检索结果-内蒙古大学图书馆

FSCIL-EACA: Few-Shot Class-Incremental Learning network Based on Embedding Augmentation and Classifier Adaptation for Image Classification

引用

Chinese Journal of Electronics 2024年第1期33卷 139-152页

作者： Ruru ZHANG Haihong E Meina SONG School of Computer Science Beijing University of Posts and Telecommunications Education Department Information Network Engineering Research Center Beijing University of Posts and Telecommunications

The ability to learn incrementally is critical to the long-term operation of AI systems. Benefiting from the power of few-shot class-incremental learning(FSCIL), deep learning models can continuously recognize new classes with only a few samples. The difficulty is that limited instances of new classes will lead to overfitting and exacerbate the catastrophic forgetting of the old classes. Most previous works alleviate the above problems by imposing strong constraints on the model structure or parameters, but ignoring embedding network transferability and classifier adaptation(CA), failing to guarantee the efficient utilization of visual features and establishing relationships between old and new classes. In this paper, we propose a simple and novel approach from two perspectives: embedding bias and classifier bias. The method learns an embedding augmented(EA) network with cross-class transfer and class-specific discriminative abilities based on self-supervised learning and modulated attention to alleviate embedding bias. Based on the adaptive incremental classifier learning scheme to realize incremental learning capability,guiding the adaptive update of prototypes and feature embeddings to alleviate classifier bias. We conduct extensive experiments on two popular natural image datasets and two medical datasets. The experiments show that our method is significantly better than the baseline and achieves state-of-the-art results.

关键词： Few-shot class-incremental learning Embedding augmentation Classifier adaptation Image classification

来源：评论

学校读者我要写书评

暂无评论

Secure beamforming and deployment design for rate-splitting multiple access-based UAV communications

引用

Science China(Information Sciences) 2025年第1期68卷 342-356页

作者： Bing WANG Xiaofeng TAO Shujun HAN Huici WU Kai YANG Zhu HAN National Engineering Research Center of Mobile Network Technologies Beijing University of Posts and Telecommunications Peng Cheng Laboratory Department of Electrical and Computer Engineering University of Houston

Rate-splitting multiple access(RSMA) has recently gained attention as a potential robust multiple access(MA)scheme for upcoming wireless networks. Given its ability to efficiently utilize wireless resources and design interference management strategies, it can be applied to unmanned aerial vehicle(UAV) networks to provide convenient services for large-scale access ground users. However, due to the line-of-sight(LoS) broadcast nature of UAV transmission, information is susceptible to eavesdropping in RSMA-based UAV networks. Moreover, the superposition of signals at the receiver in such networks becomes complicated. To cope with the challenge, we propose a two-user multi-input single-output(MISO) RSMA-based UAV secure transmission framework in downlink communication networks. In a passive eavesdropping scenario, our goal is to maximize the sum secrecy rate by optimizing the transmit beamforming and deployment location of the UAV-base station(UAV-BS),while considering quality-of-service(QoS) constraints, maximum transmit power, and flight space limitations. To address the non-convexity of the proposed problem, the optimization problem is first decoupled into two subproblems. Then, the successive convex approximation(SCA) method is employed to solve each subproblem using different propositions. In addition, an alternating optimization(AO)-based location RSMA(L-RSMA) beamforming algorithm is developed to implement joint optimization to obtain the suboptimal solution. Numerical results demonstrate that(1) the proposed L-RSMA scheme yields a28.97% higher sum secrecy rate than the baseline L-space division multiple access(SDMA) scheme;(2) the proposed L-RSMA scheme improves the security performance by 42.61% compared to the L-non-orthogonal multiple access(NOMA) scheme.

关键词： rate-splitting multiple access UAV communication precoding optimization deployment location physical layer security

来源：评论

学校读者我要写书评

暂无评论

Distribution-Based Time-Varying Ensemble Scientific Data Reduction for Uncertainty Visualization and Analysis

引用

Journal of Information Science and Engineering 2024年第2期40卷 397-419页

作者： Lin, Zhi-Rong Hsiao, An-Lung Li, Guan Wang, Ko-Chih Department of Computer Science and Information Engineering National Taiwan Normal University Taipei106 Taiwan Computer Network Information Center Chinese Academy of Sciences Beijing100083 China

Scientists often study physical phenomena using computer simulation models. The same simulation can generate different datasets because of different input parameter configurations or internal random variables. Therefore, each grid point is represented by multiple data values from simulation runs, and we call this type of data ensemble dataset. To gain insight into the physical phenomenon, scientists often have to visualize and analyze the ensemble datasets with uncertainty. Distribution-based data representation is a popular approach to handling the ensemble dataset and supporting uncertainty visualization. However, storing a time-varying ensemble dataset needs hundreds or even thousands of times storage size. Given the size of the time-varying ensemble dataset, it is natural to develop storage-reduced data representation to facilitate the time-varying ensemble data exploration. We propose a novel data representation to compactly represent the time-varying scientific data for uncertainty visualization and analysis. Our approach decouples data on the temporal domain into two types of distributions and stores. One distribution summarizes the data values on the temporal domain, and the other distribution describes the occurrence probability of a data value on the temporal domain. Our approach can provide time-varying ensemble scientific data analysis with uncertainty quantification and detailed temporal feature evolution with less storage requirement. © 2024 Institute of Information Science. All rights reserved.

关键词： Probability distributions

来源：评论

学校读者我要写书评

暂无评论

Confidence Elicitation Improves Selective Generation in Black-Box Large Language Models

Confidence Elicitation Improves Selective Generation in Blac...

引用

2024 IEEE International Conference on Systems, Man, and Cybernetics, SMC 2024

作者： Liu, Sha Yue, Zhaojuan Li, Jun School of Computer Network Information Center Chinese Academy of Sciences and University of Chinese Academy of Sciences China Chinese Academy of Sciences Department of Computer Network Information Center China

ISBN: (纸本)9781665410205

Large language models (LLMs) exhibit impressive capabilities across various domains in natural language processing. However, LLMs can produce fictional content, which we refer to as hallucinations, and it makes the LLMs unreliable. An important research topic is how to make LLMs accurately express the confidence to their answers, so they can refrain from outputting or regenerate output in cases of low-confidence predictions. It facilitates the application of LLMs in high-stakes areas. Currently, research on eliciting calibrated confidence from LLMs is still insufficient. Additionally, methods for estimating uncertainty in responses based on internal parameters of LLMs become unavailable, as many existing LLMs are black boxes served via APIs. Therefore, we analyze the existing confidence elicitation methods and propose COVO, a new confidence elicitation method that allows the black-box LLMs to output their confidence levels by letting the LLM itself judges whether the answer comes from a reliable source. Our method does not require external knowledge and it has high computational efficiency. Experiments show that COVO achieves better calibration and effectively reduces hallucinations in LLMs through selective generation. Additionally, the confidence scores enhance the reliability of the LLMs' responses. © 2024 IEEE.

关键词： Digital elevation model

来源：评论

学校读者我要写书评

暂无评论

Towards real-time ML-based DDoS detection via cost-efficient window-based feature extraction

引用

Science China(Information Sciences) 2023年第5期66卷 114-129页

作者： Haibin LI Yi ZHAO Wenbing YAO Ke XU Qi LI Department of Computer Science and Technology Tsinghua University Beijing National Research Center for Information Science and Technology (BNRist) Institute for Network Sciences and Cyberspace Tsinghua University

Distributed denial of service(DDoS) detection is still an open and challenging problem. In particular, sophisticated attacks, e.g., attacks that disguise attack packets as benign traffic always appear, which can easily evade traditional signature-based methods. Due to the low requirements for computing resources compared to deep learning, many machine learning(ML)-based methods have been realistically deployed to address this issue. However, most existing ML-based DDo S detection methods are highly dependent on the features extracted from each flow, which incur remarkable detection delay and computation overhead. This article investigates the limitations of typical ML-based DDo S detection methods caused by the extraction of flow-level features. Moreover, we develop a cost-efficient window-based method that extracts features from a fixed number of packets periodically, instead of per flow, aiming to reduce the detection delay and computation overhead. The newly proposed window-based method has the advantages of well-controlled overhead and wide support of common routers due to its simplicity and high efficiency by design. Through extensive experiments on real datasets, we evaluate the performance of flow-based and window-based *** experimental results demonstrate that our proposed window-based method can significantly reduce the detection delay and computation overhead while ensuring detection accuracy.

关键词： DDoS attack machine learning feature extraction detection delay cost-efficiency

来源：评论

学校读者我要写书评

暂无评论

An Efficient Autoscaling Cross-Browser Testing Cloud Platform based on Selenium Grid, Kubernetes and KEDA

引用

Journal of Information Science and Engineering 2023年第5期39卷 1061-1077页

作者： Lin, Chia-Yu Lee, Shin-Jie Department of Computer Science and Information Engineering National Cheng Kung University Tainan701 Taiwan Computer and Network Center National Cheng Kung University Tainan701 Taiwan

Cross-browser testing not only is one of the most common non-functional testing methods in the field of software testing, but also the testing method that requires large amounts of resources, in terms of hardware and time. Basically, based on Selenium Grid, Kubernetes and KEDA auto-scaler, a cross-browser testing platform can be quickly built. However, through our empirical study of this style of platform, we observed three signifi-cant problems in terms of its reliability and efficiency: the Health-Check problem, the Ses-sion-Queue problem, and the Cooldown problem. This paper suggests solutions to these problems. The experimental result shows a 2.27 times improvement in reliability and a decrease in execution time for 61.5%. Moreover, the overall execution time is also 54.2% less comparing with Selenium's Dynamic Grid. © 2023 Institute of Information Science. All rights reserved.

关键词： Selenium

来源：评论

学校读者我要写书评

暂无评论

Federated Transfer Learning for On-Device LLMs Efficient Fine Tuning Optimization

引用

Big Data Mining and Analytics 2025年第2期8卷 430-446页

作者： Chuantao Li Bruce Gu Zhigang Zhao Youyang Qu Guomao Xin Jidong Huo Longxiang Gao Key Laboratory of Computing Power Network and Information Security Shandong Computer Science Center(National Supercomputer Center in Jinan)Qilu University of Technology(Shandong Academy of Sciences)Jinan 250000Chinaand also with Shandong Provincial Key Laboratory of Computer NetworksJinan 250000China Key Laboratory of Computing Power Network and Information Security Shandong Computer Science Center(National Supercomputer Center in Jinan)Qilu University of Technology(Shandong Academy of Sciences)Jinan 250000Chinaand also with Department of Computer ScienceFudan UniversityShanghai 200000China TelChina Group Co.Ltd. Jinan 250000China Key Laboratory of Computing Power Network and Information Security Shandong Computer Science Center(National Supercomputer Center in Jinan)Qilu University of Technology(Shandong Academy of Sciences)Jinan 250000Chinaand also with College Oceanography and Space InformaticsChina University of Petroleum(East China)Qingdao 266000China

The proliferation of Large Language Models(LLMs)has catalyzed the growth of various *** is therefore imperative to ensure the controlled and beneficial application of LLMs across specific domains for downstream tasks through transfer learning,while preserving their general *** propose a novel and on-device efficient fine-tuning optimization algorithm for LLMs,utilizing federated transfer ***,we introduce the Fusion of low Rank Adaptation(FoRA)optimization algorithm from a micro perspective,which enhances multi-dimensional feature aggregation through the addition of efficient *** a meso perspective,we extend the application of the FoRA algorithm across all linear layers within the Transformer architecture to facilitate downstream task ***,from a macro perspective and with a focus on the medical domain,we incorporate quantization techniques into the federated learning framework to achieve on-device efficient fine-tuning optimization,thereby offering dual protection for data and model *** results indicate that,compared to existing state-of-the-art methods,our algorithm significantly improves LLM performance while ensuring dual privacy protection of both data and models.

关键词： Federated learning fine-tuning deep learning Large Language Models(LLMs)

来源：评论

学校读者我要写书评

暂无评论

Cluster-preserving sampling algorithm for large-scale graphs

引用

Science China(Information Sciences) 2023年第1期66卷 60-76页

作者： Jianpeng ZHANG Hongchang CHEN Dingjiu YU Yulong PEI Yingjun DENG National Digital Switching System E&T Research Center Information Engineering University Network Systems Department of the Strategic Support Force School of Computer Science and Technology Eindhoven University of Technology Center for Applied Mathematics Tianjin University

Graph sampling is a very effective method to deal with scalability issues when analyzing largescale graphs. Lots of sampling algorithms have been proposed, and sampling qualities have been quantified using explicit properties(e.g., degree distribution) of the sample. However, the existing sampling techniques are inadequate for the current sampling task: sampling the clustering structure, which is a crucial property of the current networks. In this paper, using different expansion strategies, two novel top-leader sampling methods(i.e., TLS-e and TLS-i) are proposed to obtain representative samples, and they are capable of effectively preserving the clustering structure. The rationale behind them is to select top-leader nodes of most clusters into the sample and then heuristically incorporate peripheral nodes into the sample using specific expansion strategies. Extensive experiments are conducted to investigate how well sampling techniques preserve the clustering structure of graphs. Our empirical results show that the proposed sampling algorithms can preserve the population's clustering structure well and provide feasible solutions to sample the clustering structure from large-scale graphs.

关键词： graph sampling clustering structure top-leader nodes expansion strategies large-scale graphs

来源：评论

学校读者我要写书评

暂无评论

Research on Security Data of Federated Learning Privacy LSTM Model Based on Generative Adversarial network

Research on Security Data of Federated Learning Privacy LSTM...

引用

2023 International Conference on Machine Learning and Cybernetics, ICMLC 2023

作者： Xu, Wang Zhuoming, Liang School of Computer Neusoft Institute Guangdong Department of Network Engineering China South China Nomal University Network Center China

ISBN: (纸本)9798350303780

As a new wireless transmission architecture technology, cSO can effectively reduce the influence of fading channels and provide cSO links for low channel quality. Secondly, in the ultra-large traffic cSO, The fitting of data and the constraint on the model must be considered at the same time. Generally speaking, regularization is the most common method, because in the process of inversion, it is far from enough to only use data fitting terms as constraints. In order to make the inversion results stable, model constraint terms should be added to the objective function on the basis of data fitting terms, and there are usually different model constraint methods for different practical problems. When the physical properties of anomalous body are different from those of surrounding rock, the minimum model constraint can be used. When inversion of the moderocess of EMM-connected state was consistent with the determination of the coverage enhancement level. © 2023 IEEE.

关键词： Fading channels

来源：评论

学校读者我要写书评

暂无评论

Neural network-Based State Estimation for Nonlinear Systems With Denial-of-Service Attack Under Try-Once-Discard Protocol

引用

IEEE/CAA Journal of Automatica Sinica 2024年第10期11卷 2182-2184页

作者： Xueli Wang Shangwei Zhao Ming Yang Xin Wang Xiaoming Wu the Key Laboratory of Computing Power Network and Information Security Ministry of EducationShandong Computer Science CenterQilu University of Technology(Shandong Academy of Sciences)Jinan 250014 Shandong Provincial Key Laboratory of Computer Networks Shandong Fundamental Research Center for Computer ScienceJinan 250014China Department of Automation Shanghai Jiao Tong UniversityShanghai 200240China

Dear Editor,This letter deals with state estimation issues of discrete-time nonlinear systems subject to denial-of-service(DoS)attacks under the try-once-discard(TOD)*** specifically,to reduce the communication burden,a TOD protocol with novel update rules on protocol weights is designed for scheduling measurement *** addition,unknown nonlinear functions vulnerable to DoS attacks are considered due to the openness and vulnerability of the network.

关键词： Service DoS service

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：