检索结果-内蒙古大学图书馆

RaSA: Robust and Adaptive Secure Aggregation for Edge-Assisted Hierarchical Federated Learning

IEEE Transactions on Information Forensics and Security 2025年 20卷 4280-4295页

作者： Wang, Lingling Huang, Mei Zhang, Zhengyin Li, Meng Wang, Jingjing Gai, Keke Qingdao University of Science and Technology School of Information Science and Technology Shandong Qingdao266061 China Hefei University of Technology School of Computer Science and Information Engineering Anhui Hefei230601 China Beijing Institute of Technology School of Cyberspace Science and Technology Beijing100081 China

Secure Aggregation (SA), in the Federated Learning (FL) setting, enables distributed clients to collaboratively learn a shared global model while keeping their raw data and local gradients private. However, when SA is implemented in edge-intelligence-driven FL, the open and heterogeneous environments will hinder model aggregation, slow down model convergence speed, and decrease model generalization ability. To address these issues, we present a Robust and adaptive Secure Aggregation (RaSA) protocol to guarantee robustness and privacy in the presence of non-IID data, heterogeneous system, and malicious edge servers. Specifically, we first design an adaptive weights updating strategy to address the non-IID data issue by considering the impact of both gradient similarity and gradient diversity on the model aggregation. Meanwhile, we enhance privacy protection by preventing privacy leakage from both gradients and aggregation weights. Different from previous work, we address system heterogeneity in the case of malicious attacks, and the malicious behavior from edge servers can be detected by the proposed verifiable approach. Moreover, we eliminate the influence of straggling communication links and dropouts on the model convergence by combining efficient product-coded computing with repetition-based secret sharing. Finally, we perform a theoretical analysis that proves the security of RaSA. Extensive experimental results show that RaSA can ensure model convergence without affecting the generalization ability under non-IID scenarios. Moreover, the decoding efficiency of RaSA achieves 1.33x and 6.4x faster than the state-of-the-art product-coded and one-dimensional coded computing schemes. © 2005-2012 IEEE.

关键词： Differential privacy

来源：评论

学校读者我要写书评

暂无评论

Composite spectral spatial pixel CNN for land-use hyperspectral image classification with hybrid activation function

引用

Multimedia Tools and Applications 2025年第12期84卷 10527-10550页

作者： Banerjee, Anasua Swain, Satyajit Rout, Minakhi Bandyopadhyay, Mainak School of Computer Engineering Kalinga Institute of Industrial Technology-Deemed-to-be-University Bhubaneswar India Department of Computer Science and Engineering National Institute of Technology Silchar India

Deep learning methods have played a prominent role in the development of computer visualization in recent years. Hyperspectral imaging (HSI) is a popular analytical technique based on spectroscopy and visible imaging which examines the pattern of light in a target and recognizes objects based on varying spectral properties. However, in remote sensing, detecting surface material via HSI analysis is a critical and difficult task. The performance of spectral-spatial data exploitation is well established to outperform typical spectral pixel-wise techniques. Because of its great feature extraction ability, convolutional neural networks (CNN) have emerged as a potent deep learning approach. CNN translates the input features of an image into an equivalent CNN feature map, in addition to naturally combining spectral and spatial information. However, spectral-spatial properties when combined with pixel-wise extraction can learn more minute details of objects present on the earth’s terrain. In this paper, a noble Composite Spectral Spatial Pixel CNN model for the classification of hyperspectral data is presented which is an amalgamation of 3D-2D-1D CNN. While the 3D and 2D CNN exploit the spectral-spatial features effectively, 1D CNN works on pixel-wise feature extraction. Further, to optimize the classification performance of the proposed model, a new hybrid activation function Flatten-T Swish is also used which is the combination of ReLU and Swish function. The proposed model is compared with other state-of-the-art models based on three popular HSI datasets, and it is found that the proposed model performs better among others in terms of classification and computation time, giving 98.87% accuracy for Indian Pines, 99.92% accuracy for Pavia University, and 99.99% accuracy for Salinas Valley dataset. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Extraction

来源：评论

学校读者我要写书评

暂无评论

Multilabel Black-Box Adversarial Attacks Only With Predicted Labels

IEEE Transactions on Artificial Intelligence

引用

IEEE Transactions on Artificial Intelligence 2025年第5期6卷 1284-1297页

作者： Kong, Linghao Luo, Wenjian Ye, Zipeng Zhou, Qi Jia, Yan Harbin Institute of Technology Guangdong Provincial Key Laboratory of Novel Security Intelligence Technologies School of Computer Science and Technology Shenzhen518055 China Peng Cheng Laboratory Shenzhen518055 China

Multilabel adversarial examples have become a threat to deep neural network models (DNNs). Most of the current work on multilabel adversarial examples are focused on white-box environments. In this article, we focus on a black-box environment where the available information is extremely limited: a label-only black-box environment. Under the label-only black-box environment, the attacker can only obtain the predicted labels, and cannot obtain any other information such as the model's internal structure, parameters, the training dataset, and the output prediction confidence. We propose a label-only black-box attack framework, and through this framework to implement two black-box adversarial attacks: multi-label boundary-based attack (ML-BA) and multilabel label-only black-box attack (ML-LBA). The ML-BA is developed by transplanting the boundary-based attack in the multiclass domain to the multilabel domain, and the ML-LBA is based on differential evolution. Experimental results show that both the proposed algorithms can achieve the hiding single label attack in label-only black-box environments. Besides, ML-LBA requires fewer queries and its perturbations are significantly less. This demonstrates the effectiveness of the proposed label-only black-box attack framework and the advantageous of differential evolution in optimizing high-dimensional problems. © 2020 IEEE.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

Comprehensive analysis of various imputation and forecasting models for predicting PM2.5 pollutant in Delhi

引用

Neural Computing and Applications 2025年 1-18页

作者： Karnati, Hemanth Soma, Anuraag Alam, Adnan Kalaavathi, B. School of Computer Science and Engineering Vellore Institute of Technology Tamil Nadu Vellore632014 India Department of IoT School of Computer Science and Engineering Vellore Institute of Technology Tamil Nadu Vellore632014 India

Understanding and predicting air quality is pivotal for public health and environmental management, especially in urban areas like Delhi. This study utilizes a comprehensive dataset from the Central Pollution Control Board, detailing air pollutant concentrations at the ITO (Income Tax Office), Delhi station, from 2017 to 2023. Our focus on pollutants such as PM2.5, PM10, NO2, NH3, SO2, ozone, and CO highlights the challenges posed by missing data in environmental studies. Through the application of several imputation models, including KNN, linear regression, forward fill + backward fill, Fourier + KNN, linear interpolation, and statistical methods, we identified combination of forward fill and backward fill as the most effective method for addressing data gaps, specifically for ozone measurements. Building on this imputed dataset, we employ various forecasting models to predict PM2.5 levels, a critical pollutant. Our exploration includes time series and deep learning approaches, such as LSTM, Bi-LSTM, CNN-LSTM, GRU, Deep AR, WaveNet, TCN, ARIMA, SARIMA, and Prophet. The performance of these models is evaluated using daily data, with Bi-LSTM and LSTM with attention emerging as the top performers due to their accuracy in predicting PM2.5 concentrations. The implications of our findings extend beyond academic interest, offering practical insights for environmental policy and health advisories in Delhi. By demonstrating the efficacy of specific imputation and forecasting techniques, our research contributes to the broader effort of improving air quality monitoring and prediction. The success of Bi-LSTM and LSTM with attention models, in particular, suggests a promising avenue for future studies aiming to enhance the precision of environmental forecasts. © The Author(s) 2025.

关键词： Time series

来源：评论

学校读者我要写书评

暂无评论

14-Bit SAR ADC with on-Chip Digital Bubble Sorting Calibration technology

引用

Chinese Journal of Electronics 2025年第1期34卷 125-136页

作者： Hua Fan Zhuorui Chen Tongrui Xu Franco Maloberti Qi Wei Quanyuan Feng School of Integrated Circuit Science and Engineering(Exemplary School of Microelectronics) University of Electronic Science and Technology of China Shenzhen Institute for Advanced Study University of Electronic Science and Technology of China Chongqing Institute of Microelectronics Industry Technology University of Electronic Science and Technology of China Department of Electrical Computer and Biomedical Engineering University of Pavia Department of Precision Instrument Tsinghua University Institute of Microelectronics Southwest Jiaotong University

This article designs a 14-bit successive approximation register analog-to-digital converter（SAR ADC）.A novel digital bubble sorting calibration method is proposed and applied to eliminate the effect of capacitor mismatch on the linearity of the SAR ADC. To reduce the number of capacitors, a hybrid architecture of a high 8-bit binary-weighted capacitor array and a low 6-bit resistor array is adopted by the digital-to-analog（DAC）. The common-mode voltage VCM-based switching scheme is chosen to reduce the switching energy and area of the DAC. The time-domain comparator is employed to obtain lower power consumption. Sampling is performed through a gate voltage bootstrapped switch to reduce the nonlinear errors introduced when sampling the input signal. Moreover, the SAR logic and the whole calibration is totally implemented on-chip through digital integrated circuit（IC） tools such as design compiler, IC compiler, etc. Finally, a prototype is designed and implemented using 0.18 μm bipolar-complementary metal oxide semiconductor（CMOS）-double-diffused MOS 1.8 V CMOS technology. The measurement results show that the SAR ADC with on-chip bubble sorting calibration method achieves the signal-to-noise-and-distortion ratio of 69.75 dB and the spurious-free dynamic range of 83.77 dB.

关键词： Semiconductor device measurement Capacitors Linearity Switches Calibration System-on-chip Frequency measurement Time-domain analysis Sorting Signal to noise ratio

来源：评论

学校读者我要写书评

暂无评论

ZVEFusion: Zero-Shot Visual Enhancement Fusion for Infrared and Visible Images in Low Light

ZVEFusion: Zero-Shot Visual Enhancement Fusion for Infrared ...

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Liu, Duo Shi, Yiqi Zhang, Guoyin Li, Sizhao Zhang, Liguo School of Computer Science and Technology Harbin Engineering University China

ISBN: (纸本)9798350368741

Infrared and visible image fusion (IVIF) aims to generate fused images with prominent targets and rich scene information. However, in low-light conditions, visible images lose accurate texture and color, reducing their ability to provide detailed scene information for fusion. Existing IVIF methods often overlook illumination degradation and cause color distortion when incorporating infrared information. To address these problems, we propose a novel visually enhanced IVIF method tailored for low-light environments. Our method combines low-light image enhancement (LLIE) and IVIF into a single module. First, we adaptively enhance the low-light visible image, ensuring rich texture and color for fusion. Additionally, we introduce a three-channel fusion coefficient map to transform infrared information into visible image, preventing color distortion and highlighting key targets while maintaining details of the fused image. Since infrared and visible images are from different modalities, we map them into the same high-dimensional feature space. We then propose the feature difference to integrate complementary information, producing a fused image with complete content and no redundancy. Notably, our method is zero-shot, requiring only a pair of test infrared and visible images for training. This better meets the complexity of IVIF in various low-light scenes. Extensive experiments show that in low-light conditions, our method surpasses other state-of-the-art (SOTA) methods by providing more natural colors, richer textures, and better alignment with human visual perception. © 2025 IEEE.

关键词： Image Fusion Low Light Visually Enhanced Zero-Shot

来源：评论

学校读者我要写书评

暂无评论

Evaluation of Operating Comfort of Biological Microscopes 26th

Evaluation of Operating Comfort of Biological Microscopes

引用

26th International Conference on Human-computer Interaction, HCII 2024

作者： Wang, Yong Zhang, Tianmei Tan, Jiubin School of Architecture and Design Harbin Institute of Technology Harbin150001 China School of Instrumentation Science and Engineering Harbin Institute of Technology Harbin150001 China

ISBN: (纸本)9783031768088

The operating comfort of biological microscopes is directly related to the operators’ health and the accuracy of observation results. In this paper, the neck and arm comfort of biological microscope operators was researched. Firstly, the evaluation experiments included two parts: one was the evaluation of operating comfort for the neck and shoulder, the other was for the arm. The surface electromyography (sEMG) data of the measured muscles such as the splenius capitis, sternocleidomastoid, trapezius, flexor carpi and brachioradialis were acquired within the range of 0–40° neck angle and 90–150° elbow angle. In addition, the rating of perceived exertion (RPE) of operators was taken as the subjective evaluation results for different operating postures. Secondly, the root mean square (RMS) data of different subjects’ sEMG were normalized to the corresponding RMS of 20%MVC. It can be seen that the load of the splenius capitis, sternocleidomastoid and trapezius all increased with the increase of the head bow angle, and the load of measured muscles reached the maximum in 35–45°. And when the angle between the forearm and the upper arm was 115–125°, the load on the flexor carpi radialis, brachioradialis and trapezius was minimal. Finally, the ANOVA statistical analysis was proceeded to the RMS data of the measured muscles. It was found that the load of the splenius capitis and sternocleidomastoid was positively correlated with the bow angle, and the load range was 4%–30%MVC. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Muscle

来源：评论

学校读者我要写书评

暂无评论

A Novel Lightweight IoT Intrusion Detection Model Based on Self-knowledge Distillation

引用

IEEE Internet of Things Journal 2025年第11期12卷 16912-16930页

作者： Wang, Zhendong Zhou, Renqiang Yang, Shuxin He, Daojing Chan, Sammy Jiangxi University of Science and Technology School of Information Engineering Ganzhou341000 China Nanchang University Software Engineering Institute Nanchang330047 China Harbin Institute of Technology School of Computer Science and Technology Shenzhen518055 China City University of Hong Kong Department of Electrical Engineering Hong Kong

The Internet of Things (IoT) environment contains many different types of devices, each with different functionalities, communication protocols, and security capabilities, which makes the IoT a complex challenge for security protection. Therefore, network intrusion detection (NID) is needed to detect intrusions in the network to secure the IoT. In recent years, deep learning-based intrusion detection systems have achieved excellent results, but they tend to require high computational resources and storage space, which is not feasible for most IoT devices. In this paper, we propose a lightweight intrusion detection model based on self-knowledge distillation, namely TBCLNN, which improves the detection accuracy while also reducing the number of model parameters and computational cost. Specifically, we use the binary Harris Hawk optimization algorithm for dimensionality reduction of traffic features. We use lightweight convolution, such as Tied Block Convolution (TBC), to design lightweight neural network models with residual and inverse residual structures. Moreover, we propose an improved self-knowledge distillation loss function to solve the sample imbalance problem and compensate for the performance degradation caused by lightweight neural networks. The multiclassification accuracy of our proposed method exceeds 99% on all three publicly available IoT datasets. The experimental results show that our method has a small model size and requires only low computational resources, making it suitable for resource-constrained IoT intrusion detection. © 2014 IEEE.

关键词： Network intrusion

来源：评论

学校读者我要写书评

暂无评论

Revolutionizing signature scheme: the enhanced Edward Elgamal extreme performance accumulate signature approach for IoT and blockchain applications

引用

Soft Computing 2025年第3期29卷 1473-1496页

作者： Anusha, R. Saravanan, R. School of Computer Science and Engineering Vellore Institute of Technology Tamilnadu Vellore632014 India

Digital signatures, essential for establishing trust in the digital realm, have evolved in their application and importance alongside emerging technologies such as the Internet of Things (IoT), Blockchain, and cryptocurrency. These advancements necessitate improvements in performance, security, and efficiency. This article examines and compares the Elliptic Curve Digital Signature Algorithm with the Hyper Elliptic Curve Digital Signature Algorithm and the Edwards Curve Digital Signature Algorithm. We highlight its superior capabilities for blockchain and IoT applications and advocate for its potential to deliver immediate enhancements in security and performance. Our study introduces a novel digital signature scheme specifically designed to enhance non-repudiation in blockchain ecosystems. Utilizing the Optimized Extreme Performance Edwards Curve Accumulated Signature scheme, our approach significantly reduces signing and verification times by 10% and 13%, respectively, compared to traditional signatures. Additionally, it offers a 10% boost in transaction throughput and block validation efficiency. Experiments conducted within various blockchain-integrated IoT setups demonstrate the scheme's effectiveness, consistently achieving improvements across diverse IoT sensor data. This highlights the innovative contribution of our scheme to the efficiency and security of blockchain technology. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2025.

关键词： Blockchain Edward curve Elliptic curve Hyperelliptic curve IoT Optimization

来源：评论

学校读者我要写书评

暂无评论

LLDif: Diffusion Models for Low-Light Facial Expression Recognition 27th

LLDif: Diffusion Models for Low-Light Facial Expression Rec...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Wang, Zhifeng Zhang, Kaihao Sankaranarayana, Ramesh College of Engineering and Computer Science Australian National University CanberraACT Australia School of Computer Science and Technology Harbin Institute of Technology Shenzhen China

ISBN: (纸本)9783031782008

This paper introduces LLDif, a novel diffusion-based facial expression recognition (FER) framework tailored for extremely low-light (LL) environments. Images captured under such conditions often suffer from low brightness and significantly reduced contrast, presenting challenges to conventional methods. These challenges include poor image quality that can significantly reduce the accuracy of emotion recognition. LLDif addresses these issues with a novel two-stage training process that combines a Label-aware CLIP (LA-CLIP), an embedding prior network (PNET), and a transformer-based network adept at handling the noise of low-light images. The first stage involves LA-CLIP generating a joint embedding prior distribution (EPD) to guide the LLformer in label recovery. In the second stage, the diffusion model (DM) refines the EPD inference, ultilising the compactness of EPD for precise predictions. Experimental evaluations on various LL-FER datasets have shown that LLDif achieves competitive performance, underscoring its potential to enhance FER applications in challenging lighting conditions. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Emotion Recognition

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：