parameters. Nevertheless, the effectiveness of such methods under the context of data augmentation, a common strategy to improve learning under low data regimes, has not been fully explored. In this paper, we examine the effectiveness of several popular task-agnostic data augmentation techniques, i.e., EDA, Back Translation, and Mixup, when using two general parameter efficient tuning methods, P-tuning v2 and LoRA, under data scarcity. We show that data augmentation can be used to boost the performance of P-tuning and LoRA models, but the effectiveness of each technique varies and certain methods can lead to a notable degradation in performance, particularly when using larger models and on harder tasks. We further analyze the sentence representations of P-tuning compared to fine-tuning to help understand the above behaviour, and reveal how P-tuning generally presents a more limited ability to separate the sentence embeddings from different classes of augmented data. In addition, it displays poorer performance on heavily altered data. However, we demonstrate that by adding a simple contrastive loss function it can help mitigate such issues for prefix tuning, resulting in sizable improvements to augmented data performance. Copyright © 2023, The Authors. All rights reserved.

关键词： Machine learning

来源：评论

学校读者我要写书评

暂无评论

A Comparative Analysis of Federated Learning Towards Big data IoT with Future Perspectives

A Comparative Analysis of Federated Learning Towards Big dat...

引用

Computing and Information Technology (ICCIT-1441), 2020 International Conference on

作者： G. Anitha A. Jegatheesan E. Baburaj Department of Computer Science and Engineering Saveetha School of Engineering Saveetha Institute of Medicaland Technical Sciences Chennai India Department of Electrical and Computer Engineering College of Engineering and Technology Bule Hora University Ethiopia

Over the past few years, big data has come a long way, allowing us to make sense of the immense amounts of information being produced by cutting-edge services and apps, as well as a plethora of IoT devices. Only through extracting value from huge data through data analytics can big data's full potential be accomplished [6]. Machine learning (ML) plays a highly essential part in this process because of its capacity to know patterns and then deliver insights through learning from facts. Since the data may include private information about individuals, governments, or financial accounts, using a more traditional technique raises serious concerns about data confidentiality. Recently, Federated learning (FL) seemed like a potential learning procedure that could help us get beyond this obstacle. In FL, the central ML algorithm is only given the parameters from the local devices so that it can train and make predictions on a global scale. The aforementioned problems with huge data may be amenable to resolution thanks to this aspect of FL. Furthermore, a thorough assessment of FL for big data applications has forward done, leaving a void in the literature. This article delivers a summary of FL, big data, and the motives for the procedure of FL for big data by surveying its use in big data applications. In particular, it is important to thoroughly examine the usage of FL for major big data services with applications like the smart city, remote surgery, transportation, smart grid, and social networks. Additionally, briefly discuss the most pressing problems associated with this engaging subject and the most potential approaches to fixing them.

关键词：

来源：评论

学校读者我要写书评

暂无评论

An Actor-Critic Reinforcement Learning Scheme for Reactive 3D Optimal Motion Planning Based on Fluid Dynamics

An Actor-Critic Reinforcement Learning Scheme for Reactive 3...

引用

IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS)

作者： Marios Malliaropoulos Panagiotis Rousseas Charalampos P. Bechlioulis Kostas J. Kyriakopoulos School of Mechanical Engineering Control Systems Laboratory National Technical University of Athens Department of Electrical and Computer Engineering University of Patras Center of AI & Robotics (CAIR) New York University Abu Dhabi

ISBN: (数字)9798350377705

ISBN: (纸本)9798350377712

This work proposes a novel and provably correct method for three-dimensional optimal motion planning in complex environments. Our approach models the 3D motion planning problem by solving streamlines of the potential fluid flow, filling a gap in traditional motion planning techniques by guaranteeing a closed-loop, smooth and natural-looking navigation solution. Special emphasis is given to an inherent challenge of artificial potential field (APF) methods, namely establishing proofs of safety and stability over the entire optimization process. A model-based actor-critic reinforcement learning algorithm is introduced to approximate the optimal solution to the Hamilton-Jacobi-Bellman equation and update the controller parameters in a deterministic manner. Through a series of ROS-Gazebo software-in-the-loop simulations the proposed methodology demonstrates robustness and outperforms widely used methods such as the RRT ∗ , highlighting its contribution to the field of 3D optimal motion planning.

关键词： Solid modeling Three-dimensional displays Navigation Reinforcement learning Mathematical models Stability analysis Robustness Planning Safety Optimization

来源：评论

学校读者我要写书评

暂无评论

Another Lattice Attack Against ECDSA with the wNAF to Recover More Bits per Signature 1

引用

18th EAI International Conference on Security and Privacy in Communication Networks, SecureComm 2022

作者： Ma, Ziqiang Li, Shuaigang Lin, Jingqiang Cai, Quanwei Fan, Shuqin Zhang, Fan Luo, Bo School of Information Engineering Ningxia University Yinchuan China State Key Laboratory of Information Security Institute of Information Engineering Chinese Academy of Sciences Beijing China School of Cyber Security University of Science and Technology of China Hefei China State Key Laboratory of Cryptology Beijing China School of Cyber Science and Technology College of Computer Science and Technology Zhejiang University Hangzhou310027 China Key Laboratory of Blockchain and Cyberspace Governance of Zhejiang Province Hangzhou310027 China Department of Electrical Engineering and Computer Science University of Kansas Lawrence United States

ISBN: (数字)9783031255380

ISBN: (纸本)9783031255373

In the resource-constrained environment such as the Internet of Things, the windowed Non-Adjacent-Form (wNAF) representation is usually used to improve the calculation speed of the scalar multiplication of ECDSA. This paper presents a practical cache side channel attack on ECDSA implementations which use wNAF representation. Compared with existing works, our method exploits more information from the cache side channels, which is then efficiently used to construct lattice attacks in the ECDSA private key recovery. First, we additionally monitor the invert function which is related to the sign of the wNAF digits, and obtain a Double-Add-Invert chain through the Flush+Flush cache side channel. Then, we develop effective methods extracting 154.2 bits information of the ephemeral key per signature for 256-bit ECDSA from this chain, much more than the best known result which extracts 105.8 bits per signature. Finally, to efficiently use the extracted information, we convert the problem of recovering the private key to the Hidden Number Problem (HNP) and the Extended Hidden Number Problem (EHNP) respectively, which are solved by lattice reduction algorithms. We applied the attack on ECDSA with the secp256k1 curve in OpenSSL 1.1.0h. The experimental results show that only 3 signatures are enough to recover the private key. To the best of our knowledge, this work exploits the signs of the wNAF representation, along with the Double-Add chain against ECDSA, to recover the private key with the least number of signatures. © 2023, ICST Institute for computer sciences, Social Informatics and Telecommunications engineering.

关键词： Side channel attack

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：