检索结果-内蒙古大学图书馆

Emerging Smart Computing and Informatics (ESCI), Conference on

作者： Mohanapriya S Ram Akaashu B Sabarish V Mohana Saranya S Dinesh K Department of Computer Science and Engineering Kongu Engineering College Erode Tamilnadu India

ISBN: (数字)9798331515683

ISBN: (纸本)9798331515690

When incremental learning models encounter new classes, they often experience what is referred to as catastrophic forgetting, wherein previously learned information is lost during the training on new data. This happens for the following reason where deep learning models are expensive to compute and necessitate significant replay of memory, this is very relevant when it involves large datasets, such as those in the case of CIFAR-10. This work introduces a novel method called Weight Space Rotation that aims to improve recall and expedite adaptation in learning new classes by strategically modifying the model's weight space to reduce interference between previously learned and incoming knowledge. WSR does this efficiently without requiring storing earlier examples and is therefore efficient and scalable. Experiments on the CIFAR-10 dataset show that the WSR-enhanced model outperforms conventional techniques in class-incremental learning. WSR achieves this through the allocation of discrete regions in the weight space to different classes, thus enhancing knowledge retention and reducing catastrophic forgetting. This also enhances scalability and reduces computational load as it simplifies regularization processes. The results indicate that WSR significantly enhances the model's capability to learn rapidly new classes in incremental settings while retaining long-term knowledge. For the CIFAR-10 dataset, this approach integrates WSR into the training pipeline of a standard CNN architecture, successfully adapting to new classes while maintaining knowledge of previously learned ones. Experiments confirm that the WSR-enhanced model surpasses current systems by achieving better retention of prior knowledge and faster adaptation to new classes, thereby enhancing performance in class-incremental scenarios.

关键词： Training Support vector machines Adaptation models Incremental learning Accuracy Computational modeling Scalability Stability analysis Standards Load modeling

来源：评论

学校读者我要写书评

暂无评论

Efficient Mulberry Leaf Disease Detection in Bangladesh: A Lightweight Approach for Real-Time Applications

Efficient Mulberry Leaf Disease Detection in Bangladesh: A L...

引用

International Conference on Electrical, computer and Communication engineering (ECCE)

作者： Ananta Raha Quazi Samiha Tasnim Department of Computer Science and Engineering Bangladesh University of Engineering and Technology Dhaka Bangladesh

ISBN: (数字)9798350357509

ISBN: (纸本)9798350357516

Bombyx mori, the silkworm species that feeds on mulberry leaves, produces the most prevalent variant of silk worldwide. Unfortunately, mulberry trees suffer from leaf diseases that spread quickly and cause significant losses in production. Identifying diseases by experts on large scale is time-consuming and often inaccessible for rural people. Despite recent advances in deep learning and image processing, State-of-the-art models tend to be computationally expensive and must be used online. This study makes use of a regional dataset that comprises of mulberry leaf images captured from Rajshahi, the principal sericulture region in Bangladesh, to align with the environmental factors of the country. Firstly, we present an extremely efficient and lightweight approach enhanching a pre-trained MobileNetV2 model to classify among three categories of leaf photos: disease free, leaf spot, and leaf rust. Upon comparison with base MobileNetV2 and existing works on this dataset, this model exhibited efficacy in accuracy, recall, precision, and F1-score, attaining remarkable values of 98.63%, 98.53%, 98.63%, and 98.63%, respectively after 5-fold cross-validation. Then, we performed an explainability analysis utilizing Grad-CAM visualization to validate the capability of our suggested method. We also assessed the impact of preprocessing techniques on classification outcomes. The proposed architecture, having only 145K parameters, remains surprisingly lightweight and outperforms the base MobileNetV2 model in our experiments. According to the results, our suggested approach correctly identifies mulberry leaf diseases, supporting sericulture growth in rural areas.

关键词： Deep learning Measurement Visualization Image processing Rural areas computer architecture Production Real-time systems Diseases Resistance heating

来源：评论

学校读者我要写书评

暂无评论

Cybersecurity Assessment of Smart Grid Exposure Using a Machine Learning Based Approach

arXiv

引用

arXiv 2025年

作者： Jeje, Mofe O. Department of Computer Science School of Electrical Engineering and Computer Science University of North Dakota Grand FolksND United States

Given that disturbances to the stable and normal operation of power systems have grown phenomenally, particularly in terms of unauthorized access to confidential and critical data, injection of malicious software, and exploitation of security vulnerabilities in a poorly patched software among others;then developing, as a countermeasure, an assessment solutions with machine learning capabilities to match up in real-time, with the growth and fast pace of these cyber attacks, is not only critical to the security, reliability and safe operation of power system, but also germane to guaranteeing advanced monitoring and efficient threat detection. Using the Mississippi State University and Oak Ridge National Laboratory dataset, the study used an XGB Classifier modeling approach in machine learning to diagnose and assess power system disturbances, in terms of Attack Events, Natural Events and No-Events. As test results show, the model, in all the three sub-datasets, generally demonstrates good performance on all metrics, as it relates to accurately identifying and classifying all the three power system events. © 2025, CC BY-NC-ND.

关键词： Adversarial machine learning

来源：评论

学校读者我要写书评

暂无评论

Mobile Edge Computing for Efficient Vehicle Management in Smart City

Mobile Edge Computing for Efficient Vehicle Management in Sm...

引用

International Conference on Communication Systems and Network Technologies (CSNT)

作者： Asem Debala Chanu Saylee Shelar Shubha Brata Nath Department of Computer Science and Engineering Indian Institute of Information Technology Guwahati Guwahati India Department of Computer Science and Engineering Ajeenkya DY Patil University Pune India

ISBN: (数字)9798331531935

ISBN: (纸本)9798331531942

Mobile edge computing (MEC) promises to be a useful technology to enable applications in a smart city. However, the task offloading becomes challenging as the MEC infrastructure has limited resources. To solve the challenges of MEC, we have proposed a system architecture with an optimization algorithm to minimize the task delay in MEC. A Genetic Algorithm based task allocation algorithm finds the optimal set of edge servers for task deployment and processing. The proposed system efficiently optimizes the task delay in MEC as analyzed from the experiments performed by taking virtual machines as MEC servers. We show the usefulness of the proposed system by taking an application scenario for efficient vehicle management in a smart city.

关键词： Multi-access edge computing Smart cities Scalability Systems architecture Virtual machines Delays Servers Resource management Optimization Genetic algorithms

来源：评论

学校读者我要写书评

暂无评论

Distributed Multi-Head Learning Systems for Power Consumption Prediction

arXiv

引用

arXiv 2025年

作者： Syu, Jia-Hao Chun-Wei Lin, Jerry Yu, Philip S. Department of Computer Science and Information Engineering National Taiwan University Taiwan Department of Computer Science Electrical Engineering and Mathematical Sciences Western Norway University of Applied Sciences Bergen Norway Department of Computer Science University of Illinois Chicago United States

As more and more automatic vehicles, power consumption prediction becomes a vital issue for task scheduling and energy management. Most research focuses on automatic vehicles in transportation, but few focus on automatic ground vehicles (AGVs) in smart factories, which face complex environments and generate large amounts of data. There is an inevitable tradeoff between feature diversity and interference. In this paper, we propose Distributed Multi-Head learning (DMH) systems for power consumption prediction in smart factories. Multi-head learning mechanisms are proposed in DMH to reduce noise interference and improve accuracy. Additionally, DMH systems are designed as distributed and split learning, reducing the client-to-server transmission cost, sharing knowledge without sharing local data and models, and enhancing the privacy and security levels. Experimental results show that the proposed DMH systems rank in the top-2 on most datasets and scenarios. DMH-E system reduces the error of the state-of-the-art systems by 14.5% to 24.0%. Effectiveness studies demonstrate the effectiveness of Pearson correlation-based feature engineering, and feature grouping with the proposed multi-head learning further enhances prediction performance. © 2025, CC BY-NC-SA.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Investigating Application of Deep Neural Networks in Intrusion Detection System Design

arXiv

引用

arXiv 2025年

作者： Jeje, Mofe O. Department of Computer Science School of Electrical Engineering and Computer Science University of North Dakota Grand ForksND United States

Despite decades of development, existing IDSs still face challenges in improving detection accuracy, evasion, and detection of unknown attacks. To solve these problems, many researchers have focused on designing and developing IDSs that use Deep Neural Networks (DNN) which provides advanced methods of threat investigation and detection. Given this reason, the motivation of this research then, is to learn how effective applications of Deep Neural Networks (DNN) can accurately detect and identify malicious network intrusion, while advancing the frontiers of their optimal potential use in network intrusion detection. Using the ASNM-TUN dataset, the study used a Multilayer Perceptron modeling approach in Deep Neural Network to identify network intrusions, in addition to distinguishing them in terms of legitimate network traffic, direct network attacks, and obfuscated network attacks. To further enhance the speed and efficiency of this DNN solution, a thorough feature selection technique called Forward Feature Selection (FFS), which resulted in a significant reduction in the feature subset, was implemented. Using the Multilayer Perceptron model, test results demonstrate no support for the model to accurately and correctly distinguish the classification of network intrusion. © 2025, CC BY-NC-ND.

关键词： Network intrusion

来源：评论

学校读者我要写书评

暂无评论

Supply chain tracking of drugs using Ethereum

Supply chain tracking of drugs using Ethereum

引用

Artificial Intelligence and Data engineering (AIDE), International Conference on

作者： Raghunandan K R Dhanya Avisha V Shetty Akshata D Bhat Devishree Bangera Nalini N Department of Computer Science and Engineering NMAM Institute of Technology (Deemed to be University) Udupi India Department of Computer Science and Engineering Nitte Meenakshi Institute of Technology Bangalore India

ISBN: (数字)9798331527518

ISBN: (纸本)9798331527525

Counterfeiting, increased fake intermediate products, lack of transparency, unauthorized alteration of data, and other related problems throughout the supply chain have made it necessary for the modern pharmaceutical industry to ensure integrity and security. Thus, the above-mentioned problem can be solved by a decentralized blockchain-based system, which ensures security and traceability. In the proposed solution, there are seven stakeholders including manufacturers and consumers. Authorized manufacturers are allowed to store the drugs in the blockchain network, and corresponding QR and bar codes are generated. This generates a reliable record of drug movements by constructing an architecture that is decentralized to avoid alteration of the data by anyone. Consumers have the opportunity to scan QR codes to ensure that their products are genuine and trace their origin. By addressing the challenges of the conventional pharmaceutical supply chain that provides high transparency, minimal counterfeits, and assured data security, this system addresses the aforementioned disadvantages.

关键词： Drugs Data security Supply chains QR codes Data engineering Blockchains Stakeholders Reliability Pharmaceutical industry Counterfeiting

来源：评论

学校读者我要写书评

暂无评论

SSE: A Speaking Style Extractor Based on Fine-Grained Contrastive Learning between Speech and Descriptive Text

SSE: A Speaking Style Extractor Based on Fine-Grained Contra...

引用

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

作者： Zhang, Zixing Wu, Yimeng Dong, Zhongren Xiang, Wulong Shen, Shengfan Schuller, Björn W. College of Computer Science and Electronic Engineering Hunan University China Shenzhen Research Institute Hunan University China GLAM Department of Computing Imperial College London United Kingdom

ISBN: (纸本)9798350368741

Effective extraction of paralinguistic features from speech, such as emotion, accent, and age, remains a challenging task in speech processing. Traditional methods typically address each type of paralinguistic information with separate classification or regression tasks-e. g., emotion recognition, accent detection, and age estimation. These approaches can result in fragmented and incomplete descriptions of speech style, failing to capture the full range of paralinguistic attributes in an integrated way. To address these limitations, we propose the Speech Style Extractor (SSE), a novel approach that aims to provide a more comprehensive extraction of speech style features. SSE leverages an optimized fine-grained contrastive learning scheme, enhancing the extraction of diverse paralinguistic features. The optimization approach outperforms the baseline on 11 out of 13 datasets, with an average improvement of 2.1% in terms of the R@1 evaluation metric. We also propose a complete data generation framework using large language models to create 199k text samples across 9 paralinguistic categories, such as emotion, deception, stuttering, and accent, to fill the gap in speaking style descriptions in public datasets. This extensive dataset not only facilitates our research but also serves as a valuable resource for the speech processing community. © 2025 IEEE.

关键词： Contrastive Learning Fine-Grained Loss Prompt engineering Speaking Styles

来源：评论

学校读者我要写书评

暂无评论

SciTexC: A Framework for Scientific Text Classification in Bengali using Transformer-Based Techniques

SciTexC: A Framework for Scientific Text Classification in B...

引用

International Conference on Robotics,Electrical and Signal Processing Techniques (ICREST)

作者： Fahmida Nahiyan Mohammed Moshiul Hoque Department of Computer Science & Engineering Chittagong University of Engineering and Technology Chattogram Bangladesh

ISBN: (数字)9798331530952

ISBN: (纸本)9798331530969

Scientific text classification (STC) in NLP automatically categorizes texts into predefined scien-tific domains. Although there is enormous progress in STC regarding high-resource languages, it must be explored in low-resource languages, specifically Ben-gali. This study introduces a framework, SciTexC (Scientific Text Classifier), which uses a transformer-based technique that classifies Bengali texts into six scientific domains: Physics, Chemistry, Biology, Information and Communication Technology (ICT), Mathematics, and Others. A new corpus, STexD (scientific text dataset), containing 6000 Bengali texts, was also developed for classification tasks. Five transformer-based models were used to perform the task, including mBERT, dmBERT, Bangla BERT, Indic BERT, and XLM-RoBERTa. The evaluation results demonstrate that XLM-RoBERTa outperformed the other exploited models, achieving the highest F1 score (0.9158).

关键词： Error analysis Text categorization Signal processing Transformers Mathematical models Data models Natural language processing Information and communication technology Physics Robots

来源：评论

学校读者我要写书评

暂无评论

Contact-Rich Task Learning on an Articulated Soft Robot Arm Through Simulation 25th

Contact-Rich Task Learning on an Articulated Soft Robot Ar...

引用

25th Annual Conference on Towards Autonomous Robotic Systems, TAROS 2024

作者： Elstner, Laurenz Kyrkjebø, Erik Stoelen, Martin F. Department of Computer Science Electrical Engineering and Mathematical Sciences Western Norway University of Applied Sciences Førde Norway Fieldwork Robotics Ltd. Cambridge United Kingdom

ISBN: (纸本)9783031720611

Traditional rigid robots face limitations in exploration during learning processes in dynamic and unknown environments where contact and impacts are common. Soft robots that leverage compliant materials, for example as part of variable stiffness actuators (VSAs), offer promising avenues for enhanced adaptability and safety. Deep Reinforcement Learning (RL), while proven to be a powerful tool, requires a lot of training data that is time-consuming to gather. This paper introduces a novel approach to simulating robots driven by antagonistic VSAs, as well as full-task learning with a soft robot arm in three environments with different contact types: path following (no contact), door opening (kinematic constraint), and surface wiping (continuous contact). We show that the tendon-driven robot can learn all three tasks, even if learning in a very low-level actuation space. The stretchable tendons enables it to perform relatively well on the contact-rich tasks. Further, using the proposed variable stiffness controller the robot is able to develop a stable behaviour in terms of forces applied to the environment. The results are promising for extending the approach to the physical robot. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Tendons

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：