检索结果-内蒙古大学图书馆

arXiv 2025年

作者： Luo, Kairong Wen, Haodong Hu, Shengding Sun, Zhenbo Liu, Zhiyuan Sun, Maosong Lyu, Kaifeng Chen, Wenguang Department of Computer Science and Technology Tsinghua University China Qian Xuesen College Xi’an Jiaotong University China Simons Institute University of California Berkeley United States Peng Cheng Laboratory China

Training large models is both resource-intensive and time-consuming, making it crucial to understand the quantitative relationship between model performance and hyperparameters. In this paper, we present an empirical law that describes how the pretraining loss of large language models evolves under different learning rate schedules, such as constant, cosine, and step decay schedules. Our proposed law takes a multi-power form, combining a power law based on the sum of learning rates and additional power laws to account for a loss reduction effect induced by learning rate decay. We extensively validate this law on various model sizes and architectures, and demonstrate that after fitting on a few learning rate schedules, the law accurately predicts the loss curves for unseen schedules of different shapes and horizons. Moreover, by minimizing the predicted final pretraining loss across learning rate schedules, we are able to find a schedule that outperforms the widely used cosine learning rate schedule. Interestingly, this automatically discovered schedule bears some resemblance to the recently proposed Warmup-Stable-Decay (WSD) schedule (Hu et al., 2024) but achieves a slightly lower final loss. We believe these results could offer valuable insights for understanding the dynamics of pretraining and designing learning rate schedules to improve efficiency.1 © 2025, CC BY.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Enhancing Fintech Fraud Detection through Graph Convolutional Neural Networks

Enhancing Fintech Fraud Detection through Graph Convolutiona...

引用

Intelligent Systems and Computational Networks (ICISCN), International Conference on

作者： Pavan V Omprakash Omprakash Bhagya Prasad Bugge S Thangamani Anvesh Perada Sampathirao Suneetha Manipal Law School Manipal Academy of Higher Education India School of Law Jain University Bengaluru India Department of Electronics and Communication Engineering S.R.K.R Engineering College Bhimavaram India Department of Information Technology Nandha Engineering College Erode India Department of Electrical and Computer Engineering Drexel University Philadelphia Pennsylvania USA Department of Computer Science and Engineering Koneru Lakshmaiah Education Foundation Vijayawada India

ISBN: (数字)9798331529246

ISBN: (纸本)9798331529253

The swift advancement of fintech has transformed financial services, offering users seamless digital experiences. Nonetheless, this expansion has also presented intricate cybersecurity difficulties, especially in fraud detection. This research investigates novel methods for addressing fintech fraud through the utilization of GCNN and the ResNet101-C model. GCNN adeptly addresses dynamic fraud trends via graph-based feature aggregation, whereas ResNet101-C, augmented with dropout and fully connected layers, attains enhanced classification performance. Preprocessing procedures, such as imputation, noise reduction, and feature standardization, enhance data quality, whereas feature extraction emphasizes essential transaction variables like frequency and value. Assessment of synthetic datasets reveals that ResNet101-C achieves an accuracy of 93.46%, surpassing leading models. These results underscore the transformational capability of GCNN and ResNet101-C in tackling the dynamic nature of fraud, providing scalable, precise, and privacy-preserving solutions for the financial sector.

关键词： Training Industries Accuracy Federated learning Fintech Data preprocessing Feature extraction Fraud Convolutional neural networks Synthetic data

来源：评论

学校读者我要写书评

暂无评论

Deep learning models for early and accurate diagnosis of ventilator-associated pneumonia in mechanically ventilated neonates

引用

computers in Biology and Medicine 2025年 189卷 109942-109942页

作者： Hsu, Jen-Fu Lin, Ying-Chih Lin, Chun-Yuan Chu, Shih-Ming Cheng, Hui-Jun Xu, Fan-Wei Huang, Hsuan-Rong Liao, Chen-Chu Fu, Rei-Huei Tsai, Ming-Horng Division of Neonatology and Pediatric Hematology/Oncology Department of Pediatrics Chang Gung Memorial Hospital Yunlin Taiwan Division of Pediatric Neonatology Department of Pediatrics Chang Gung Memorial Hospital Taoyuan Taiwan College of Medicine Chang Gung University Taoyuan Taiwan Department of Computer Science and Information Engineering Asia University Taichung Taiwan Department of Applied Mathematics Feng Chia University Taichung Taiwan Department of Artificial Intelligence Application Minth University of Science and Technology HsinChu County Taiwan

Background: Early and accurate confirmation of critically ill neonates with a suspected diagnosis of ventilator-associated pneumonia (VAP) can optimize the therapeutic strategy and avoid unnecessary use of empirical antibiotics. We aimed to examine whether deep learning (DL) methods can assist the diagnosis of VAP of intubated neonates in the neonatal intensive care unit (NICU). Methods: A total of 670 neonates with mechanical ventilation were prospectively observed in a tertiary-level NICU in Taiwan between October 2017 and March 2022, during which image data were collected. All neonates with clinically suspected VAP were enrolled, and various DL methods were used to test the prediction ability of VAP diagnosis. The accuracy, precision, sensitivity, specificity, F1-score, and area under curves (AUCs) of several DL methods were compared. Results: A total of 900 chest X-ray images derived from 670 neonates with VAP and/or bronchopulmonary dysplasia (BPD) were enrolled, including 399 images from patients with definite diagnosis of VAP based on the strict criteria and 501 images from neonates without VAP. Compared with conventional DNN models such as ResNet, VGG, DenseNet, the RegNetX80 achieved the best specificity of 0.8378, which facilitates a low false positive rate. For accurate diagnosis of neonatal VAP, a combinatorial model of ResNet50 and RegNetX80, created through ensemble learning, further enhanced the AUC to 0.8023 for neonates with VAP on mechanical ventilation. In addition, the consistent XAI results in the left-lower region of chest X-ray image provided informative feedback and increased confidence to AI-assisted doctors. Conclusions: Deep learning methods are applicable with good predictive accuracy using chest X-ray images to help diagnosis of VAP in the NICU, which can help clinicians make decisions regarding the choices of empiric antibiotics for critically ill neonates. Future prospective trials are warranted to document its clinical usefulness and

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Real-Time Blood Pressure Monitoring and Hypertension Prediction based on Multitask Temporal Convolutional Network

Real-Time Blood Pressure Monitoring and Hypertension Predict...

引用

Intelligent Systems and Computational Networks (ICISCN), International Conference on

作者： Muntather Musnmusawi R. Deepa Smitha Sharath Shankar Nirupashri G P. Katheeja Khanam Department of Computers Techniques Engineering College of Technical Engineering The Islamic University of Al Diwaniyah Al Diwaniyah Iraq Department of Information Technology St.Joseph's Institute of Technology Chennai India Department of Electronics and Communication Engineering Nitte Meenakshi Institute of Technology Bengaluru India Department of Bio Medical Engineering Dhanalakshmi Srinivasan College of Engineering and Technology India Department of Computer Science and Engineering Mohan Babu University Tirupati India

ISBN: (数字)9798331529246

ISBN: (纸本)9798331529253

In recent days, blood pressure monitoring and hypertension prediction have become essential for effective measurement and management. Continuously tracking blood pressure helps in making informed decisions like changing medication or dosage. However, some challenges are raised while monitoring such as blood pressure readings being prone to errors, variability in measurements if a patient moves and environmental noise. Therefore, this research proposes Multitask Temporal Convolutional Network (MTCN) for realtime blood pressure monitoring and hypertension prediction. Initially, data is collected from Medical information Mart for Intensive Care III (MIMIC-III) and preprocessed by using Min- Max Normalization which rescales numeric data to definite range. After that, features like time frequency coefficients, power spectral density, Scalogram features are extracted by using Continuous Wavelet Transform (CWT). Finally, blood pressure monitoring and hypertension prediction is done by using proposed MTCN which provides accurate hypertension prediction and personalized risk assessments. The proposed MTCN achieved better results in terms of Mean Absolute Error (MAE) of 2.01, Root Mean Square Error (RMSE) of 3.26 when compared with existing Random Forest Regression model.

关键词： Hypertension Time-frequency analysis Continuous wavelet transforms Feature extraction Real-time systems Pressure measurement Convolutional neural networks Risk management Biomedical monitoring Monitoring

来源：评论

学校读者我要写书评

暂无评论

An Enhanced SVM Model for Eye Cancer Diagnosis Using Image Feature 3

An Enhanced SVM Model for Eye Cancer Diagnosis Using Image F...

引用

3rd IEEE International Conference on Automation and Computation, AUTOCOM 2025

作者： Panigrahi, Lakshmisree Khanna, Lalit Vinay Kumar, S.B. Agarwal, Ankita Pushpalatha, T. Mahajan, Shriya Department of Computer Science and Engineering Odisha Bhubaneswar India Chitkara Centre for Research and Development Chitkara University Himachal Pradesh Baddi174103 India Karnataka Bangalore India Maharishi School of Engineering & Technology Maharishi University of Information Technology Uttar Pradesh India Bengaluru India Centre of Research Impact and Outcome Chitkara University Punjab Rajpura140417 India

ISBN: (纸本)9798331542375

Eye cancer., also called ocular cancer., affects the eye and its structure. As with other conditions., early detection and correct diagnosis are key to effective treatment;however., this can be difficult for healthcare professionals. It makes diagnosis time-consuming and a bit subjective as well. Hence., a routine and diagnostic tool is required. This paper presents an improved SVM (Support Vector Machine) model for image feature-based eye cancer diagnosis. In this model., eye images are first processed to extract features., and then these extracted features will be classified into cancerous or noncancerous using SVM. Additionally, a feature selection algorithm is incorporated into this model to detect essential features that help further curtail the data while increasing classification performance. We tested our model on 500 images and achieved an accuracy of - 92.0%. Besides, we have also compared our model with other state-of-the-art methods and achieved better accuracy, sensitivity, and specificity. Conclusions: In summary, we found that our proposed improved SVM model of eye cancer diagnosis did good results, which are acceptable and may be useful to medical doctors for a valid early detection of eye cancer. And this can, in turn, lead to improved treatment results for patients. © 2025 IEEE.

关键词： Diseases

来源：评论

学校读者我要写书评

暂无评论

An Optimized Deep Learning Framework for Semantic Interpretation Of Complex Single-Line Diagram Imagery

SSRN

引用

SSRN 2025年

作者： Bhanbhro, Hina Yew, Kwang Hooi Kusakunniran, Worapan Bin Zakaria, Mohammad Nordin Hashmi, Syed Abdul Moiz Amur, Zaira Hassan Memon, Vengas Computer and Information Science Department Universiti Teknologi Petronas Seri Iskandar Perak Darul Ridzuan Malaysia Faculty of Information and Communication Technology Mahidol University Nakhon Pathom Thailand Chemical Engineering Department Universiti Teknologi Petronas Seri Iskandar Perak Darul Ridzuan Malaysia

Accurate interpretation of single-line diagrams (SLDs) is crucial for analyzing electrical systems, as they encapsulate vital information about operational safety and efficiency in a simplified format. Traditional SLD processing methods rely on manual inspection and basic image analysis, which are computationally intensive, error-prone, and require extensive preprocessing. Although deep learning has been applied to symbol classification, existing models often fail to capture fine-grained symbol details, leading to misclassification. To address these limitations, this study proposes a hybrid deep learning-based symbol classification method. A newly created dataset was benchmarked using state-of-the-art deep learning models, and an optimal model was systematically designed, developed, and tested. The proposed approach integrates a Hybrid Residual Attention Module (HRAM) to enhance the model’s ability to identify fine-grained symbol details and a Proximity-aware Loss Function to improve performance in cluttered regions by penalizing missed classifications based on the spatial proximity of neighboring symbols. These innovations result in an optimized method for semantic processing in symbol classification tasks. The proposed model achieves 92.5% classification accuracy, a 3.8% improvement over the top-performing baseline, alongside a 19.6% reduction in model parameters. These advancements contribute to more efficient and accurate semantic processing of SLDs, paving the way for improved analysis of electrical system diagrams. © 2025, The Authors. All rights reserved.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

Link Adaptation Techniques for Massive MIMO Communication Systems 3

Link Adaptation Techniques for Massive MIMO Communication Sy...

引用

3rd IEEE International Conference on Automation and Computation, AUTOCOM 2025

作者： Das, Prativa Dhingra, Lovish Traisa, Roopa Sahoo, Girija Shankar Maddikayala, Veera Nagaiah Sharma, Divya Department of Computer Science and Engineering Odisha Bhubaneswar India Chitkara Centre for Research and Development Chitkara University Himachal Pradesh Baddi174103 India Karnataka Bangalore India Maharishi School of Engineering & Technology Maharishi University of Information Technology Uttar Pradesh India Bengaluru India Centre of Research Impact and Outcome Chitkara University Punjab Rajpura140417 India

ISBN: (纸本)9798331542375

Huge MIMO communication structures are a promising technology for future Wi-Fi networks that can obtain significantly higher information costs and progress spectral efficiency in comparison to traditional structures. But, to completely take advantage of the ability of today's big MIMO, green and reliable hyperlink version strategies are required. Link model refers to the choice of trendy suitable transmission parameters, consisting of modulation and coding scheme, to evolve to changing channel conditions, a good way to optimize the machine's overall performance. In large MIMO structures, the massive number of trendy antennas at both the transmitter and receiver permits extra accurate channel estimation, which can be applied for greater superior hyperlink edition strategies. This technical summary provides a complete evaluation of the link model strategies for large MIMO systems. These techniques include preceding-based totally adaptive modulation and coding, adaptive beam forming, and closed-loop spatial multiplexing. The advantages and demanding situations of cutting-edge approaches are mentioned, and their overall performance is evaluated using theoretical analysis and simulation consequences. Furthermore, the impact cutting-edge large MIMO channel traits, along with correlation, fading, and interference, at the effectiveness of these strategies are likewise investigated. This research aims to offer insights into the potential modern-day hyperlink model for huge MIMO structures and guide the layout of cutting-edge green and reliable hyperlink model algorithms for destiny Wi-Fi networks. © 2025 IEEE.

关键词： Internet

来源：评论

学校读者我要写书评

暂无评论

Parallel Selective Meta-Heuristic Algorithms for Real-Time Non-Convex Optimization in Big Data Analytics 3

Parallel Selective Meta-Heuristic Algorithms for Real-Time N...

引用

3rd IEEE International Conference on Automation and Computation, AUTOCOM 2025

作者： Ghosal, Shreya Shankar, Amar Agarwal, Trapty Suchithra, R. Raj, Nimesh Hansda, Prity Chitkara Centre for Research and Development Chitkara University Himachal Pradesh Baddi174103 India Karnataka Bangalore India Maharishi School of Engineering & Technology Maharishi University of Information Technology Uttar Pradesh India Bengaluru India Centre of Research Impact and Outcome Chitkara University Punjab Rajpura140417 India Department of Computer Science and Engineering Odisha Bhubaneswar India

ISBN: (纸本)9798331542375

With big data in modern analytics becoming more common, a lot of it is real-time optimization, which is an essential problem. This is because big data requires efficient and accurate optimization methods, leading to developing parallel selective viable meta-heuristic algorithms for nonconvex optimization in large analytics. They can be used to find optimal solutions rapidly by merging the complementary advantages of meta-heuristics and parallel computing. Algorithms do this using an optimization process, but not always for the whole data set used, saving computational performance. As information is updated in real-time, its approach will constantly be improved, resulting in a better and more accurate solution. A parallel implementation of these algorithms makes them effective for real-time optimization using large datasets. However, the performance of big data analytics may be significantly improved by implementing these algorithms because they have achieved satisfactory results in a number of applications, including machine learning, data mining, and pattern recognition. © 2025 IEEE.

关键词： Large datasets

来源：评论

学校读者我要写书评

暂无评论

EAPD Diagnosis Diagnosis Using AI

EAPD Diagnosis Diagnosis Using AI

引用

Intelligent Control, Computing and Communications (IC3), International Conference on

作者： P. Sruthi Jajimoggala Sravanthi Renukadevi. M Zayd Balasse Shravanthi Mantra Prithvi Raj B. R. Supreeth Department of CSE(AI&ML) CMR College of Engineering & Technology Hyderabad Telangana India Department of CSE Gokaraju Rangaraju Institute of Engineering and Technology Hyderabad Telangana India Department of Computer Science Engineering Sethu Institute of Technology Kariapatti Tamil Nadu India Department of computers Techniques engineering College of technical engineering The Islamic University Najaf Iraq ENT Saveetha Medical College and Hospital Saveetha Institute of Medical and Technical Sciences Saveetha University Chennai Tamil Nadu India Department of Computer Science Oxford Degree and PG College Sir. M. Visveshwaraiah Layout Bangalore

ISBN: (数字)9798331527495

ISBN: (纸本)9798331527501

The motivation behind this undertaking is to take a gander at the capability of computerized reasoning (A1) in giving dependable and fast diagnostics to pediatric problems. Pediatric disease determination accurately and speedily are basic for effective therapy and further developed wellbeing outcomes Nevertheless, the intricacies of these signs, as well as the shortage of pediatric specialists, make ID difficult Al has shown fantastic outcomes in different clinical fields, including assessment, treatment, and medicine find This study examines flow progresses in involving Al strategies for pediatric sickness location, evaluating their exactness and efficacy This examination accentuates the limitations and cutoff points of Al techniques, stressing the prerequisite for more innovative work to work on their accuracy and constancy, The exploration contains cases demonstrating the viability of Al in the analysis of pediatric afflictions like leukemia, mental imbalance, and flu. The discoveries propose that Al strategies have a likelihood to further develop finding exactness and immediately, limit botches, and empower brief treatment for pediatric sicknesses. In any case, further review is expected to evaluate the appropriateness of Al methods in certifiable medical services conditions while handling issues like as information quality, interpretability, and moral worries. The paper goes on by proposing possible turn of events and examination ways for Al techniques for the specific finding of pediatric infections, underlining the significance of interdisciplinary coordinated effort and moral issues.

关键词： Ethics Reviews Medical treatment Leukemia Cognition Artificial intelligence Medical diagnostic imaging Intelligent control information integrity Diseases

来源：评论

学校读者我要写书评

暂无评论

Optimizing CNN Architectures for Early Detection of Lung Cancer Using CT Images 3

Optimizing CNN Architectures for Early Detection of Lung Can...

引用

3rd IEEE International Conference on Automation and Computation, AUTOCOM 2025

作者： Devi, Swagatika Swarup, Lakshya Sahana, B.S. Naval, Preeti Jaswani, Neha Singh, Jaskirat Department of Computer Science and Engineering Odisha Bhubaneswar India Chitkara Centre for Research and Development Chitkara University Himachal Pradesh Baddi174103 India Karnataka Bangalore India Maharishi School of Engineering & Technology Maharishi University of Information Technology Uttar Pradesh India Bengaluru India Centre of Research Impact and Outcome Chitkara University Punjab Rajpura140417 India

ISBN: (纸本)9798331542375

Lung cancer remains a primary cause of cancer-related deaths globally, and early diagnosis is crucial to improve patient outcomes. Whole-body computed tomography (CT) imaging holds promise as a diagnostic tool for the early detection of lung cancer. Still, malignancies' highly variable and spatially complex properties require refined image analysis techniques. Although convolutional neural networks (CNNs) have become great tools for image analysis, performance relies highly on network architecture design. This project focuses on CNN architecture optimization for lung cancer detection from CT scan images at a very early stage. We use this method to train and evaluate different CNN architectures on a big dataset of lung CT images. Architecture [(TODO) link] Add dropout, experiments with network designs and hyperparameters (layer size, filter sizes & activations ReLu or PReLU Conclusions [same recommendations?] (Previous fin data discussion)] ((Same records for whatever it's worth)) We will use data augmentation to try and make our models a bit more general. Once we get the best CNN model, we can assess its capability to identify lung cancer images earlier than other state-of-the-art methods and present our results by comparing them with them on a separate testing dataset. The project can potentially improve diagnostic tests for early lung cancer detection, thus improving patient survival. © 2025 IEEE.

关键词： Diseases

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：