检索结果-内蒙古大学图书馆

A Joint Secure Mechanism of Multi-task Learning For A UAV Team Under FDI Attacks

IEEE Transactions on Mobile Computing 2025年

作者： Zeng, Rongfei Jiang, Chenyang Wang, Xingwei Li, Baochun Northeastern University Software College Shenyang China AliCloud Hangzhou China Northeastern University Department of Computer Science and Engineering Shenyang China University of Toronto Department of Electrical and Computer Engineering Canada

A UAV team shows tremendous potential for various mobile scenarios. However, some evidences reveal their vulnerability to False Data Injection (FDI) attacks, which can significantly jeopardize the flight security or even lead to catastrophic incidents. Existing studies primarily focus on detecting or defending against FDI attacks at the trajectory control of individual UAVs, leaving a gap in a comprehensive secure mechanism that can simultaneously detect, localize, and compensate for such attacks across an entire UAV team. The complexity of developing such a solution is magnified by the multiple design goals, the inherent sophistication of UAV team, and practical attack assumptions. In this paper, we propose a joint secure framework based on multi-task deep learning to simultaneously detect FDI attacks, localize the compromised components, and compensate control signals to mitigate the impact of FDI attacks on promising UAV teams. Specifically, we design an all-in-one deep learning model framework with a temporal-spatial information extraction module and a hierarchical multi-task module to perform three tasks simultaneously. Moreover, we introduce an iterative learning method with experience replay to counteract knowledge decay during model training. Extensive experiments and real flight demonstrations are presented to validate the improved performance and the benefits of our proposed secure method. A demonstration video and our source code can be accessed via https://***/WingFeiTsang/JS/. © 2025 IEEE. All rights reserved.

关键词： Unmanned aerial vehicles (UAV)

来源：评论

学校读者我要写书评

暂无评论

FLEKE: Federated Locate-then-Edit Knowledge Editing

arXiv

引用

arXiv 2025年

作者： Zhao, Zongkai Xu, Guozeng Li, Xiuhua Wei, Kaiwen Zhong, Jiang School of Big Data & Software Engineering Chongqing University China College of Computer Science Chongqing University China

Locate-then-Edit Knowledge Editing (LEKE) is a key technique for updating large language models (LLMs) without full retraining. However, existing methods assume a single-user setting and become inefficient in real-world multi-client scenarios, where decentralized organizations (e.g., hospitals, financial institutions) independently update overlapping knowledge, leading to redundant mediator knowledge vector (MKV) computations and privacy concerns. To address these challenges, we introduce Federated Locate-then-Edit Knowledge Editing (FLEKE), a novel task that enables multiple clients to collaboratively perform LEKE while preserving privacy and reducing computational overhead. To achieve this, we propose FedEdit, a two-stage framework that optimizes MKV selection and reuse. In the first stage, clients locally apply LEKE and upload the computed MKVs. In the second stage, rather than relying solely on server-based MKV sharing, FLEKE allows clients retrieve relevant MKVs based on cosine similarity, enabling knowledge re-edit and minimizing redundant computations. Experimental results on two benchmark datasets demonstrate that FedEdit retains over 96% of the performance of non-federated LEKE while significantly outperforming a FedAvg-based baseline by approximately twofold. Besides, we find that MEMIT performs more consistently than PMET in the FLEKE task with our FedEdit framework. Our code is available at https://***/zongkaiz/FLEKE. Copyright © 2025, The Authors. All rights reserved.

关键词： Benchmarking

来源：评论

学校读者我要写书评

暂无评论

VGGBM-Net: A Novel Pixel-Based Transfer Features engineering for Automated Coffee Bean Diseases Classification

引用

IEEE Access 2025年 13卷 83580-83590页

作者： Hashmi, Muhammad Shadab Alam Qadri, Azam Mehmood Raza, Ali Ullah, Saleem Smerat, Aseel Kim, Changgyun Syafrudin, Muhammad Fitriyani, Norma Latif Khwaja Fareed University of Engineering and Information Technology Department of Computer Science Rahim Yar Khan Pakistan University of Lahore Department of Software Engineering Lahore 54000 Pakistan Al-Ahliyya Amman University Faculty of Educational Sciences Amman 19328 Jordan Chitkara University Institute of Engineering and Technology Centre for Research Impact & Outcome Chitkara University Punjab Rajpura 140401 India Mazaya University College Computer Technologies Engineering Nasiriyah Iraq Kangwon National University Department of Artificial Intelligence & Software Samcheok 25913 South Korea Sejong University Department of Artificial Intelligence and Data Science Seoul 05006 South Korea

Although coffee is a crucial commodity for the economies of developing nations, defects in processing can significantly impact the safety, commercial value, and quality of coffee beans. Traditional coffee bean grading methods are labor-intensive and prone to error, necessitating automated and accurate classification of bean diseases. This study proposes a novel transfer learning-based deep neural network, VGGBM-Net, for the classification of coffee bean diseases. It focuses on the timely and precise identification of diseases to improve export quality without compromising safety during production. In this research, the proposed approach is trained using a dataset USK-Coffee of coffee beans to classify healthy and unhealthy beans. The performance of multiple machine learning classifiers, including Random Forest (RF), Logistic Regression (LoR), LightGBM (LGBM), and K-Nearest Neighbor Classifier (KNC), is evaluated alongside neural network techniques such as Convolutional Neural Networks (CNN) and VGG-19. A novel transformation of the VGG-19 model for feature engineering based on transfer learning is introduced, where spatial features extracted from coffee bean images are transformed into class prediction probabilities using LGBM. These enhanced features are then used as inputs for advanced machine-learning algorithms. Unlike traditional models this feature extraction enhances classification accuracy and robustness. This transformation improves feature extraction by capturing more discriminative patterns, leading to superior performance compared to conventional methods. Experimental results highlight the superior performance of the LGBM classifier, achieving an impressive 99% accuracy, recall, f1, and precision score of 98% with a computational runtime of just 0.084 seconds. K-fold cross-validation ensured the robustness of the models, and optimization techniques were applied to fine-tune parameters for maximum accuracy. This research establishes a highly effective fra

关键词： Coffee Bean Disease Deep Learning Feature engineering Image Processing Transfer Learning

来源：评论

学校读者我要写书评

暂无评论

POS tagging of low-resource Pashto language: annotated corpus and BERT-based model

引用

Language Resources and Evaluation 2025年 1-23页

作者： Haq, Ijazul Zhang, Yingjie Qadri, Intakhab Alam Shien-Ming Wu School of Intelligent Manufacturing South China University of Technology Guangzhou China Artificial Intelligence Department Guangdong CAS Angels Biotechnology Co. Ltd. Foshan China College of Computer Science and Software Engineering Shenzhen University Shenzhen 518060 China

This paper presents the development of a comprehensive part-of-speech (POS) annotated corpus for the low-resource Pashto language, along with a deep learning model for automatic POS tagging. The corpus comprises approximately 700K words (30K sentences), labeled for word boundaries, considering Pashto lacks explicit delimiters for word segmentation. The corpus was then annotated for POS information using a concise and pragmatic tagset of 36 grammatical categories. We utilized this corpus to train a supervised POS tagging model. For the model development, we employed a multilingual BERT (Bidirectional Encoder Representations from Transformers), fine-tuning it for this specific task. The BERT-based model performance was evaluated against an RNN-based model that employs the BiLSTM-CRF network and word embeddings (Word2Vec, fastText, and GloVe). Experimental results demonstrated that the BERT-based model achieved the best performance, attaining an accuracy of 96.24% and an F1-score (weighted average) of 96.22%. The model’s performance is highly satisfactory, making it useful in practical applications. Furthermore, the applications of the annotated corpus are not limited to this study only;it can be employed in various NLP applications, including named entity recognition (NER), text proofing, and constituency and dependency parsing. © The Author(s), under exclusive licence to Springer Nature B.V. 2025.

关键词： Artificial Intelligence BERT Corpus Linguistics LLMs Low-resource Languages Machine Learning NLP Pashto POS Tagging Transformers

来源：评论

学校读者我要写书评

暂无评论

WQuatNet: Wide range quaternion-based head pose estimation

引用

Journal of King Saud University - computer and Information sciences 2025年第3期37卷 1-14页

作者： Algabri, Redhwan Shin, Hyunsoo Abdu, Ahmed Bae, Ji-Hun Lee, Sungon Dept. of Computer Science & Engineering College of AI Convergence Sejong University Seoul 05006 South Korea Research Institute of Engineering and Technology Hanyang University Ansan 15588 South Korea School of Software Northwestern Polytechnical University Xi’an 710072 China Department of Robotics Hanyang University Ansan 15588 South Korea

Head pose estimation (HPE) is a critical task for numerous applications ranging from human-computer interaction, healthcare, and robotics, to surveillance. Most existing methods employ Euler angles as a representation, which often face challenges such as a gimbal lock, especially in full-range rotation scenarios or rotation matrices that require nine parameters. This study introduces WQuatNet, a novel deep learning-based model that leverages the quaternion representation, which uses only four parameters, to avoid this challenge. WQuatNet was designed based on a landmark-free HPE method to predict head poses across the full-range angles of 360∘ from images. Landmark-free methods bypass the need for explicit detection of facial landmarks;instead, they leverage the entire image to estimate the head orientation. The model incorporates a RepVGG-D2se backbone for robust feature extraction and introduces two loss functions tailored for quaternion predictions. Our experimental results on multiple HPE datasets covering both narrow- and full-range angles demonstrate that WQuatNet outperforms the state-of-the-art (SOTA) approaches in terms of accuracy. The performance of the proposed HPE was evaluated using the CMU, AGORA, BIWI, AFLW2000, and 300W-LP datasets. We also perform ablation studies and error analyses to validate the significance of each component of the model. © The Author(s) 2025.

关键词： Deep neural network Full range of rotation Head pose estimation Quaternion

来源：评论

学校读者我要写书评

暂无评论

Novel Load Balancing Mechanism Based on Deep Reinforcement Learning for Human-Centric Applications in Data Centers

引用

Human-centric Computing and Information sciences 2025年 15卷

作者： Hu, Jinbin Luo, Wangqing He, Yi Rao, Shuying Alqahtani, Fayez Tolba, Amr Zhu, Min Wang, Jin School of Computer and Communication Engineering Changsha University of Science and Technology Changsha China Software Engineering Department College of Computer and Information Sciences King Saud University Riyadh Saudi Arabia Department of Computer Science Community College King Saud University Riyadh Saudi Arabia College of Information Science and Technology Zhejiang Shuren University Zhejiang China

Human-computer interaction applications require ultra-low response delays to achieve a better user experience. With the rapid growth of data center network (DCN) topologies, the number of equal-cost paths has increased dramatically. It is imperative to use load balancing between parallel paths to improve link utilization. However, the existing load balancing mechanisms cannot make appropriate forwarding decisions for reacting to the dynamic traffic and path congestion status in large-scale DCNs, resulting in poor network performance. In this paper, we have proposed a dynamic load balancing (DLB) mechanism that uses deep reinforcement learning (DRL) to monitor the status of parallel paths in real time and uses the depth deterministic policy gradient algorithm to make the path state weight (PSW) automatically adapt to different network load scenarios. It then automatically configures the PSW through the DRL agent to guide the switch to select the best forwarding path to balance the load. The experimental results show that DLB significantly reduces the average flow completion by up to 36% compared to state-of-the-art load balancing schemes such as DRILL, LetFlow, and RPS under the asymmetric topology scenario, the result of which significantly enhances the efficiency of human-centric interaction. © (2024), (Korea Information Processing Society). All Rights Reserved.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

A Practical Learning-Based FTL for Memory Constrained Mobile Flash Storage

A Practical Learning-Based FTL for Memory Constrained Mobile...

引用

Design, Automation and Test in Europe Conference and Exhibition

作者： Zelin Du Kecheng Huang Tianyu Wang Xin Yao Renhai Chen Zili Shao Department of Computer Science and Engineering The Chinese University of Hong Kong Hong Kong SAR College of Computer Science and Software Engineering Shenzhen University Shenzhen China Huawei Technologies China

ISBN: (数字)9783982674100

ISBN: (纸本)9798331534646

The rapidly growing mobile market is pushing flash storage manufacturers to expand capacity into the terabyte range. However, this presents a significant challenge for mobile storage management: more logical-to-physical page mappings are desired to be efficiently managed and cached while the available caching space is extremely limited. This motivates us to shift toward a new learning-based paradigm: rather than maintaining mappings for individual pages, the learning-based approach can represent mapping relationships for a set of continuous pages. However, to construct linear models, existing methods that either consume the already-limited memory space or reuse flash garbage collection demonstrate poor model construction capabilities or significantly degrade flash performance, making them impractical for real-world use. In this paper, we propose LFTL, a practical, learning-based on-demand flash translation layer design for flash management in mobile devices. In contrast to prior work that centered around gathering sufficient mappings for linear model construction, our key insight is that linear patterns can be extracted and refined by leveraging the orderly, LPA-aligned write stream typical of mobile devices. By doing this, highly accurate linear models can be constructed regardless of the constraints of mobile device's cache limitation. We have implemented a fully functional prototype of LFTL based on FEMU. Our evaluation results show that LFTL is more adaptable to memory-constrained storage devices than state-of-the-art learning-based approaches.

关键词： Performance evaluation Adaptation models Maximum likelihood detection Translation Accuracy Storage management Prototypes Nonlinear filters Mobile handsets Load modeling

来源：评论

学校读者我要写书评

暂无评论

Enhanced Security Protocol for VLSI Systems: Modified AES Algorithm for Robust Data Transmission 3

Enhanced Security Protocol for VLSI Systems: Modified AES Al...

引用

3rd International Conference on Integrated Circuits and Communication Systems, ICICACS 2025

作者： Lawrence, Jinsha Budagam, Dinesh Kumar Mukilan, P. Veerappan, Kavitha Chandrasekhar, E. Barathi, K. Faculty of Engineering Department of Computer Science and Engineering Coimbatore India Visa Department of Ms in Software Engineering CA United States Dhanalakshmi Srinivasan College of Engineering Department of Electronics and Communication Engineering Coimbatore India Kongu Engineering College Department of Electronics and Communication Engineering Erode India Department of Electronics and Communication Engineering Hyderabad India Vel Tech Rangarajan Dr. Sagunthala R&d Institute of Science and Technology Department of Electrical and Electronics Engineering Chennai India

ISBN: (纸本)9798331508456

The development of compact and efficient devices has been made possible by the growth of Very Large Scale Integration (VLSI) technologies, which has transformed modern electronics. However, there are cautions regarding the security of data, especially when it comes to transmission due to the extensive usage of technology. To address these, this study proposes a modified Advanced Encryption Standard (AES) algorithm as a solution for improved security protocol in VLSI systems. The proposed algorithm are intended to support the encryption procedure, increasing stronger production against cryptographic attacks while preserving the system's effectiveness and speed factors that are critical aspects for VLSI applications. Through extensive simulations and testing, the modified AES algorithm demonstrated significant security enhancements the operational efficiency of the VLSI system. The attained outcomes of the proposed work shows that the Modified AES strategy provides a workable technique to protect data in VLSI-based devices while maintaining data communication integrity and confidentiality. The proposed modified AES algorithm demonstrates a significant improvement in performance with a propagation delay of 7 ns, power consumption of 29 mW, and a computational overhead of 62 bits, leading to enhanced efficiency in cryptographic operations. This study highlights that the modified AES algorithm improve security in modern electronic systems and maintain performance. © 2025 IEEE.

关键词： Network security

来源：评论

学校读者我要写书评

暂无评论

Accurate prediction of virulence factors using pre-train protein language model and ensemble learning

引用

BMC Genomics 2025年第1期26卷 1-16页

作者： Li, Guanghui Zhou, Jian Luo, Jiawei Liang, Cheng School of Information and Software Engineering East China Jiaotong University Nanchang China College of Computer Science and Electronic Engineering Hunan University Changsha China School of Information Science and Engineering Shandong Normal University Jinan China

As bacterial pathogens develop increasing resistance to antibiotics, strategies targeting virulence factors (VFs) have emerged as a promising and effective approach for treating bacterial infections. Existing methods mainly relied on sequence similarity, and remote homology relationships cannot be discovered by sequence analysis alone. To address this limitation, we developed a protein language model and ensemble learning approach for VF identification (PLMVF). Specifically, we extracted features from protein sequences using ESM-2 and their three-dimensional (3D) structures using ESMFold. We calculated the true TM-score of the proteins based on their 3D structures and trained a TM-predictor model to predict structural similarity, thereby capturing hidden remote homology information within the sequences. Subsequently, we concatenated the sequence-level features extracted by ESM-2 with the predicted TM-score features to form a comprehensive feature set for prediction. Extensive experimental validation demonstrated that PLMVF achieved an accuracy (ACC) of 86.1%, significantly outperforming existing models across multiple evaluation metrics. This study provided an ideal tool for identifying novel targets in the development of anti-virulence therapies, offering promise for the effective prevention and control of pathogenic bacterial infections. The proposed PLMVF model offers an efficient computational approach for VF identification.

关键词： Ensemble learning Protein language model Remote homology TM-score Virulence factor prediction

来源：评论

学校读者我要写书评

暂无评论

One Gray Code Fits All: Optimizing Access Time with Bi-Directional Programming for QLC SSDs

One Gray Code Fits All: Optimizing Access Time with Bi-Direc...

引用

Design, Automation and Test in Europe Conference and Exhibition

作者： Shaoqi Li Tianyu Wang Yongbiao Zhu Chenlin Ma Yi Wang Zhaoyan Shen Zili Shao College of Computer Science and Software Engineering Shenzhen University Shenzhen China School of Computer Science and Technology Shandong University Qingdao China Department of Computer Science and Engineering The Chinese University of Hong Kong Hong Kong China

ISBN: (数字)9783982674100

ISBN: (纸本)9798331534646

Gray code, a voltage-level-to-data-bit translation scheme, is widely used in QLC SSDs. However, it causes the four data bits in QLC to exhibit significantly different read and write performance with up to 8 × latency variation, severely impacting the worst-case performance of QLC SSDs. This paper presents BDP, a novel Bi-Directional Programming scheme. Based on a fixed Gray code, BDP combines both the normal (forward) and reverse programming directions to enable runtime programming direction arbitration. Experimental results show that BDP can effectively improve the read and write performance of SSD compared to representative schemes.

关键词： Translation Runtime Europe Bidirectional control Programming Reflective binary codes

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：