检索结果-内蒙古大学图书馆

arXiv 2025年

作者： He, Yu Li, Boheng Liu, Liu Ba, Zhongjie Dong, Wei Li, Yiming Qin, Zhan Ren, Kui Chen, Chun The State Key Laboratory of Blockchain and Data Security Zhejiang University China College of Computing and Data Science Nanyang Technological University Singapore

Membership Inference Attacks (MIAs) aim to predict whether a data sample belongs to the model’s training set or not. Although prior research has extensively explored MIAs in Large Language Models (LLMs), they typically require accessing to complete output logits (i.e., logits-based attacks), which are usually not available in practice. In this paper, we study the vulnerability of pre-trained LLMs to MIAs in the label-only setting, where the adversary can only access generated tokens (text). We first reveal that existing label-only MIAs have minor effects in attacking pre-trained LLMs, although they are highly effective in inferring fine-tuning datasets used for personalized LLMs. We find that their failure stems from two main reasons, including better generalization and overly coarse perturbation. Specifically, due to the extensive pretraining corpora and exposing each sample only a few times, LLMs exhibit minimal robustness differences between members and non-members. This makes token-level perturbations too coarse to capture such differences. To alleviate these problems, we propose PETAL: a label-only membership inference attack based on PEr-Token semAntic simiLarity. Specifically, PETAL leverages token-level semantic similarity to approximate output probabilities and subsequently calculate the perplexity. It finally exposes membership based on the common assumption that members are ‘better’ memorized and have smaller perplexity. We conduct extensive experiments on the WikiMIA benchmark and the more challenging MIMIR benchmark. Empirically, our PETAL performs better than the extensions of existing label-only attacks against personalized LLMs and even on par with other advanced logit-based attacks across all metrics on five prevalent open-source LLMs1. Our study highlights that pre-trained LLMs remain vulnerable to privacy risks posed by MIAs even in the most challenging and realistic setting, calling for attention to develop more effective defenses. Copyright © 20

关键词： Semantics

来源：评论

学校读者我要写书评

暂无评论

ZnO-ITO/WO_(3-x)heterojunction structured memristor for optoelectronic co-modulation neuromorphic computation

引用

science China Materials 2024年第9期67卷 2838-2847页

作者： Jianyong Pan Tong Wu Wenhao Yang Yang Li Jiaqi Zhang Hao Kan Shandong Provincial Key Laboratory of Network Based Intelligent Computing School of Information Science and EngineeringUniversity of JinanJinan250022China School of Integrated Circuits Shandong UniversityJinan250101China Key Laboratory of Automobile Materials Ministry of Education College of Materials Science and EngineeringJilin UniversityChangchun130012China

Traditional transistors confront severe challenges of insufficient computing capability and excessive power consumption in large-scale neuromorphic *** address these critical bottlenecks,we propose an optoelectronic memristor based on zinc oxide-indium tin oxide/tungsten oxide(ZnO-ITO/WO_(3-x))heterojunctions as a promising *** applying different types of electrical and optical signals,the device successfully emulates diverse synaptic functions including short-term/long-term synaptic plasticity,alongside short-term and long-term *** the ZnO-ITO functional layer enhances the photoresponse of the WO_(3-x)-based memristor and demonstrates“learning-forgetting-relearning”behavior under optical ***,based on the photoelectric cooperative memristor array,a convolutional neural network for vehicle type recognition is constructed,which solves the problem of zero weight and negative weight *** regard to energy efficiency,the neural network built with this device operates at a power level of only 10^(-3)W,representing a reduction of more than 4 orders of magnitude compared with a standard central ***,the photoelectric memristor proposed in this work provides a new idea for neuromorphic computing and is expected to promote the development of energy-efficient brain-like computing.

关键词： ZnO-ITO/WO_(3-x)heterojunction optoelectronic memristor synaptic plasticity neuromorphic computing convolutional neural network

来源：评论

学校读者我要写书评

暂无评论

An improved branching algorithm for the proper interval edge deletion problem

引用

Frontiers of Computer science 2022年第2期16卷 178-180页

作者： Wenjun LI Xiaojing TANG Yongjie YANG School of Computer and Communication Engineering Hunan Provincial Key Laboratory of Intelligent Processing of Big Data on TransportationChangsha University of Science and TechnologyChangsha 410015China School of Computer Science and Engineering Central South UniversityChangsha 410083China

1 Introduction For a graph class G,the G-EDGE DELETION problem is to determine whether a given graph can be transformed into a graph in G by deleting at most k *** G-EDGE DELETION problem for a large body of graph cla... 详细信息

关键词： branching problem transformed

来源：评论

学校读者我要写书评

暂无评论

Analysis of Factors Influencing Driver Lane-Changing Intentions Based on a Naturalistic Trajectory data Set on Highways 24

Analysis of Factors Influencing Driver Lane-Changing Intenti...

引用

24th COTA International Conference of Transportation Professionals: Resilient, intelligent, Connected, and Lowcarbon Multimodal Transportation, CICTP 2024

作者： Wang, Guangchen Lu, Guangquan Wang, Jinghua Liu, Miaomiao School of Transportation Science and Engineering Beihang Univ. Beijing China Beijing Key Laboratory for Cooperative Vehicle Infrastructure Systems and Safety Control National Engineering Laboratory for Comprehensive Transportation Big Data Application Technology Beijing Advanced Innovation Center for Big Data and Brain Computing Beihang Univ. Beijing China

ISBN: (数字)9780784485484

ISBN: (纸本)9780784485484

Accurately analyzing and predicting driver lane-changing intentions is of paramount importance, as it significantly enhances the safety of self-driving vehicles in their decision-making processes, holding great promise for the future of intelligent transportation systems. This study extracts fundamental lane-changing characteristic parameters from the Next Generation Simulation (NGSIM) trajectory data set and employs Structural Equation Modeling (SEM) to construct a model that elucidates the factors influencing drivers’ lane-changing intentions. Furthermore, it provides insight into how lane-changing intentions are influenced by lane-changing behavior and explores the intricate connection between intention and associated risks. The results emphasize the substantial negative impact of the longitudinal speed difference between the self-vehicle and the preceding vehicle on lane-changing intentions. Additionally, the consideration of lane-changing risk plays a pivotal role in drivers’ decisions to engage in lane-changing behavior. These findings have significant implications for a deeper comprehension of lane-changing intentions, guiding lane-changing decisions, and enhancing overall traffic safety. © ASCE.

关键词： Autonomous vehicles

来源：评论

学校读者我要写书评

暂无评论

Identification of key Genes as Potential Drug Targets for Gastric Cancer

引用

Tsinghua science and Technology 2023年第4期28卷 649-664页

作者： Md.Tofazzal Hossain Md.Selim Reza Yin Peng Shengzhong Feng Yanjie Wei Center for High Performance Computing Joint Engineering Research Center for Health Big Data Intelligent Analysis TechnologyShenzhen Institute of Advanced TechnologyChinese Academy of SciencesShenzhen 518055China School of Computer Science and Technology University of Chinese Academy of SciencesBeijing 100049China Department of Statistics Bangabandhu Sheikh Mujibur Rahman Science and Technology UniversityGopalganj 8100Bangladesh Guangdong Provincial Key Laboratory for Genome Stability and Disease Prevention and Regional Immunity and Diseases Department of PathologyShenzhen UniversityShenzhen 518060China

Gastric cancer(GC)is one of the most common cancers and ranks the third in cancer mortality all over the *** goal of this study was to identify potential hub-genes,highlighting their functions,signaling pathways,and candidate drugs for the treatment of GC *** used publicly available next generation sequencing(NGS)data to identify differentially expressed(DE)*** top DE genes were mapped to STRING database to construct the protein-protein interaction(PPI)network and top hub genes were selected for further *** found a total of 1555 DE genes with 870 upregulated and 685 downregulated genes in *** selected the top 400(200 upregulated and 200 downregulated)genes to construct a PPI network and extracted the top 15 hub *** gene ontology(GO)term and kyoto encyclopedia of genes and genomes(KEGG)pathway enrichment analyses of the 15 hub genes exposed some important functions and signaling pathways that were significantly associated with GC *** survival analysis of the hub genes disclosed that the lower expressions of the three hub genes CDH2,COL4A1,and COL5A2 were associated with better survival of GC *** three genes might be the candidate biomarkers for the diagnosis and treatment of ***,we considered 3 key proteins(genomic biomarkers)(COL4A1,CDH2,and CO5A2)as the drug target proteins(receptors),performed their docking analysis with the 102 meta-drug agents,and found Everolimus,Docetaxel,Lanreotide,Venetoclax,Temsirolimus,and Nilotinib as the top ranked 6 candidate drugs with respect to our proposed target proteins for the treatment against GC ***,the proposed drugs might play vital role for the treatment against GC patients.

关键词： gastric cancer hub genes candidate genes molecular docking candidate drugs

来源：评论

学校读者我要写书评

暂无评论

Continuous-Time Prediction of Industrial Paste Thickener System With Differential ODE-Net

引用

IEEE/CAA Journal of Automatica Sinica 2022年第4期9卷 686-698页

作者： Zhaolin Yuan Xiaorui Li Di Wu Xiaojuan Ban Nai-Qi Wu Hong-Ning Dai Hao Wang the Beijing Advanced Innovation Center for Materials Genome Engineering Institute of Artificial IntelligenceBeijing Key Laboratory of Knowledge Engineering for Materials ScienceSchool of Computer and Communication EngineeringUniversity of Science and Technology Beijing the Department of ICT and Natural Science Norwegian University of Science and Technology IEEE the Institute of Systems Engineering and Collaborative Laboratory for Intelligent Science and SystemsMacao University of Science and Technology the Department of Computing and Decision Sciences Lingnan University the Department of Computer Science Norwegian University of Science and Technology

It is crucial to predict the outputs of a thickening system,including the underflow concentration(UC)and mud pressure,for optimal control of the *** proliferation of industrial sensors and the availability of thickening-system data make this ***,the unique properties of thickening systems,such as the non-linearities,long-time delays,partially observed data,and continuous time evolution pose challenges on building data-driven predictive *** address the above challenges,we establish an integrated,deep-learning,continuous time network structure that consists of a sequential encoder,a state decoder,and a derivative module to learn the deterministic state space model from thickening *** a case study,we examine our methods with a tailing thickener manufactured by the FLSmidth installed with massive sensors and obtain extensive experimental *** results demonstrate that the proposed continuous-time model with the sequential encoder achieves better prediction performances than the existing discrete-time models and reduces the negative effects from long time delays by extracting features from historical system *** proposed method also demonstrates outstanding performances for both short and long term prediction tasks with the two proposed derivative types.

关键词： Industrial 24 paste thickener ordinary differential equation(ODE)-net recurrent neural network time series prediction

来源：评论

学校读者我要写书评

暂无评论

High color saturation and angle-insensitive ultrathin color filter based on effective medium theory

引用

Chinese Optics Letters 2023年第3期21卷 146-151页

作者：李欣婷李阳李超高嵩岳文静 School of Information Science and Engineering University of JinanJinan 250022China Shandong Provincial Key Laboratory of Network Based Intelligent Computing University of JinanJinan 250022China

An ultrathin angle-insensitive color filter enabling high color saturation and a wide color gamut is proposed by relying on magnesium hydride-hydrogenated amorphous silicon[MgH2-a-Si:H]lossy dielectric *** on effective medium theory,the MgH2-a-Si:H layer with an ultrathin thickness can be equivalent to a quasi-homogeneous dielectric layer wit an effective complex refractive index,which can be tuned by altering the thickness of MgH2to obtain the targeted value o the imaginary part,corresponding to the realization of high color *** is verified that the proposed color filte offers highly enhanced color saturation in conjunction with a wide color gamut by introducing a few-nanometer thic *** the MgH2-a-Si:H layer retains the advantages of high refractive index and tiny thickness,the proposed colo filter exhibits large angular tolerance up to±60°.In addition,MgH2with an unstable property can interconvert with Mg unde a dehydrogenation/hydrogenation reaction,which empowers the proposed color filter with dynamically tunable outpu *** proposed scheme shows great promise in color printing and ultracompact display devices with high color sat uration,wide gamut,large angular tolerance,and dynamic tunability.

关键词： color filter effective medium theory lossy dielectric layer color saturation angle insensitivity

来源：评论

学校读者我要写书评

暂无评论

Contrastive Learning based Item Representation with Asymmetric Augmentation for Sequential Recommendation 23

Contrastive Learning based Item Representation with Asymmetr...

引用

2nd International Conference on Algorithms, data Mining, and Information Technology, ADMIT 2023

作者： Zeng, Kexiang Shi, Xiaoyu Xie, Hong Shang, Mingsheng Department of Computer Science and Technology Chongqing University of Posts and Telecommunications Chongqing China Chongqing Key Laboratory of Big Data and IntelligentComputing Chongqing Institute of Green and Intelligent Technology Chongqing China

ISBN: (纸本)9798400707629

Contrastive learning has been widely applied in sequential recommendation to improve the recommendation performance. Existing contrastive learning methods focus on adjusting the views number of positive and negative samples to enhance the item representation via data-level augmentation(e.g., MMInfoRec). However, they generally ignore the sampled items still follow a long-tail distribution, i.e., a few popular items have a high-frequency occurrence in contrastive learning, while the majority of unpopular items have a low-frequency in contrastive learning. As a result, this imbalance in sample extraction leads to insufficient contrastive learning between popular and unpopular items, resulting in sub-optimal item representation. In this paper, we propose CA4Rec, a Contrastive learning based item representation with Asymmetric augmentation for Sequential Recommendation, that adopts an asymmetric augmented multi-instance contrastive learning strategy to enhance the item representation. Specifically, we first use the popularity-aware method to divide the entire item set into popular and unpopular items, then we generate k augmented views for popular items, and k + n augmented views for unpopular items. Finally, we use our proposed asymmetric multi-instance noise contrastive estimation (AMINCE) to perform contrastive learning calculation. Make sure there is sufficient contrast between popular and unpopular items to mitigate the issue. we implement the improvements on the state-of-the-art model and conduct extensive experiments on three benchmark datasets. The experimental results demonstrate that CA4Rec outperforms the state-of-the-art baselines. © 2023 Owner/Author.

关键词： Learning systems

来源：评论

学校读者我要写书评

暂无评论

Shadow Removal from a Single Image Using an Enhanced Generative Adversarial Network 19

Shadow Removal from a Single Image Using an Enhanced Generat...

引用

19th International Conference on Natural Computation, Fuzzy Systems and Knowledge Discovery, ICNC-FSKD 2023

作者： Zhang, Jing Kong, Weiwei Xi'an University of Posts and Telecommunications College of Computer Science Xi'an China Xi'an University of Posts and Telecommunications The Shaanxi Key Laboratory of Network Data Analysis and Intelligent Processing Xi'an China

ISBN: (纸本)9798350304398

Image shadow removal is challenging due to the diversity of shadows and the dependence on the background of the area covered by shadows. In this paper, we design an enhanced generative adversarial network (EGAN) to handle the shadow removal task. EGAN is a complicated generative adversarial network that may exploit previous info, like shadow masks and lightness estimation, for improving recovery capability. In addition, EGAN can effectively utilize the details of shadow regions and eliminate artifacts. EGAN owns three main advantages. First, EGAN employs stricter rules for discriminators, which can upgrade the instability of traditional GAN. Second, EGAN involves a completely unique enhanced attention module, which may efficiently pick up different types of previous info to accelerate the learning efficiency of image reconstruction. Finally, EGAN is essentially an original cascaded network, that is, DRDU, which is beneficial for efficiently detecting hierarchical features of image recovery. Simulation experiments are conducted to verify the effectiveness of the proposed method on two benchmark datasets including SRD and ISTD, the strengths of our proposed method are indicated over the current representative ones according to subjective performance and objective estimation indexes. © 2023 IEEE.

关键词： Image reconstruction

来源：评论

学校读者我要写书评

暂无评论

Unleashing the Potential of Decentralized Federated Learning in Healthcare: A Comprehensive Survey on Insights and Challenges 9

Unleashing the Potential of Decentralized Federated Learning...

引用

9th IEEE International Conference on data science in Cyberspace, DSC 2024

作者： Cheng, Haibo Liu, Wenjian Qu, Youyang Fu, Kexue Wang, Qinyuan City University of Macau Faculty of Data Science China Qilu University of Technology Shandong Academy of Sciences Key Laboratory of Computing Power Network and Information Security Ministry of Education Shandong Computer Science Center National Supercomputer Center in Jinan Jinan China

ISBN: (纸本)9798350391367

Federated learning (FL) is a distributed learning method that reduces data transmission and privacy risks. However, the existence of a central server may become a target of attackers, resulting in an increased risk of user privacy leakage. The emergence of Decentralized Federated Learning (DFL) is posited to enhance data privacy and system reliability by reducing reliance on central servers and enhancing local data control, especially in the medical field, where personal health data is protected. This research investigates the application of (DFL) in the medical domain, focusing on its limitations. This study employs a systematic literature review, adhering to PRISMA guidelines, to comprehensively understand DFL's current landscape and challenges in healthcare. The key findings highlight four principal obstacles faced by DFL: security and privacy, communication efficiency, data and model heterogeneity, and incentive mechanisms. It also explores the potential for DFL to facilitate personalized healthcare treatments by large datasets distributed across various medical institutions. This research fills a significant void by systematically reviewing DFL technologies in healthcare. It provides insights into applications, challenges, and future directions for participants seeking secure and efficient healthcare data handling. © 2024 IEEE.

关键词： Federated 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：