检索结果-内蒙古大学图书馆

Enhanced DGA botnet domain detection and family classification via n-gram analysis and Hellinger distance

computer Networks 2025年 269卷

作者： Jungmin Lee Sungju Yun Nahyun Kim Yeonjoon Lee Department of Computer Science & Engineering Major in Bio Artificial Intelligence Hanyang University Ansan 15588 Republic of Korea College of Computing Hanyang University Ansan 15588 Republic of Korea

Bot masters spread malware to create botnets and use Domain Generation Algorithms (DGAs) to evade blacklist-based detection methods with numerous generated domains, posing a significant threat to network security. Since detection alone cannot halt malware operations, classifying DGA domains into their respective botnet families is essential for enabling targeted countermeasures and addressing vulnerabilities in infected systems. However, most existing approaches focus primarily on distinguishing DGA domains from legitimate ones and face challenges when classifying domains from DGA families with similar character distributions, highlighting the need for improved techniques. In response, we expand the focus to DGA family classification and conduct in-depth analyses using eXplainable Artificial Intelligence (XAI) techniques to explore the impact of n-grams on classification performance. These analyses reveal that n-gram preprocessing and Hellinger Distance (HD)-based features derived from n-gram probability distributions can significantly enhance classification performance. Building on these insights, we propose an integrated framework with two components, an N-gram-based Multi-scale One-Dimensional Convolutional Neural Network model (N-MODCNN) and a machine learning (ML) classifier utilizing HD features, for detecting and classifying DGA domains. N-MODCNN detects DGA domains from n-gram preprocessed inputs, and detected domains are classified into their respective botnet families by a soft ensemble approach that integrates predictions from N-MODCNN and the ML classifier, enabling robust and accurate classification. Experiments on recent public datasets show that our framework achieves up to 99% detection and classification accuracy. For families with similar character distributions, it achieves F1-scores exceeding 90%, representing improvements of up to 72 percentage points over existing methods.

关键词：

来源：评论

学校读者我要写书评

暂无评论

How shape information is coded by V4 cortical response of Macaque Monkey

引用

Journal of neurophysiology 2025年第6期133卷 2025 May 14页

作者： Meilan Liu Shanshan Chang Meixuan Chen Peichao Li Anna Wang Roe Jia Ming Hu Department of Neurosurgery of the Second Affiliated Hospital of Zhejiang University School of Medicine Interdisciplinary Institute of Neuroscience and Technology Zhejiang University Hangzhou 310029 China. Key Laboratory for Biomedical Engineering of Ministry of Education College of Biomedical Engineering and Instrument Science Zhejiang University Hangzhou 310027 China. MOE Frontier Science Center for Brain Science and Brain-machine Integration School of Brain Science and Brain Medicine Zhejiang University Hangzhou 310012 China. National Key Laboratory of Brain and Computer Intelligence Zhejiang University Hangzhou 310058 China. Zhejiang Key Laboratory of Research and Transformation for Major Neurosurgical Diseases Hangzhou Zhejiang 310000 China.

Previous neural recording studies have shown that monkey V4 can process the shape information across populations of neurons. The responses recorded from each single neuron make it possible to retrieve shape information. However, these studies did not fully characterize the spatial distribution of activity in the cortex. There are multiple types of functional columns (orientation, curvature) in V4; how do these structures respond to different shapes? Here, with intrinsic optical imaging, we explored the cortical responses of V4 to contours (straight and curved) and shapes (circle and square). We found that in V4, the response of neurons to different shapes is highly dependent on the compositional features contained in the shape. A specific local network would have a higher response magnitude to its corresponding shape than other shapes. Meanwhile, the cortical response of V4 exhibits a tolerance to the shift of stimulus location. Our results suggest that two essential cortical response features in V4 are the specificity of the activated response pattern in the cortex and tolerance to the stimulus location variance. These features can help decode shape information from imaging results.

关键词： Functional domains Macaque monkey Optical imaging Shape Visual cortex

来源：评论

学校读者我要写书评

暂无评论

Improved YOLOv8 framework for efficient solar panel defect detection

引用

Journal of Building engineering 2025年 111卷

作者： Amreen Batool Yong-Won Kim Yung-Cheol Byun Department of Electronic Engineering Institute of Information Science & Technology Jeju National University Jeju 63243 South Korea Department of Computer Engineering Major of Electronic Engineering Jeju National University Institute of Information Science & Technology Jeju 63243 South Korea

The photovoltaic (PV) industry plays a vital role in the renewable energy sector, making efficient fault detection essential to ensure optimal performance and sustainability. While the existing deep learning models have significantly improved defect detection accuracy, their large size and limited feature extraction capabilities reduce detection efficiency and complicate adaptation to varying defect conditions. Therefore, this study proposed the Improved YOLOv8-SEB model, which integrates a Squeeze-and-Excitation Block (SEB) into the YOLOv8 architecture to enhance defect detection in photovoltaic (PV) panels. The SEB dynamically reweights feature maps, allowing the model to emphasize critical features relevant to the task and improve the detection of complex and subtle defects such as “Black Border” and “Broken”. Experimental evaluations using a PV-Multi-Defect dataset show that YOLOv8-SEB achieves a 2.6% increase in mean average precision at IoU 0.5 ( [email protected] ) and a 3.4% gain in [email protected] :0.95 compared to the baseline YOLOv8, while reducing DFL Loss to 0.16. The improved model demonstrates enhanced precision and recall, with fewer false positives and negatives. Although the integration of SEB introduces additional computational complexity, YOLOv8-SEB remains suitable for real-time deployment on edge devices through potential optimizations. This research offers a scalable and efficient approach for solar panel defect detection, facilitating cost-effective maintenance and supporting the operational stability of PV systems. The proposed model contributes to the advancement of intelligent inspection technologies, helping to extend the lifespan of solar panels and promote the sustainability of renewable energy infrastructures.

关键词：

来源：评论

学校读者我要写书评

暂无评论

AI-driven state of power prediction in battery systems: A PSO-optimized deep learning approach with XAI

引用

Energy 2025年 331卷

作者： Jafari, Sadiqa Byun, Yung-Cheol Department of Electronic Engineering Institute of Information Science & Technology Jeju National University Jeju63243 Korea Republic of Department of Computer Engineering Major of Electronic Engineering Jeju National University Institute of Information Science & Technology Jeju63243 Korea Republic of

Accurate prediction of the State of Power (SoP) in Battery Management Systems (BMS) is crucial for maximizing battery efficiency, especially in electric cars and renewable energy storage fields. SoP is a crucial metric defining lithium-ion batteries’ power efficiency. Contrary to the State of Charge (SoC), the SoP assessment suggests the battery is being utilized under demanding conditions. This research introduces a new hybrid deep learning model that combines Convolutional Neural Networks (CNN), Long Short-Term Memory (LSTM), and Convolutional LSTM (ConvLstm) architectures. The predictive performance of the hybrid method is improved even further by utilizing Particle Swarm Optimization (PSO) for hyperparameter adjustment. Explainable AI (XAI) techniques, particularly SHapley Additive exPlanations (SHAP) values, are utilized to understand model predictions and determine the crucial factors influencing SoP. The predictive models underwent training and evaluation using a dataset from NASA, focusing on lithium-ion batteries. The results demonstrate that the hybrid model, incorporating PSO optimization, achieves exceptional accuracy. It yields a Mean Absolute Error (MAE) of 0.09, Mean Squared Error (MSE) of 0.03, Root Mean Squared Error (RMSE) of 0.17, and an R2 value of 98%. Comparing these measures to individual models without PSO, there has been a noticeable improvement. The model's transparency and dependability are increased by using SHAP values, which provide a deep knowledge of the importance of features like battery voltage and current. The study's findings demonstrate the potential for significantly improving SoP predictions in BMS with the use of complex deep learning models optimized using PSO. This improvement can increase battery operation safety, reliability, and efficiency. © 2025

关键词： Lithium-ion batteries

来源：评论

学校读者我要写书评

暂无评论

LLaVA-docent: Instruction tuning with multimodal large language model to support art appreciation education

arXiv

引用

arXiv 2024年

作者： Lee, Unggi Jeon, Minji Lee, Yunseo Byun, Gyuri Son, Yoorim Shin, Jaeyoon Ko, Hongkyu Kim, Hyeoncheol Department of Computer Science and Engineering Korea University Korea Republic of Teaching Learning and Teacher Education University of Nebraska-Lincoln United States Poongnap Elementary School Seoul Metropolitan Office of Education Korea Republic of Department of Education Seoul National University Korea Republic of Interdisciplinary Program in Art Education Art Education Major Seoul National University Korea Republic of Department of Elementary Art Education Seoul National University of Education Korea Republic of

Despite the development of various AI systems to support learning in various domains, AI assistance for art appreciation education has not been extensively explored. Art appreciation, often perceived as an unfamiliar and challenging endeavor for most students, can be more accessible with a generative AI enabled conversation partner that provides tailored questions and encourages the audience to deeply appreciate artwork. This study explores the application of multimodal large language models (MLLMs) in art appreciation education, with a focus on developing LLaVA-Docent, a model designed to serve as a personal tutor for art appreciation. Our approach involved design and development research, focusing on iterative enhancement to design and develop the application to produce a functional MLLM-enabled chatbot along with a data design framework for art appreciation education. To that end, we established a virtual dialogue dataset that was generated by GPT-4, which was instrumental in training our MLLM, LLaVA-Docent. The performance of LLaVA-Docent was evaluated by benchmarking it against alternative settings and revealed its distinct strengths and weaknesses. Our findings highlight the efficacy of the MMLM-based personalized art appreciation chatbot and demonstrate its applicability for a novel approach in which art appreciation is taught and experienced. © 2024, CC BY-NC-SA.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Analysis of Polysemy using Variance Values for Word Embeddings by BERT

Analysis of Polysemy using Variance Values for Word Embeddin...

引用

International Conference on Technologies and Applications of Artificial Intelligence (TAAI)

作者： Yanghuizi Ou Rui Cao Jing Bai Wen Ma Hiroyuki Shinnou Major in Computer and Information Sciences Graduate School of Science and Engineering Ibaraki University Hitachi JAPAN

ISBN: (数字)9781665403801

ISBN: (纸本)9781665447379

Word embeddings output by BERT depends on the context of the sentence in which the word appears. Therefore, we collect examples of word w and calculate their variance values from word w's embeddings obtained by BERT, and it is considered that the variance values correspond to the polysemy of word w. We collected examples of polysemous words (head)", "(meaning)", "(nuclear)", "(record)", "(language)", "(chest)") and monosemous words (production)", "(politics)", "(consciousness)", "(protest)", "(scores)") and compared the variance values. We expected that variance values for polysemous and monosemous words would be large and small, respectively. We also investigated which BERT position has the most influence on polysemy. However, we found that it is difficult to measure polysemy using the previous variance values, and we considered the cause of this difficulty.

关键词： Bit error rate Production Magnetic heads Artificial intelligence Nuclear measurements

来源：评论

学校读者我要写书评

暂无评论

Retrieval-augmented visual parcel invoice understanding transformer for address correction

引用

engineering Applications of Artificial Intelligence 2025年 158卷

作者： Yeon-Bi Jeong Hong-il Seo Yoo-Hwan Kim Won-Yeol Kim Department of Computer Information and Communication Engineering Chosun University Republic of Korea Department of Electronics and Electrical Engineering Interdisciplinary Major of Maritime AI Convergence Korea Maritime and Ocean University Republic of Korea Department of Computer Science Chosun University Republic of Korea Department of Artificial Intelligence Chosun University Republic of Korea

In automated postal and logistics operations, Visual Parcel Invoice Understanding technology (VPIU) addresses the challenging task of recognizing named entities such as addresses, names, and phone numbers. In particular, the VPIU performance is significantly reduced when the addresses are extracted. Therefore, we propose a novel Retrieval-augmented-VPIU model (Re-VPIU). The proposed Re-VPIU consists of a Document Understanding Transformer (DONUT) for information extraction and a Large Language Model (LLM) with Retrieval-Augmented Generation (RAG) for information correction. First, using an end-to-end approach, the DONUT extracts information regarding addresses, names, and phone numbers from the invoice images. Next, the RAG detects and corrects errors in the extracted address information according to the Korean address database and prompts. Proposed Re-VPIU model is the first application of RAG in the context of visual document understanding. Unlike traditional models, Re-VPIU does not require retraining to incorporate new address data, as it dynamically integrates external knowledge through retrieval. Furthermore, it enables both character- and word-level corrections simultaneously, leveraging the linguistic capabilities of LLMs. To validate the proposed method's effectiveness, several comparative experiments were conducted using a Korean invoice dataset. The experiments showed that the proposed method achieved an F1 Score of 83.42 % when evaluating the accuracy of the complete extraction, which was 12.17 % higher than that of the current state-of-the-art model.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Construction of Domain-Specific DistilBERT Model by Using Fine-Tuning

Construction of Domain-Specific DistilBERT Model by Using Fi...

引用

International Conference on Technologies and Applications of Artificial Intelligence (TAAI)

作者： Jing Bai Rui Cao Wen Ma Hiroyuki Shinnou Major in Computer and Information Sciences Graduate School of Science and Engineering Ibaraki University Hitachi Ibaraki JAPAN

ISBN: (数字)9781665403801

ISBN: (纸本)9781665447379

In this paper, we point out the problem that BERT is domain dependent, and propose to construct the domain specific pre-training DistilBERT model by using fine-tuning. In particular, parameters of a DistilBERT model are initialized using a trained BERT model, and then these parameters are tuned from the specific domain corpus. Consequently, we can construct efficiently the domain-specific DistilBERT model. In the experiment, we experiment the test set for each domain; the estimation of a masked word in a sentence. Through this experiment, we show the superiority of our proposed model (domain-specific DistilBERT model) by comparing it with the general BERT model.

关键词： Bit error rate Estimation Task analysis Artificial intelligence

来源：评论

学校读者我要写书评

暂无评论

Construction of document feature vectors using BERT

Construction of document feature vectors using BERT

引用

International Conference on Technologies and Applications of Artificial Intelligence (TAAI)

作者： Hirotaka Tanaka Rui Cao Jing Bai Wen Ma Hiroyuki Shinnou Major in Computer and Information Sciences Graduate school of Science and Engineering Ibaraki University Hitachi Ibaraki JAPAN

ISBN: (数字)9781665403801

ISBN: (纸本)9781665447379

In recent years, the performance of various natural language processing systems has been greatly improved by pre-training models such as Bidirectional Encoder Representations from Transformers (BERT). The BERT model obtains word em-beddings in context by the multi-head attention of a Transformer. In document-classification tasks, document feature vector can be constructed from the BERT output. However, the length of the BERT input sequence is upper-limited, which reduces the amount of information obtainable from BERT by the standard method. This limitation is problematic when handling long documents. Against this background, we propose a method that constructs document feature vector in BERT by embedding all words in long documents.

关键词： Bit error rate Natural language processing Task analysis Artificial intelligence Standards Context modeling

来源：评论

学校读者我要写书评

暂无评论

IDSH: An Improved Deep Supervised Hashing Method for Image Retrieval

引用

computer Modeling in engineering & Sciences 2019年第11期121卷 593-608页

作者： Chaowen Lu Feifei Lee Lei Chen Sheng Huang Qiu Chen School of Optical-Electrical and Computer Engineering University of Shanghai for Science and TechnologyShanghai200093China Major of Electrical Engineering and Electronics Graduate School of EngineeringKogakuin University1-24-2Nishi-shinjukuSinjuku-kuTokyo163-8677Japan

Image retrieval has become more and more important because of the explosive growth of images on the *** image retrieval methods have limited image retrieval performance due to the poor image expression abhility of visual feature and high dimension of *** is a widely-used method for Approximate Nearest Neighbor(ANN)search due to its rapidity and ***,Convolutional Neural Networks(CNNs)have strong discriminative characteristics which are used for image *** this paper,we propose a CNN architecture based on improved deep supervised hashing(IDSH)method,by which the binary compact codes can be generated *** main contributions of this paper are as follows:first,we add a Batch Normalization(BN)layer before each activation layer to prevent the gradient from vanishing and improve the training speed;secondly,we use Divide-and-Encode Module to map image features to approximate hash codes;finally,we adopt center loss to optimize *** experimental results on four large-scale datasets:MNIST,CIFAR-10,NUS-WIDE and SVHN demonstrate the effectiveness of the proposed method compared with other state-of-the-art hashing methods.

关键词： Image retrieval convolutional neural network hash functions center loss

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：