检索结果-内蒙古大学图书馆

Video captioning using transformer-based GAN

Multimedia Tools and Applications 2025年第10期84卷 7091-7113页

作者： Babavalian, Mohammad Reza Kiani, Kourosh Electrical and Computer Engineering Department Semnan University Semnan Iran

Video captioning is the process of automatically generating natural language descriptions of video content. Historically, most video captioning methods have relied on extending Sequence-to-Sequence (Seq2Seq) models. However, such approaches possess limitations due to the sequential nature of the captions, which leads to less accurate captions. To address this limitation, this paper introduces a novel end-to-end architecture for video captioning that combines conditional Wasserstein Generative Adversarial Networks (cWGAN) with a transformer model. The proposed architecture consists of two modules: feature extraction and caption generation. The feature extraction module aims to obtain an encoded feature vector representing the video contents, while the caption generation module generates human-readable captions from encoded feature vector. To the best of our knowledge, this is the first architecture for generative video captioning that integrates a transformer model with GAN. The results of the proposed model based on the BLEU-4, METEOR, ROUGE-L, and CIDEr metrics, on two datasets, MSVD (BLEU-4 = 61.2, METEOR = 41.6) and MSR-VTT (BLEU-4 = 61.2, METEOR = 31.1), compared to state-of-the-art approaches, demonstrate the effectiveness of the transformer with generative model in generating accurate and human-readable captions. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

Semantic Feature engineering with LSA-SVM for Cyberbullying Comment Classification on Instagram

Informatica (Slovenia)

引用

Informatica (Slovenia) 2025年第15期49卷 165-178页

作者： Wulandari Makmur, Haerunnisya Surianto, Dewi Fatmarani Risal, Andi Akram Nur Budiarti, Nur Azizah Eka Zain, Satria Gunawan Wahid, Abdul Department of Computer Engineering State University of Makassar Makassar Indonesia

Social media is now an essential part of everyday life, with Instagram being one of the most popular platforms and often utilized for various purposes, one of which is to increase popularity. However, the platform also often becomes a place where acts of violence and impoliteness in commenting increase, known as cyberbullying. To address the problem, detecting and classifying cyberbullying comments on Instagram is an important step in cyberbullying prevention. However, in text classification, several challenges need to be overcome to ensure the success of the model, such as polysemy, curse of dimensionality, and selection of text representation for feature extraction. Therefore, this study aims to implement a feature engineering technique using a hybrid approach that combines word weighting with TF-IDF and LSA method to reduce feature dimensionality and capture the semantic meaning of the data, with SVM used as a classifier to classify bullying and non-bullying comments. The results showed that the proposed method using feature engineering of the LSA matrix formed from the dataset of one of the classes, yielded a significant accuracy of 97%. In comparison, the conventional method with feature engineering using TF-IDF and the use of LSA matrix formed from the dataset of both classes only achieved an accuracy of 84%. This shows that the proposed method is more effective than the baseline approach. © 2025 Slovene Society Informatika. All rights reserved.

关键词： Support vector machines

来源：评论

学校读者我要写书评

暂无评论

DTBA-net: Drug-Target Binding Affinity prediction using feature selection in hybrid CNN model

引用

Journal of computer-Aided Molecular Design 2025年第1期39卷 1-19页

作者： Mishra, Priya Vipsita, Swati Department of Computer Science and Engineering IIIT Bhubaneswar Odisha India

In drug discovery, virtual screening and repositioning rely on accurate Drug-Target Binding Affinity (DTBA) prediction to develop effective therapies. However, DTBA prediction remains challenging due to limited annotated datasets, high-dimensional biochemical data, and heterogeneous data sources, including chemical structures, biological sequences, and molecular interactions. These complexities hinder the development of unified deep-learning frameworks. To address these challenges, we propose DTBA-Net, a novel hybrid neural network model that enhances DTBA prediction accuracy and efficiency. DTBA-Net integrates optimal feature selection within a CNN architecture to predict DTBA. Protein sequences and compound structures are processed through a hybrid CNN that includes convolutional layers, a flattened layer, a Modified JAYA Algorithm for optimal feature selection, and dense blocks. The Modified JAYA algorithm selects relevant features, reduces computational overhead, and improves predictive performance. DTBA-Net was evaluated on two benchmark datasets, KIBA and DAVIS. On the DAVIS dataset, DTBA-Net attained an R-squared value of 0.95 and a Mean Absolute Error (MAE) of 0.17. Further validation using the drug Nirmatrelvir resulted in an R-squared value of 0.96, showcasing the model’s robustness and scalability. Integrating a hybrid neural network with an optimized feature selection process accelerates model training and enhances prediction accuracy. DTBA-Net demonstrates promising potential for scalable, efficient, and accurate DTBA prediction, facilitating faster and more reliable drug discovery. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2025.

关键词： Binding energy

来源：评论

学校读者我要写书评

暂无评论

Ill-condition enhancement for BC speech using RMC method

引用

International Journal of Speech Technology 2024年第4期27卷 1085-1092页

作者： Ohidujjaman Hasan, Mahmudul Zhang, Shiming Huda, Mohammad Nurul Uddin, Mohammad Shorif Computer Science and Engineering Daffodil International University Dhaka1216 Bangladesh Computer Science and Engineering Comilla University Comilla3506 Bangladesh School of Electrical and Information Northeast Agricultural University Harbin150030 China Computer Science and Engineering United International University Dhaka1212 Bangladesh Computer Science and Engineering Green University of Bangladesh Kanchon1460 Bangladesh Computer Science and Engineering Jahangirnagar University Savar1342 Bangladesh

This paper improves the ill-condition of bone-conducted (BC) speech signal by reducing the eigenvalue expansion. BC speech commonly contains a large spectral dynamic range that causes ill-condition for the classical linear prediction (LP) methods. In the field of numerical analysis, we often face the situation where an ill-conditioned case occurs in finding the solution. Principally, eigenvalue expansion causes ill-condition in numerical analysis. To mitigate this problem, the regularized least squares (RLS) technique is commonly used. Motivated by the RLS concept, we derive the regularized modified covariance (RMC) method for BC speech analysis in this study. The RMC method reduces eigenvalue expansion by compressing the spectral dynamic range of the speech signal. Thus, the RMC method resolves the ill-conditioned problem of LP. In experiments, we show that the RMC method provides compressed eigenvalue expansion than the conventional methods for BC speech where synthetic and real BC speeches are considered. The performance of the RMC method is affected by the setting of the regularization parameter. In this paper, the regularization parameter in practice is iteratively and rule-based derived. The RMC method with such a setting provides the best performance for BC speech analysis. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.

关键词： Iterative methods

来源：评论

学校读者我要写书评

暂无评论

Impact of exposure to air pollutants on the spread, severity, and mortality of viral infections: a systematic review

引用

International Journal of Environmental Studies 2025年第3期82卷 1327-1354页

作者： Gendy, Maggie Ezzat Gaber Yuce, Mehmet Rasit Electrical and Computer Systems Engineering Monash University Melbourne Australia

Recent research suggests that air pollution’s detrimental effects may influence viral transmission and severity. The COVID-19 pandemic has introduced the concept of air pollution-to-human subject transmission. Pollutants may contribute to virus spread, severity, and mortality. Numerous studies have found correlations between higher COVID-19 hospitalisation and death rates and elevated air pollution levels. This systematic review presents a comparative analysis of studies from COVID-19-affected countries, categorising them by the short-term and long-term effects of air pollution. We examine study duration, types of pollutants, statistical models, and confounding variables, such as health, socio-economic, demographic, and climate factors. Environmental pollutants have a critical role in viral transmission. Air quality monitoring can enhance future pandemic preparedness and mitigation efforts. © 2025 The Author(s). Published by Informa UK Limited, trading as Taylor & Francis Group.

关键词： COVID-19

来源：评论

学校读者我要写书评

暂无评论

NORRIS: Noise-Resilient and Resource-Efficient Semantic Encoded Point Cloud Data Transmission for Internet of Things Communications

引用

IEEE Internet of Things Journal 2025年第13期12卷 25720-25731页

作者： Safarnejad, Saleh Hosein Soheilian, Mohammad Safaei, Bardia Sharif University of Technology Department of Computer Engineering Tehran Iran

Employing point cloud and 3-D data has recently proliferated with the advent of emerging Internet of Things (IoT) applications, e.g., Autonomous Vehicles (AV), and Augmented Reality/Virtual Reality (AR/VR), with their resource-limited IoT devices equipped with Light Detection and Ranging (LiDAR) sensors. While offloading the point cloud computations could improve energy-efficiency, the imposed bandwidth and energy dissipation due to transmitting a significant volume of 3-D data is still challenging. This issue gets more complicated due to the instability caused by the noisy channel conditions, which leads to the cliff effect and reduced object detection accuracy. Accordingly, this article introduces NORRIS;a semantic-based point cloud data transmission technique that mitigates transmitted data while enhancing resilience against channel noise. NORRIS utilizes joint source-channel coding in combination with transfer learning. It emphasizes efficiency and performance, providing a lightweight model that sustains excellent performance across different channel conditions, making it ideal for resource-constraint IoT devices. While NORRIS provides high detection accuracy, it significantly reduces the inference time and model complexity. The effectiveness of NORRIS is evaluated and compared with state-of-the-art through a comprehensive set of experiments. Results demonstrate that NORRIS successfully overcomes the cliff effect while guaranteeing a higher performance in lower Signal-to-Noise Ratio (SNR) conditions. It also reduces the transmitted data volume by up to 10\times , while accelerating meaning extraction by 5\times . It also enhances accuracy by approximately 10% in various SNR scenarios. Furthermore, NORRIS maintains stable classification performance under noise, with only a 5% accuracy drop, in contrast to the 60% decrease observed in existing approaches. © 2014 IEEE.

关键词： Channel coding

来源：评论

学校读者我要写书评

暂无评论

Resource Allocation and Performance Analysis in D2D Underlay Cellular Networks with Poisson Cluster Process

引用

Journal of Communications and Information Networks 2025年第1期10卷 71-80页

作者： Shuai Chen Kunqi Guo School of Computer Science and Communications Engineering Jiangsu UniversityZhenjiang 212000China

Device-to-device(D2D)communication underlay cellular networks offers several benefits,including offloading cellular traffic and improving *** this paper,a scheme is proposed for D2D communication underlay cellular networks,by taking resource allocation algorithm into ***,a scenario where cellular users(CUs)and D2D users(DUs)coexist is *** characterize the clustering feature of DUs,the locations of CUs and DUs are modeled as Poisson cluster processes(PCPs),and the locations of base stations(BSs)are modeled as a Poisson point process(PPP).Subsequently,a resource allocation algorithm is proposed to allocate resources for CUs to DUs in the cluster,which involves the optimization of matching between D2D user and cellular ***,the coverage probability of the clustering network is *** simulation results indicate that the proposed scheme can effectively enhance system capacity and reveal the effects of various parameters on system performance.

关键词： coverage probability D2D communication Poisson cluster process resource allocation

来源：评论

学校读者我要写书评

暂无评论

Hungarian-based Heuristic for Permutation-invariant Multi-robot Path Planning

引用

Journal of Institute of Control, Robotics and Systems 2025年第4期31卷 286-294页

作者： Lee, Jeongmin Na, Hyeon-Suk School of Computer Science and Engineering Soongsil University Korea Republic of

The Hungarian algorithm is a well-known cubic-time algorithm for finding minimum-cost matchings in weighted bipartite graphs. While utilizing it for multi-agent path planning yields the minimum-total-length set of paths for n agents, these paths may intersect. We propose a quadratic-time heuristic that computes a close-to-minimum-total-length set of agent paths and enables highly efficient collision-free modification of these paths. We employ several techniques to optimize both time efficiency and total path length, including grouping nearby points and exploiting the geometric properties of these groups. First, we utilize k-d trees to group nearby starting/destination points. We then use the Hungarian algorithm to match the centroids of these groups, followed by a second-round Hungarian matching of the individual points within each paired group. This hierarchical approach effectively reduces the time complexity from cubic to quadratic. Second, we accelerate the collision-free refinement of the computed paths by identifying disjoint group pairs through a simple geometric test and excluding them from collision-checking. Based on experimental results, our heuristic algorithm is highly effective in terms of both time efficiency and total path length;it achieves 1.001-optimal solutions for 4K agents in 0.4 s and for 250K agents in 875 s, whereas the Hungarian spends 192 s on computing only the minimum-total-length paths (before collision-checking) for 4K agents. We compare our method against four algorithms: Hungarian and greedy matchings, both with brute-force collision-checking and modification, and two previous algorithms developed by ourselves. © ICROS 2025.

关键词： Multipurpose robots

来源：评论

学校读者我要写书评

暂无评论

Evaluation of Decomposition to Microservices with Optimized Inter-Service Communication Strategy

IAENG International Journal of Computer Science

引用

IAENG International Journal of computer Science 2025年第6期52卷 1907-1920页

作者： Weerasinghe, L.D.S.B. Perera, Indika Department of Computer Science & Engineering University of Moratuwa Sri Lanka

During the early stages of software development, monolithic architectures predominated within software systems. However, their limitations in effectively accommodating dynamic user demands prompted the widespread adoption of microservices architecture. Microservices offer solutions by anticipating changes in user needs and technological advancements. Most enterprise-grade software is now deployed in a cloud-native environment and transitioning towards a microservice-based architecture. This study investigates how inter-service communication affects the performance of the overall microservices architecture. In order to find the optimal communication segments, the researchers have conducted several experiments. The research experiments also considered software quality attributes such as maintainability, scalability, portability, testability, and reliability. The evaluations’ findings indicate that a discernible decrease in overall system performance occurs when more than six microservices are communicated to fulfil user requirements. This emphasizes how crucial it is for architects to consider communication patterns when designing or transitioning to microservices prudently. The findings and insights from this study are anticipated to contribute valuable knowledge and guidance for future advancements in microservices developments. © (2025), (International Association of Engineers). All rights reserved.

关键词： Software design

来源：评论

学校读者我要写书评

暂无评论

Lightweight Model for Occlusion Removal from Face Images

Annals of Emerging Technologies in Computing

引用

Annals of Emerging Technologies in Computing 2024年第2期8卷 1-14页

作者： John, Sincy Danti, Ajit Department of Computer Science and Engineering Christ University India

In the realm of deep learning, the prevalence of models with large number of parameters poses a significant challenge for low computation device. Critical influence of model size, primarily governed by weight parameters in shaping the computational demands of the occlusion removal process. Recognizing the computational burdens associated with existing occlusion removal algorithms, characterized by their propensity for substantial computational resources and large model sizes, we advocate for a paradigm shift towards solutions conducive to low-computation environments. Existing occlusion riddance techniques typically demand substantial computational resources and storage capacity. To support real-time applications, it's imperative to deploy trained models on resource-constrained devices like handheld devices and internet of things (IoT) devices possess limited memory and computational capabilities. There arises a critical need to compress and accelerate these models for deployment on resource-constrained devices, without compromising significantly on model accuracy. Our study introduces a significant contribution in the form of a compressed model designed specifically for addressing occlusion in face images for low computation devices. We perform dynamic quantization technique by reducing the weights of the Pix2pix generator model. The trained model is then compressed, which significantly reduces its size and execution time. The proposed model, is lightweight, due to storage space requirement reduced drastically with significant improvement in the execution time. The performance of the proposed method has been compared with other state of the art methods in terms of PSNR and SSIM. Hence the proposed lightweight model is more suitable for the real time applications with less computational cost. © 2024 by the author(s).

关键词： Generative adversarial networks

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

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

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：