检索结果-内蒙古大学图书馆

Pruning techniques for artificial intelligence networks: a deeper look at their engineering design and bias: the first review of its kind

引用

Multimedia Tools and Applications 2025年第11期84卷 9591-9665页

作者： Mohanty, Lopamudra Kumar, Ashish Mehta, Vivek Agarwal, Mohit Suri, Jasjit S. School of Computer Science Engineering and Technology Bennett University Gr. Noida India Department of Computer Science ABES Engineering College UP Ghaziabad India Stroke Diagnostic and Monitoring Division AtheroPoint™ RosevilleCA United States

Trained Artificial Intelligence (AI) models are challenging to install on edge devices as they are low in memory and computational power. Pruned AI (PAI) models are therefore needed with minimal degradation in performance. This is the first study of its kind where we have characterized three types of PAI paradigms, namely, weight pruning (WP), channel pruning (CP), and hybrid pruning (HP). Further, the study also analyses the risk of bias (RoB) in PAI models, and explainable PAI (XPAI) networks. PRISMA model was adapted for the selection of the best 145 studies. The review had two hypotheses: (i) PAI had bias and (ii) the order of performance for the three PAI is: HP > CP > WP in terms of performance. Thirty-two AI attributes using each of two experts were categorized into four clusters, namely, architecture, optimization, dataset design, and scientific validation. The RoB was analyzed in these PAI studies using AP(ai)Bias 2.0 (AtheroPoint™, CA, USA), which was based on a scoring-and-ranking strategy. This was then compared against the Butterfly model, composed of radial and regional biases, and analyzed using a Venn diagram in three independent bins based on low-moderate and moderate-high cut-offs. Recommendations were laid out for low-bias studies. PAI studies that lack adequate AI attributes were more inclined to the RoB. It has been observed that 19.64%, 21.42%, and 58.92% studies were in low-bias, moderate-bias, and high-bias bins, respectively. Pruning made inference faster and reduced sparsity. We have observed low participation of the PAI studies in the XPAI framework. The study presented the challenges and five-step recommendations for mitigating bias in PAI models. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2024.

关键词： Risk assessment

来源：评论

学校读者我要写书评

暂无评论

Unlocking Domain Specificity: Fine-Tuning Llama 2 for Enhanced Performance on Custom Datasets

Unlocking Domain Specificity: Fine-Tuning Llama 2 for Enhanc...

引用

International IoT, Electronics and Mechatronics Conference, IEMTRONICS 2024

作者： Bhattacharya, Swarnadwip Bhattacharjee, Anindita Das Roy, Pranab Singha Samanta, Tapas Department of Computer Science and Engineering Institute of Engineering and Management Kolkata700091 India Department of CSE IEM Centre of Excellence for InnovAI Institute of Engineering and Management Kolkata700091 India Department of Computer and Informatics Variable Energy Cyclotron Centre Kolkata700064 India

ISBN: (纸本)9789819747795

This paper aims to improve Llama 2’s performance by using personalized and modified datasets. Despite the impressive capabilities of large language models (LLMs) such as Llama 2, their effectiveness may be limited in specialized domains. The proposed method entails fine-tuning Llama 2 on custom datasets to optimize performance efficiently. The study focuses on the impact of quantization-aware low-rank adapter layers (QLoRA) on a single GPU’s resource-efficient fine-tuning performance. Furthermore, the study looks into the design of instruction datasets to guide the model toward desired behaviors. When Llama 2 is fine-tuned with QLoRA, performance improves significantly across tasks such as text summarization, question answering, and natural language generation in a variety of domains. The paper concludes by highlighting the broader implications of the findings. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Custom datasets Fine-tuning Instruction datasets Large language models (LLMs) Llama 2 Personalized datasets QLoRA Quantization-aware low-rank adapter layers (QLoRA) Resource-efficient fine-tuning Specialized domains

来源：评论

学校读者我要写书评

暂无评论

DeepFakeGuard: Safeguarding Digital Platforms Against Fake Profiles Using AI 4

DeepFakeGuard: Safeguarding Digital Platforms Against Fake P...

引用

4th International Conference on Sentiment Analysis and Deep Learning, ICSADL 2025

作者： Swarna Sudha, M. Manjula, S. Bharathi, Indira Krishnasamy, Valarmathi Vijayalakshmi, K. Ramco Institute of Technology Department of Computer Science and Engineering Rajapalayam India School of Computer Science and Engineering Vellore Institute of Technology Chennai India PSR Engineering College Department of Electronics and Communication Engineering Sivakasi India

ISBN: (纸本)9798331523923

This research introduces DeepFakeGuard, a hybrid deep learning framework designed to detect fake profiles on social media platforms, addressing the growing threat of fraudulent accounts online. DeepFakeGuard integrates multiple data modalities, including textual, visual, behavioral, and network features, to accurately classify profiles as real or fake. Leveraging advanced deep learning models such as BERT for textual analysis, ResNet for image processing, and LSTM for modeling behavioral patterns, DeepFakeGuard provides a comprehensive and robust solution for identifying fake profiles across diverse platforms. The framework processes a rich set of profile characteristics, including account information, posting behavior, post sentiment, and network relationships, ensuring a multi-dimensional analysis of social media accounts. The hybrid architecture of DeepFakeGuard merges outputs from individual sub-models, which are then processed through dense layers for final classification using a sigmoid activation function. This multi-faceted approach significantly enhances detection accuracy, precision, and recall compared to traditional methods. The framework is highly scalable and adaptable to various social media platforms, offering a versatile tool for strengthening online security. With strong performance metrics, including high classification accuracy, DeepFakeGuard demonstrates its potential for real-world applications in mitigating the risks associated with fake profiles. By fostering secure and trustworthy social media environments, this research contributes to the ongoing fight against fraudulent accounts, paving the way for safer online interactions. © 2025 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Optimizing task scheduling in cloud environments: a hybrid golden search whale optimization algorithm approach

引用

Neural Computing and Applications 2025年 1-23页

作者： Acharya, Biswaranjan Panda, Sucheta Das, Satyabrata Majhi, Santosh Kumar Gerogiannis, Vassilis C. Kanavos, Andreas Department of Computer Science and Engineering Veer Surendra Sai University of Technology Odisha Burla India Bilaspur India Department of Digital Systems University of Thessaly Larissa Greece Department of Informatics Ionian University Corfu Greece

Managing fluctuating workloads and optimizing resource utilization in cloud environments pose significant challenges, particularly in fields requiring real-time data processing, such as healthcare. This paper introduces a novel hybrid metaheuristic technique, the Golden Search Whale Optimization Algorithm (GSWOA), specifically designed for scheduling independent dynamic biomedical data. GSWOA merges the strengths of Golden Search Optimization (GSO) and Whale Optimization Algorithm (WOA), optimizing numerical function optimization and achieving a balance between exploration and exploitation. The algorithm’s effectiveness was assessed using MATLAB by applying standard benchmark functions and further evaluated on a real-world biomedical dataset within the CloudSim environment. The performance evaluations demonstrate that GSWOA significantly outperforms existing metaheuristic and traditional scheduling techniques, achieving a 42.71% increase in resource utilization and a 14.17% reduction in makespan compared to conventional methods. These results highlight GSWOA’s potential to enhance scheduling efficiency substantially in cloud computing infrastructures, suggesting it is a powerful tool for complex task allocations. Future research will explore the scalability of GSWOA and its applicability across other data-intensive sectors. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., part of Springer Nature 2025.

关键词： Scheduling algorithms

来源：评论

学校读者我要写书评

暂无评论

A D-Band CMOS Linearity-Enhanced Low-Noise Amplifier Using Multiple-Gated Transistors Technique

引用

IEEE Access 2025年 13卷 82874-82880页

作者： Zhang, Jingbo Seo, Munkyo Sungkyunkwan University Department of Electrical and Computer Engineering Suwon16419 Korea Republic of Sungkyunkwan University School of Electronic and Electrical Engineering Suwon16419 Korea Republic of

In this paper, a D-band low-noise amplifier (LNA) using the linearity-enhanced technique in 40-nm CMOS process is presented. In this D-band LNA, multiple-gated transistors are used to improve the third-order intercept point (IIP3) while maintaining minimal small-signal gain reduction and nearly identical DC power consumption. This linearized LNA includes three conventional capacitive neutralization common-source differential stages and one multiple-gated transistor (MGTR) stage as the final stage. From the two-tone large-signal measurement results, the linear mode demonstrates a 5 dB improvement in the third-order intercept point (IIP3). In the one-tone large-signal measurement, a 4 dB improvement of the input 1-dB compression point (IP1dB) can be observed in the linear mode. For the small-signal measurement, this LNA achieves peak small-signal gains of 16.5 dB in normal mode and 14.2 dB in linear mode. The 3-dB bandwidth is 35 GHz (110-145 GHz) in both operating modes. Using the Y-factor method, the measured noise figure is 8.8 dB in normal mode and 9.5 dB in linear mode at 140 GHz. The DC consumption is 42 mW and 43 mW at 0.9 V for normal mode and linear mode. The whole chip area is 0.32 mm2. To the author's knowledge, this is the first implementation of the MGTR technique in CMOS circuits operating beyond 100 GHz. © 2013 IEEE.

关键词： CMOS integrated circuits

来源：评论

学校读者我要写书评

暂无评论

Optimized Neural Network for Automated Cardiovascular Disease Classification: Meta-heuristic Enabled Model

引用

SN computer Science 2025年第1期6卷 1-14页

作者： Kusuma, S. Jothi, K.R. Department of Computer Science & Engineering (Data Science) Madanapalle Institute of Technology & Science Madanapalle India School of Computer Science and Engineering Vellore Institute of Technology Vellore India

Cardiovascular disease remains a major issue for mortality and morbidity, making accurate classification crucial. This paper introduces a novel heart disease classification model utilizing Electrocardiogram (ECG) signals. The model consists of two main steps: feature extraction using a refined Lion Algorithm with New Mutated evaluation (LA-NM) and classification through an ensemble approach combined with a Neural Network (NN). The proposed model significantly outperforms traditional classifiers. At 80% training data, it achieves accuracy improvements of 8.4% to 3.11% over Neural Network (NN), K-Nearest Neighbors (KNN), Support Vector Machine (SVM), Decision Tree (DT), Random Forest (RF), and ensemble models. At 90% training data, it shows precision improvements ranging from 1.03% to 0.31% better than the comparison models and specificity improvements ranging from 13.92% to 0.52%. These results underscore the effectiveness of the model and suggest its potential for further development in personalized treatment recommendations based on patient history. © The Author(s), under exclusive licence to Springer Nature Singapore Pte Ltd. 2024.

关键词： Ensemble models Feature extraction Heart disease classification Neural network Optimization

来源：评论

学校读者我要写书评

暂无评论

IensNet: A novel and efficient approach for iris spoof detection via ensemble of deep models

引用

Multimedia Tools and Applications 2025年 1-30页

作者： Sharma, Deepika Selwal, Arvind School of Computer Science Engineering and Technology Bennett University Greater Noida201310 India Department of Computer Science and Information Technology Central University of Jammu Samba India

Iris biometrics allow contactless authentication, which makes it widely deployed human recognition mechanisms since the couple of years. Susceptibility of iris identification systems remains a challenging task due to diversity in spoof or presentation attacks (PAs) that fails to assure consistency while adopting them in real life scenarios. Hence, iris PAs are the growing concerns that gained significant attention in recent past decade. To alleviate these attacks or recognize presentation attack instruments (PAIs), iris presentation attacks detection (IPAD) algorithms are designed to distinguish a real and fabricated iris trait. Aiming at the efficient iris spoof detection mechanism, in this research work we expound a novel ensemble learning-enabled model (IensNet) that learns three pre-trained and fined-tuned deep models (i.e. DenseNet161, ResNet and VGGNet) for better accuracy and generalized performance. The novel IensNet approach offers several merits (i.e. consolidated strengths of multiple models, improved generalization ability, etc.) as compared to a simple transfer learning strategy where the knowledge is drawn from single pre-trained model. Finally, our approach learns a novel fully-connected dual layer classifier via outcome of three fine-tuned models to yield a final classification result as bonafide or spoof iris trait. Our approach is evaluated on Notre Dame LivDet iris 2017 and Notre Dame contact lenses 2015 anti-spoofing datasets. The experimental analysis of IensNet offers outstanding performance with a lower ACER of 0.2% and 1.4% for Iris-LivDet-2017 and Notre Dame contact lenses 2015 dataset respectively. Besides, IensNet exhibit promising results in cross-dataset environment with an ACA of 91.46%. © The Author(s), under exclusive licence to Springer Science+Business Media, LLC, part of Springer Nature 2025.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Direction-of-Arrival Estimation with Arbitrary Linear Array: A Turbo-HS Approach 5

Direction-of-Arrival Estimation with Arbitrary Linear Array:...

引用

5th IEEE International Conference on Power, Electronics and computer Applications, ICPECA 2025

作者： Chen, Zhendong Huang, Yongfeng Li, Dingzhao Lu, Xicheng Sun, Haixin School of Informatics Xiamen University Xiamen China School of Electronic Engineering and Computer Science Queen Mary University of London London United Kingdom

ISBN: (纸本)9798331533694

Conventional subspace-based direction-of-arrival (DOA) estimation algorithms require optimal environments to achieve satisfactory estimation accuracy. With the advancement of sparse signal recovery theory, sparse optimization-based DOA estimation algorithms have exhibited commendable performance. One of the popular sparse optimization method is sparse Bayesian learning (SBL), due to the fact that it does not require regularization parameters. However, existing SBL algorithms do not adequately address the issue of noise modeling. In order to more accurately model the received signal covariance matrix under finite snapshots, we transform the signal covariance matrix into the form of sparse representation, normalize the variance of the observation matrix and vectorize it. Based on this, we introduce the Turbo-CS framework, which divides the DOA estimation problem into two subproblems: LMMSE estimation for external noise separation and a novel prior for internal SBL DOA estimation. As demonstrated by the experimental results, the proposed algorithm performs optimally in a variety of scenarios. © 2025 IEEE.

关键词： Direction of arrival

来源：评论

学校读者我要写书评

暂无评论

Automated Classification and Segmentation of Brain Tumor Images using Backboned ResUNet Framework 6

Automated Classification and Segmentation of Brain Tumor Ima...

引用

6th International Conference on Mobile Computing and Sustainable informatics, ICMCSI 2025

作者： Prasanalakshmi, B. Anitha, Cuddapah Balamanigandan, R. Mahaveerakannan, R. Gade, N.V. Ratnakishor Adjunct Faculty Department of Computer Science and Engineering Saveetha School of Engineering Saveetha Institute of Medical and Technical Sciences Tamil Nadu Chennai India Department of Computer Science and Engineering School of Computing Mohan Babu University Andhra Pradesh Tirupati India Department of Computer Science and Engineering Saveetha School of Engineering Saveetha Institute of Medical and Technical Sciences Tamil Nadu Chennai India

ISBN: (纸本)9798331522667

One of the most challenging issues in computer imaging is the automated segmentation of brain tumors using Magnetic Resonance Images (MRI). Several approaches are explored using Deep Neural Networks in image segmentation since they perform well in automated tumor segmentation using tumor scans. Due to the complexity, training deeper neural networks usually takes a long time and many processing resources. An automated approach for classifying and segmenting images of Brain tumor MRI is proposed. As a hybrid transfer learning approach, modified ResNet50 identifies the class of MRI images as Benign or malignant. After identifying the Benign images, the tumor’s location is segmented using a hybrid network involving ResUNet with the Xception network as Backbone. This combination approach is chosen after its comparative analysis with other backbones. Various combination approaches are carried out, and performance metrics like Jaccard, Dice, and Tversky coefficient are done to prove the ResUNet backboned with Xception plays a better role concerning accuracy and speed. The proposed approach was tested on the dataset Cancer Imaging Archive to validate its efficiency. The proposed model has an average dice similarity coefficient score of 0.94, outperforming the conventional transfer learning training method with all the layers. © 2025 IEEE.

关键词： Magnetic resonance imaging

来源：评论

学校读者我要写书评

暂无评论

Comparative Analysis of Machine Learning Algorithm;for Conversion Predictors of Clinically Isolated Syndrome (CIS) to Multiple Sclerosis (MS) 4

Comparative Analysis of Machine Learning Algorithm;for Conve...

引用

4th International Conference on Sentiment Analysis and Deep Learning, ICSADL 2025

作者： Mendanha, Serena Gudibandi, Keerthi Reddy Hajarathaiah, Koduru Gudibandi, Bharat Reddy Enduri, Murali Krishna School of Computer Science and Engineering VIT-AP University Amaravati India School of Electronics Engineering VIT-AP University Amaravati India Srm University Department of Computer Science and Engineering AP India

ISBN: (纸本)9798331523923

The central nervous system is impacted by multiple sclerosis (MS), a chronic neurological condition that causes significant cognitive and physical deficits. Better disease management and prompt intervention depend on early and precise detection. Effective diagnosis of MS is hampered by a number of factors, such as small datasets, significant clinical presentation variability, difficult feature selection, model generalization problems, and integration of multimodal data such as magnetic resonance imaging and genetic markers. Several machine learning (ML) models are assessed in this study in order to predict the development of clinically isolated syndrome (CIS) to multiple sclerosis. Using clinical and magnetic resonance data from patients with CIS at risk of developing MS, we evaluate the effectiveness of support vector machines (SVM), K nearest neighbor (KNN), decision trees (DT), random forests (RF), logistic regression (LR), Gaussian naive Bayes (Gaussian NB) and XGBoost (XG). Evaluation is performed using performance criteria including F1 score, recall, precision, and precision. According to our research, Random Forest has the best prediction accuracy, which makes it a potentially useful tool for helping doctors diagnose and treat MS patients early. Notwithstanding these developments, issues like model interpretability and data scarcity still exist. In order to improve diagnosis precision, future research will concentrate on enhancing these models by integrating deep learning methods, genetic markers,, and more advanced imaging modalities. © 2025 IEEE.

关键词： Magnetic resonance imaging

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：