检索结果-内蒙古大学图书馆

An active learning framework for adversarial training of deep neural networks

Neural Computing and Applications 2025年第9期37卷 6849-6876页

作者： Ghosh, Susmita Chatterjee, Abhiroop Fiondella, Lance Department of Computer Science and Engineering Jadavpur University Kolkata India Department of Electrical and Computer Engineering University of Massachusetts Dartmouth United States

This article introduces a novel approach to bolster the robustness of Deep Neural Network (DNN) models against adversarial attacks named "Targeted Adversarial Resilience Learning (TARL)". The initial evaluation of a baseline DNN model reveals a significant accuracy decline when subjected to adversarial examples generated through techniques like FGSM, PGD, Carlini Wagner, and DeepFool attacks. To address this vulnerability, the article proposes an active learning framework, wherein the model iteratively identifies and learns from the most uncertain and misclassified instances. The key components of this approach include uncertainty estimation score in predicting the class of the input sample, selecting challenging samples based on this uncertainty score, labeling these challenging examples and augmenting them into the training set, and thereafter retraining the model with the expanded training set. The iterative active learning process, governed by parameters such as the number of iterations and batch size, demonstrates the potential to systematically enhance the resilience of DNN against adversarial threats. The proposed methodology has been investigated on several popular datasets such as the SARS-CoV-2 CT scan, MNIST, CIFAR-10, and Caltech-101, and demonstrated to be effective. Experiments illustrate that the learning framework improves the adversarial accuracies from 17.4% to 98.71% for the SARS-CoV-2 dataset, from 8.4% to 99.89% for the MNIST dataset, 1.6% to 78.84% for the CIFAR-10, and 12% to 92.92% for Caltech-101. Further, comparative analysis with several state-of-the-art methods suggests that the proposed framework offers superior defense against various attack methods and offers promising defensive mechanisms to deep neural networks. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2025.

关键词： Active learning

来源：评论

学校读者我要写书评

暂无评论

Impact of Different Integration Times on Distributions of Rain Rates for Predictions of Rain Attenuation

引用

Progress in Electromagnetics Research C 2025年 154卷 229-238页

作者： Hassan, Mohammad Rofiqul Rafiqul, Islam Md. Habaebi, Mohamed Hadi Suriza, Ahmad Zabidi Badron, Khairayu Budalal, Asma Ali Mahfuz, Md. Mahmudul Hasan Department of Electrical and Computer Engineering Faculty of Engineering International Islamic University Malaysia Jalan Gombak Kuala Lumpur53100 Malaysia College of Electrical and Electronics Technology Benghazi Libya Department of Electrical and Computer Engineering Concordia University Canada

All wireless communication systems are moving towards higher and higher frequencies day by day which are severely attenuated by rains in outdoor environment. To design a reliable RF system, an accurate prediction method of rain attenuation is established and used globally based on local rain intensity measurement. Required rain intensity used for attenuation prediction is generally measured at a point with 1-min integration time or converted from higher integration time to 1-min. Recent measurements of rain intensity with a 10-second integration time indicate that intensity is not uniform over a 1-minute duration. Consequently, the statistics of rain intensity distribution and attenuation predictions are influenced by measurements with integration times shorter than 1 minute. It has been established that an integration time of 0.01% provides the optimal fit for actual rain rate data. This paper presents the rain intensity distributions from data measured with 2-min, 1-min, 30-sec, 20-sec, and 10-sec integration times, and it has impact on rain rate distributions as well as rain attenuation predictions. © 2025, Electromagnetics Academy. All rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Stability Enhancement of Grid Connected Power System Using Type-II Neurofuzzy Wavelet Control

引用

International Journal of Fuzzy Systems 2025年 1-26页

作者： Dilshad, Saad Badar, Rabiah Haider, Aun Abas, Naeem Department of Electrical Engineering University of Management and Technology Lahore Sialkot Campus Sialkot Pakistan Department of Electrical and Computer Engineering COMSATS University Islamabad Islamabad Pakistan Department of Electrical and Computer Engineering International Islamic University Islamabad Islamabad Pakistan Department of Electrical Engineering University of Gujrat Hafiz Hayat Campus Gujrat Pakistan

Conventional methods like Automatic Voltage Regulators (AVRs) and Power System Stabilizers (PSSs) fall short of effectively damping Low-Frequency Oscillations (LFOs), necessitating the exploration of advanced solutions. This paper addresses the challenge of power system stability by utilizing advanced soft computing techniques in Supplementary Damping Control (SDC) for Static Synchronous Compensator (STATCOM) to mitigate low frequency oscillations in multi-machine power system. The proposed Type-II NeuroFuzzy Wavelet Controls (NFWC) employ Gaussian Type-II membership functions, with uncertainties in mean and standard deviation parameters, in the antecedent part and Wavelet Neural Networks (WNNs) in the consequent part. A multi-machine power system with STATCOM is considered for the performance evaluation of the proposed controllers. The proposed Type-II NFWC-1 and Type-II NFWC-2 outperform Adaptive NeuroFuzzy TSK Control (ANFTSKC) in terms of convergence speed and damping performance. Type-II NFWC-2 exhibits and maintains its superior performance over Type-II NFWC-1 and ANFTSK, highlighting the effectiveness of incorporating additional uncertainty in both mean and standard deviation parameters. © The Author(s) under exclusive licence to Taiwan Fuzzy Systems Association 2025.

关键词： Frequency stability

来源：评论

学校读者我要写书评

暂无评论

Analysis of the effect of different scattering mechanisms on the excess noise behavior of Sb-based avalanche photodiodes

引用

Optics Express 2025年第4期33卷 7337-7344页

作者： Karimi, Hannaneh Herrera, Daniel J. Dadey, Adam A. Wei, Dongxia Mcarthur, Andrew Bank, Seth R. Campbell, Joe. C. Department of Electrical and Computer Engineering University of Virginia CharlottesvilleVA22903 United States Department of Electrical and Computer Engineering University of Texas AustinTX78758 United States

In optical applications where avalanche photodiodes (APDs) provide the benefit of high sensitivity, Sb-based materials systems such as AlInAsSb and AlGaAsSb have shown extremely low excess noise factors. The Monte Carlo (MC) model described in this work reveals the effect of high alloy scattering on the excess noise factor in these material systems. A comparison between the excess noise factor of different combinations of scattering rates is investigated. Furthermore, the ionization path length probability distribution is presented for AlInAsSb APDs with different scattering mechanisms. Good agreement is achieved between simulation results and experimental measurements. © 2025 Optica Publishing Group under the terms of the Optica Open Access Publishing Agreement.

关键词： III-V semiconductors

来源：评论

学校读者我要写书评

暂无评论

A High Power Density Active-Clamp Forward-Flyback Converter with Three Planar Transformer for xEV Low Voltage DC-DC Converter

引用

Transactions of the Korean Institute of electrical Engineers 2025年第3期74卷 468-476页

作者： Kwon, Gyeonghyun Lee, Dongin Jeong, Seongwook Jeong, Yeonho Youn, Hanshin Dept. of Electrical Engineering Incheon National University Korea Republic of Dept. of Electrical Computer and Biomedical Engineering University of Rhode Island KingstonRI United States

This paper proposes an Active-Clamp Forward-Flyback (ACFF) converter with three planar transformers to achieve 4 kW output power. As autonomous driving technology rapidly advances, the demand for higher output power from low-voltage DC-DC converters (LDC) in electric vehicles has significantly increased. Conventionally, two 2 kW ACFF converters with four magnetic cores connected in series-parallel have been used to meet 4 kW power requirement, but this configuration has proven inefficient in terms of size, power density, and cost. The proposed converter improves power density by integrating the four magnetic cores of the conventional 4 kW ACFF converter into three, reducing the overall magnetic volume and allowing for a more compact design. Furthermore, the use of planar magnetic components enables the winding to be embedded in a 4-layer PCB. Copyright © The Korean Institute of electrical Engineers.

关键词： High frequency transformers

来源：评论

学校读者我要写书评

暂无评论

Binary Error-Correcting Codes With Minimal Noiseless Feedback

引用

IEEE Transactions on Information Theory 2025年第5期71卷 3424-3446页

作者： Gupta, Meghal Guruswami, Venkatesan Yun Zhang, Rachel Department of Electrical and Computer Engineering BerkeleyCA94720 United States Massachusetts Institute of Technology Department of Electrical Engineering and Computer Science CambridgeMA02139 United States

In the setting of error-correcting codes with feedback, Alice wishes to communicate a k-bit message x to Bob by sending a sequence of bits over a channel while noiselessly receiving feedback from Bob. It has been long known (Berlekamp, 1964) that in this model, Bob can still correctly determine x even if ≈ 1/3 of Alice’s bits are flipped adversarially. This improves upon the classical setting without feedback, where recovery is not possible for error fractions exceeding 1/4 . In the corresponding setting of erasures rather than bit flips, feedback improves the error resilience from 1/2 -Ε to 1 - Ε for any Ε > 0 . The original feedback setting assumes that after transmitting each bit, Alice knows (via feedback) what bit Bob received. In this work, our focus in on the limited feedback model, where Bob is only allowed to send a few bits at a small number of pre-designated points in the protocol. For any desired Ε > 0 , we construct a coding scheme that tolerates a 1/3 -Ε fraction of bit flips (respectively a 1 −Ε fraction of erasures) relying only on OΕ (log k) bits of feedback from Bob sent in a fixed OΕ (1) number of rounds. We complement this with a matching lower bound showing that Ω (log k) bits of feedback are necessary to recover from an error fraction exceeding 1/4 (respectively 1/2 for erasures), and for schemes resilient to a 1/3 -Ε fraction of bit flips (respectively a 1 −Ε fraction of erasures), the number of rounds must grow as Ε → 0 © 1963-2012 IEEE.

关键词： Binary codes

来源：评论

学校读者我要写书评

暂无评论

Multi-stage Segmentation of Diabetic Foot Ulcers Using Self-supervised Learning 4th

Multi-stage Segmentation of Diabetic Foot Ulcers Using Self...

引用

4th Diabetic Foot Ulcers Grand Challenge, DFUC 2024 held in Conjunction 27th International Conference on Medical Image Computing and computer Assisted Intervention, MICCAI 2024

作者： Gupta, Neetika Nangia, Ayush Khan, Naimul Mefraz Department of Electrical Computer and Biomedical Engineering Toronto Metropolitan University TorontoON Canada

ISBN: (纸本)9783031808708

Diabetic Foot Ulcers (DFUs) pose a significant healthcare challenge due to their potential to lead to severe complications, including infections and ischemia, often necessitating limb amputation or causing mortality. Accurate diagnosis of ulcer regions, and prompt response can prevent amputations and fatalities through early and reliable DFU detection. However, manual examination of DFUs is labor-intensive and prone to variability. In response to these challenges, we propose DFUC_CAM, a novel method for instance segmentation of DFUs using self-supervised learning (SSL). The proposed approach leverages a multi-stage methodology. Initially, a convolutional neural network (CNN) trained for classification is utilized to generate Class Activation Maps (CAMs), which localize and highlight discriminative regions within DFU images. These CAMs serve as pseudo-labels to train a segmentation network using a SSL framework. For the MICCAI 2024 Diabetic Foot Ulcer Challenge (DFUC 2024), we trained DFUC_CAM using the DFUC 2024 and FUSeg datasets. The FUSeg dataset primarily serves for the quantitative evaluation of our method. During the validation phase of DFUC 2024, our team, ***, achieved a mean Dice score of 0.342, and in the final testing phase, we improved this to 0.385. These results demonstrate the potential of our approach for enhancing DFU detection and segmentation in clinical settings. The code is available at: https://***/RyersonMultimediaLab/DFUC_CAM. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Self-supervised learning

来源：评论

学校读者我要写书评

暂无评论

Maximally Flat Negative Group Delay Prototype Filter Based on Capped Reciprocal Transfer Function of Classical Bessel Filter

引用

Progress in Electromagnetics Research B 2025年 110卷 91-105页

作者： Kandic, Miodrag Bridges, Greg E. Department of Electrical and Computer Engineering University of Manitoba WinnipegMB Canada

A prototype filter design exhibiting Negative Group Delay (NGD) is presented, based on the ratio of two low-pass classical Bessel filter transfer functions of the same order, but with different 3 dB-bandwidths. The resulting design is a reciprocal-Bessel filter transfer function, capped at a finite out-of-band gain. The proposed capped reciprocal-Bessel design is based on a similar concept applied to previously reported capped reciprocal-Butterworth and reciprocal-Chebyshev NGD designs, which use ratios of corresponding classical low-pass filter transfer functions. It is shown that within the in-band frequency range, the synthesized NGD transfer function exhibits a maximally flat group delay characteristic (Bessel-like property). Due to its near-flat in-band group delay characteristic, the design is suitable for constant phase shifter applications. For high design orders, it is shown that the achieved NGD-bandwidth product has an upper asymptotic limit, given by the square root of the out-of-band gain in decibels. When the prototype baseband transfer function is translated to a non-zero center frequency, it is demonstrated that resonator-based implementations are feasible via Sallen-Key, as well as all-passive ladder topologies. A combined in-band magnitude/phase distortion metric is evaluated for selected design examples and applied Gaussian and sinc input waveforms, and it is shown to be proportional to the design order and out-of-band gain. The proposed design's distortion metric is also shown to be generally lower than the previously reported capped reciprocal-Butterworth and reciprocal-Chebyshev designs. © (2025), (Electromagnetics Academy). All rights reserved.

关键词： Group delay

来源：评论

学校读者我要写书评

暂无评论

Deep Learning-Enhanced Ultrasound Analysis: Classifying Breast Tumors Using Segmentation and Feature Extraction

引用

IEEE Access 2025年 13卷 83528-83541页

作者： Hamza, Ali Mezl, Martin Brno University of Technology Faculty of Electrical Engineering and Communication Department of Biomedical Engineering Brno616 00 Czech Republic

Breast cancer remains a significant global health challenge, requiring accurate and effective diagnostic methods for timely treatment. Ultrasound imaging is a valuable diagnostic tool for breast cancer because of its affordability, accessibility, and non-ionizing radiation properties. This study proposes a classification method for breast ultrasound images that integrates segmentation and feature extraction. Initially, ultrasound images are pre-processed to enhance quality and reduce noise, followed by segmentation using the U-Net++ architecture. Feature extraction is then performed using MobileNetV2, and these features are used to train and validate classification models to differentiate between malignant and benign breast masses. The model’s performance is assessed using accuracy, precision, recall, Mean IoU, and Dice Score metrics. The U-Net++ model achieved superior segmentation performance with a Dice Score of 0.911 and a Mean IoU of 0.838, outperforming related methods such as U-Net (0.888 Dice, 0.79 IoU) and Efficient U-Net (0.904 Dice, 0.80 IoU). In the classification task, MobileNetV2 when paired with the ANN classifier, produced the highest test accuracy at 96.58%, with a precision of 97% and recall of 96%. Our approach demonstrates superior performance compared to other models, such as RMTL-Net, which achieved 91.02% accuracy, and hybrid CAD models with 94% accuracy. This highlights the benefits of combining advanced segmentation and feature extraction techniques, with MobileNetV2 proving to be the better model, offering superior accuracy and robustness in classification tasks. This approach has the potential to support promise for supporting radiologists, enhance diagnostic accuracy, and ultimately improve outcomes for breast cancer patients. In the future, we will use comprehensive datasets to validate our methodology. © 2013 IEEE.

关键词： Image segmentation

来源：评论

学校读者我要写书评

暂无评论

Knowledge distillation on spatial-temporal graph convolutional network for traffic prediction

引用

International Journal of computers and Applications 2025年第1期47卷 45-56页

作者： Izadi, Mohammad Safayani, Mehran Mirzaei, Abdolreza Department of Electrical and Computer Engineering Isfahan University of Technology Isfahan Iran

Efficient real-time traffic prediction is crucial for reducing transportation time. To predict traffic conditions, we employ a spatio-temporal graph neural network (ST-GNN) to model our real-time traffic data as temporal graphs. Despite its capabilities, it often encounters challenges in delivering efficient real-time predictions for real-world traffic data. Recognizing the significance of timely prediction due to the dynamic nature of real-time data, we employ knowledge distillation (KD) as a solution to enhance the execution time of ST-GNNs for traffic prediction. In this paper, We introduce a cost function designed to train a network with fewer parameters (the student) using distilled data from a complex network (the teacher) while maintaining its accuracy close to that of the teacher. We use knowledge distillation, incorporating spatial-temporal correlations from the teacher network to enable the student to learn the complex patterns perceived by the teacher. However, a challenge arises in determining the student network architecture rather than considering it inadvertently. To address this challenge, we propose an algorithm that utilizes the cost function to calculate pruning scores, addressing small network architecture search issues, and jointly fine-tunes the network resulting from each pruning stage using KD. Ultimately, we evaluate our proposed ideas on two real-world datasets, PeMSD7 and PeMSD8. The results indicate that our method can maintain the student's accuracy close to that of the teacher, even with the retention of only (Formula presented.) of network parameters. © 2024 Informa UK Limited, trading as Taylor & Francis Group.

关键词： Cost functions

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：