检索结果-内蒙古大学图书馆

arXiv 2025年

作者： Lv, Peizhuo Xiahou, Yiran Li, Congyi Sun, Mengjie Zhang, Shengzhi Chen, Kai Zhang, Yingjun Institute of Information Engineering Chinese Academy of Sciences China Department of Computer Science Metropolitan College Boston University United States Institute of Software Chinese Academy of Sciences China

LoRA (Low-Rank Adaptation) has achieved remarkable success in the parameter-efficient fine-tuning of large models. The trained LoRA matrix can be integrated with the base model through addition or negation operation to improve performance on downstream tasks. However, the unauthorized use of LoRAs to generate harmful content highlights the need for effective mechanisms to trace their usage. A natural solution is to embed watermarks into LoRAs to detect unauthorized misuse. However, existing methods struggle when multiple LoRAs are combined or negation operation is applied, as these can significantly degrade watermark performance. In this paper, we introduce LoRAGuard, a novel black-box watermarking technique for detecting unauthorized misuse of LoRAs. To support both addition and negation operations, we propose the Yin-Yang watermark technique, where the Yin watermark is verified during negation operation and the Yang watermark during addition operation. Additionally, we propose a shadow-model-based watermark training approach that significantly improves effectiveness in scenarios involving multiple integrated LoRAs. Extensive experiments on both language and diffusion models show that LoRAGuard achieves nearly 100% watermark verification success and demonstrates strong effectiveness. Copyright © 2025, The Authors. All rights reserved.

关键词： Watermarking

来源：评论

学校读者我要写书评

暂无评论

Deep Learning-Based Image Analysis for Detecting Nutrient Deficiencies of Tomato Plants

Deep Learning-Based Image Analysis for Detecting Nutrient De...

引用

Advancements in Computing (ICAC), International Conference on

作者： K. C. Nirasha Jayasiri Sanjeevi Chandrasiri Shalini Rupasinghe K.O. Rivindu Karunanayake Abineshh Uthayachandran M.I. Fathima Zihara Dept. of Computer Science and Software Engineering Sri Lanka Institute of Information Technology Sri Lanka

Tomato is one of the most popular crops worldwide. The success of a tomato crop is highly dependent on the health of the plants. Nutrient deficiency surveillance is typically conducted through visual inspections, which can be challenging for inexperienced farmers and home gardeners to work accurately. There is a growing need for a quick, reliable, and accurate nutrient deficiency identification system to address this issue. The proposed method is based on image processing and deep learning technologies that are highly effective for image classification tasks. Two models were trained using a dataset collected from tomato plants in Sri Lanka and evaluated using Mask Region-based Convolutional Neural Network (Mask R-CNN) and, You Only Look Once (YOLO) for deficiency classification, and results obtained 92% and 98% accuracy, respectively. Deficiency dispersion level expressed as a percentage using Mask R-CNN and followed by image processing techniques. Overall, this proposed system offers a convenient and accessible tool for farmers and home gardeners to monitor and maintain the health of their tomato plants, enabling them to achieve optimal yields and ensure profitable returns.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Beyond pixels:Learning from multimodal hyperspectral superpixels for land cover classification

引用

science China(Technological sciences) 2022年第4期65卷 802-808页

作者： HONG DanFeng WU Xin YAO Jing ZHU XiaoXiang Key Laboratory of Computational Optical Imaging Technology Aerospace Information Research InstituteChinese Academy of SciencesBeijing 100094China School of Computer Science(National Pilot Software Engineering School) Beijing University of Posts and TelecommunicationsBeijing 100876China Remote Sensing Technology Institute German Aerospace Center(DLR)Wessling 82234Germany Data Science in Earth Observation Technical University of MunichMunich 80333Germany

Despite tons of advanced classification models that have recently been developed for the land cover mapping task,the monotonicity of a single remote sensing data source,such as only using hyperspectral data or multispectral data,hinders the classification accuracy from being further improved and tends to meet the performance *** this reason,we develop a novel superpixel-based subspace learning model,called Supace,by jointly learning multimodal feature representations from HS and MS superpixels for more accurate LCC *** can learn a common subspace across multimodal RS data,where the diverse and complementary information from different modalities can be better combined,being capable of enhancing the discriminative ability of to-be-learned features in a more effective *** better capture semantic information of objects in the feature learning process,superpixels that beyond pixels are regarded as the study object in our Supace for *** experiments have been conducted on two popular hyperspectral and multispectral datasets,demonstrating the superiority of the proposed Supace in the land cover classification task compared with several well-known baselines related to multimodal remote sensing image feature learning.

关键词： classification hyperspectral image land cover multimodal multispectral image remote sensing subspace learning superpixels

来源：评论

学校读者我要写书评

暂无评论

Smart System to Support Hearing Impaired Students in Tamil

Smart System to Support Hearing Impaired Students in Tamil

引用

2023 International Research Conference on Smart Computing and Systems engineering, SCSE 2023

作者： Thahseen, Ahamed Tissera, Wishalya Vidhanaarachchi, Samitha Aaron, N. Rajapaksha, D.C.N. Fernando, P.V. Dias, Thisuru Fernando, Thatamathy Sri Lanka Institute of Information Technology New Kandy Road Department of Information Technology Malabe Sri Lanka Sri Lanka Institute of Information Technology New Kandy Road Department of Computer Science & Software Engineering Malabe Sri Lanka YMCA Young Men's Christian Association 26/4A Boundary Road South Batticaloa Sri Lanka

ISBN: (纸本)9798350341454

This research paper introduces a groundbreaking smart system designed to assist hearing-impaired students in comprehending spoken Tamil, the second language in Sri Lanka. The system addresses the challenges faced by these students by incorporating cutting-edge deep learning techniques, including Convolutional Neural Networks, Hidden Markov Models, and Recurrent Neural Networks, for efficient feature extraction and sequence modeling within an Automatic Speech Recognition framework. Additionally, the paper proposes a novel method for automatically recognizing Tamil Sign Language gestures using advanced hand gesture recognition algorithms and a comprehensive dataset of Tamil Sign Language. The system encompasses four primary classification approaches, enabling the conversion of Tamil Sign Language to Text. Text to Tamil Sign Language, lip reading to Tamil Sign Language, Normal voice to Tamil Sign Language -Sign Language to Normal Voice and physical object identification to both text and Tamil Sign Language. Notably, the system achieves remarkable results, boasting an impressive accuracy rate of 0.99% surpassing existing Automatic Speech Recognition and Text-to-Speech systems. This significant breakthrough holds immense potential in enhancing the learning experience of hearing-impaired students in Tamil-speaking regions. Furthermore, the system's adaptability allows for future expansion to support additional languages, making it highly versatile for diverse educational and communication settings. © 2023 IEEE.

关键词： Students

来源：评论

学校读者我要写书评

暂无评论

Smart System to Optimize Organic Crop Rotation Using Precision Agriculture Data

Smart System to Optimize Organic Crop Rotation Using Precisi...

引用

International Symposium on Multidisciplinary Studies and Innovative Technologies (ISMSIT)

作者： Ransinghe R.A.D. M Samarakoon S.M.A. D Waththaladeniya N. M Ihalagedara I.H.U. B Udara Srimath S. Samaratunge Arachchilllage N. Kodagoda Department of Computer Science and Software Engineering Sri Lanka Institute of Information Technology Malabe Sri Lanka

Optimizing organic crop rotation and effectively managing pests and diseases remain a significant challenge for farmers. In this research, we propose a Smart System to Optimize Organic Crop Rotation using Precision Agriculture Data. The system leverages Internet of Things (IoT) technology, machine learning (ML) algorithms, and cloud computing to enhance decision-making and improve productivity in organic farming. The proposed system integrates multiple components to provide farmers with informed decisions for optimized crop rotation. Soil sensors collect data on soil health, which is then analyzed using a multi-objective optimization technique to determine the best crop for a given soil sample. Real-time weather data is incorporated to enable climate-resilient farming practices and help farmers make educated choices in selecting crops and implementing rotation plans. Additionally, a cloud computing-based model is developed for pest and disease identification, providing farmers with effective solutions to combat these challenges in organic methodologies. By leveraging IoT, ML, and cloud computing, our system proposes farmers a more efficient and effective approach to managing their crops. The system provides real-time data-driven recommendations on fertilizer selection and crop rotation, leading to improved crop growth, increased yields, and reduced environmental impact. Moreover, by adopting organic farming practices, the system contributes to the sustainable development of the agricultural sector.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Effectiveness of Deep Learning Technologies to Assist Educational Inclusion for Children with Down Syndrome

Effectiveness of Deep Learning Technologies to Assist Educat...

引用

Advancements in Computing (ICAC), International Conference on

作者： Shehan Silva Anjula Jayasinghe Rishitha Dilshan Chandur Dissanayake Samitha Vidhanaarachchi Dilshan De Silva Department of Computer Science and Software Engineering Sri Lanka Institute of Information Technology Malabe Sri Lanka

Down Syndrome (DS) is a genetic disorder causing intellectual disability and developmental delays. Despite instances of discrimination, several individuals with DS have achieved success through proper education and community support. This research addresses the challenges faced by children with DS in terms of motor skill development, cognitive impairments, and speech difficulties with the use of Deep Learning Technologies such as LSTM, CNN machine learning models. To bridge the educational gap and improve speech intelligibility, this study focuses on oral motor exercises that enhance articulation skills. Cognitive development in individuals with DS is also a concern, prompting the proposal of a self-learning system to foster cognitive growth. Fine motor skills improvement is explored through yoga exercises, and interventions like interactive technologies are considered to enhance learning experiences. This paper discusses the development of a user-friendly mobile application for early intervention in DS, integrating interactive activities using advanced technologies and a parent therapy dashboard. The system's potential for improving developmental skills and future expansion is emphasized.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Predictive Modeling for Early Identification of Disease Severity in Acute Respiratory Infections: A Case Study with COVID-19

Predictive Modeling for Early Identification of Disease Seve...

引用

Advancements in Computing (ICAC), International Conference on

作者： Vindhya Kalapuge Dharshana Kasthurirathna Department of Computer Science & Software Engineering Sri Lanka Institute of Information Technology Malabe Sri Lanka

The COVID-19 pandemic, declared a global emergency by WHO in March 2020, has had a profound impact on global health. With millions affected, the pandemic's magnitude became evident. Even advanced healthcare systems grappled with resource shortages, from protective gear to ventilators. Late diagnosis and the inability to adapt to new SARS-CoV-2 variants led to misdiagnoses and high casualties. Overburdened healthcare facilities and the absence of a centralized patient data system hindered swift decision-making. The lack of real-time monitoring meant many didn't receive timely, personalized care, underscoring the need for an integrated healthcare response during global emergencies. To tackle these challenges and improve hospital resource management during pandemics, this study presents a web approach harnessing big data analytics to develop a machine learning predictive model. This model has been developed to enable hospitals to assess patients' severity in real-time, taking into account a wide range of factors such as symptoms, medical history, and demographics. To acquire data regularly, a streaming data pipeline was established using Apache Kafka and Apache Streaming. In predicting the severity levels of patients, both Multinomial Logistic Regression and Multilayer Perceptron (MLP) models were employed. One noteworthy finding was that using datasets with balanced classes yielded significantly higher accuracy for the MLP model compared to unbalanced datasets. Specifically, MLP achieved the highest accuracy of 60.4% when trained on balanced classes. This model's real-time severity insights have the potential to revolutionize decision-making and resource allocation in healthcare. It represents a significant step towards a more effective response to future pandemics and the strengthening of the global healthcare system.

关键词：

来源：评论

学校读者我要写书评

暂无评论

VC-IDS: An Ensemble Learning Method based on Voting Classifier for Intrusion Detection System using Machine Learning Algorithms 2

VC-IDS: An Ensemble Learning Method based on Voting Classifi...

引用

2nd International Conference on information and Communication Technology for Sustainable Development, ICICT4SD 2023

作者： Palash, Md Istakiak Adnan Ulla, Sana Islam, A. K.M. Mohaiminul Mridha, Sabbir Ahamed Al Nahian, Md Jaber Kaiser, M. Shamim Bangladesh University of Professionals Department of ICT Dhaka Bangladesh Daffodil International University Department of Software Engineering Dhaka Bangladesh University of Calgary Department of Computer Science AB Canada Institute of Information Technology Jahangirnagar University Dhaka Bangladesh

ISBN: (纸本)9798350358667

The number of illegal intruders is growing at the same rate as the number of computer networks and the amount of data on those networks. They try to steal important and private information from the network about its users. An intrusion detection system (IDS) blocks all malicious or unauthorized activity from a computer network. However, it is important to detect illegal activity as early as possible;otherwise, attackers might access all the resources of the system. So, an effective and efficient IDS can be very useful for that. In this paper, we propose an ensemble learning method based on a voting classifier technique for detecting intrusion into a system. We have used the NSL-KDD dataset for our study. First, we preprocess the dataset. Then we implemented different machine learning algorithms. Based on the results, the Extra Trees Classifier, Decision Tree, and Random Forest algorithms achieved the highest accuracy scores of 98.68%, 98.58%, and 98.70% respectively. In the next step, we applied voting classifier techniques using these top three models, and we achieved an accuracy of 98.73% which indicates that this method can effectively increase the performance of the models. Finally, we have evaluated our models using different performance evaluation metrics. © 2023 IEEE.

关键词： Intrusion detection

来源：评论

学校读者我要写书评

暂无评论

Deep learning-based multi-head self-attention model for human epilepsy identification from EEG signal for biomedical traits

引用

Multimedia Tools and Applications 2024年第33期83卷 80201-80223页

作者： Dutta, Ashit Kumar Raparthi, Mohan Alsaadi, Mahmood Bhatt, Mohammed Wasim Dodda, Sarath Babu Prashant, G.C. Sandhu, Mukta Patni, Jagdish Chandra Department of Computer Science and Information Systems College of Applied Sciences AlMaarefa University Ad Diriyah Riyadh13713 Saudi Arabia Alphabet Life Science DallasTX7506 United States Department of Computer Science Al-Maarif University College Al Anbar31001 Iraq Model Institute of Engineering and Technology J&K Jammu India Software Engineer Central Michigan University DallasTX75009 United States Department of Computer Science Texas Tech University Lubbock79409 United States Skill Faculty of Engineering & amp Technology Shri Vishwakarma Skill University Haryana Palwal India CSE School of advance Computing Alliance University Bengaluru India

The neurological condition epilepsy is demanding and even fatal. Electroencephalogram (EEG)-based epilepsy detection still faces various difficulties. EEG readings fluctuate, and different patients have various seizure activity patterns. EEG signal detection is time-consuming and labour-intensive, which puts a strain on medical staff and raises the possibility of erroneous detections. Usually, electrodes are inserted into the scalp or inside the brain for a brief period of time in order to obtain EEG data. It is essential to research efficient cross-patient automatic epilepsy detection techniques. The multi-head self-attention mechanism recognises long-distance dependencies with the same proficiency as it does temporal dynamic correlations between short-term temporal pattern characteristics and sequential relationships. The contextual representations are inputted into a bidirectional long short-term memory network (BiLSTM) so that information can be extracted in both directions. Classification and training are carried out utilising the log SoftMax algorithm. The experiments utilised scalp EEG data from the CHB-MIT database. Sensitivity, specificity, F1-score, and accuracy were computed to be 96.5 percent, 97.04 percent, 96.6 percent, and 96.2 percent, respectively. The results of the experiment show how well the technique works for detecting seizures in several patients utilising multi-channel EEG recordings. The results also demonstrate the method's improved generalisation capabilities and resilience and consistency in collecting seizure patterns. This is particularly critical for the tertiary diagnosis of epilepsy, and the findings indicate that the proposed method significantly improves the accuracy of detection. © The Author(s), under exclusive licence to Springer science+Business Media, LLC, part of Springer Nature 2024.

关键词： Electroencephalography

来源：评论

学校读者我要写书评

暂无评论

Impact of COVID-19 on Predicting 2020 US Presidential Elections on Social Media 16th

Impact of COVID-19 on Predicting 2020 US Presidential Electi...

引用

16th International Conference on information Technology and Applications, ICITA 2022

作者： Khan, Asif Zhang, Huaping Boudjellal, Nada Hayat, Bashir Dai, Lin Ahmad, Arshad Al-Hamed, Ahmed School of Computer Science and Technology Beijing Institute of Technology Beijing100081 China The Faculty of New Information and Communication Technologies University Abdelhamid Mehri Constantine 2 Constantine25000 Algeria Institute of Management Sciences Peshawar Peshawar Pakistan Institute of Software Systems Engineering Johannes Kepler University Linz4040 Austria Department of IT and Computer Science Pak-Austria Fachhochschule: Institute of Applied Sciences and Technology Mang Khanpur Road Haripur22620 Pakistan

ISBN: (纸本)9789811993305

By the beginning of 2020, the world woke up to a global pandemic that changed people’s everyday lives and restrained their physical contact. During those times Social Media Platforms (SMPs) were almost the only mean of individual-to-individual and government-to-individuals communications. Therefore, people’s opinions were more expressed on SM. On the other hand, election candidates used SM to promote themselves and engage with voters. In this study, we investigate how COVID-19 affected voters’ opinions through the months of the US presidential campaign and eventually predict the 2020 US Presidential Election results using Twitter’s data. Mainly two types of experiments were conducted and compared;(i) transformer-based, and (ii) rule-based sentiment analysis (SA). In addition, vote shares for the presidential candidates using both approaches were predicted. The results show that the rule-based approach nearly predicts the right winner, Joe Biden with MAE 2.1, outperforming the predicted results from CNBC, Economist/YouGov, and transformer-based (BERTweet) approach, except for RCP (MAE 1.55). © 2023, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： COVID-19

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：