检索结果-内蒙古大学图书馆

arXiv 2024年

作者： Franco, Gabriel Crovella, Mark Department of Computer Science Boston University United States Department of Computer Science Faculty of Computing & Data Sciences Boston University United States

Many papers have shown that attention heads work in conjunction with each other to perform complex tasks. It's frequently assumed that communication between attention heads is via the addition of specific features to token residuals. In this work we seek to isolate and identify the features used to effect communication and coordination among attention heads in GPT-2 small. Our key leverage on the problem is to show that these features are very often sparsely coded in the singular vectors of attention head matrices. We characterize the dimensionality and occurrence of these signals across the attention heads in GPT-2 small when used for the Indirect Object Identification (IOI) task. The sparse encoding of signals, as provided by attention head singular vectors, allows for efficient separation of signals from the residual background and straightforward identification of communication paths between attention heads. We explore the effectiveness of this approach by tracing portions of the circuits used in the IOI task. Our traces reveal considerable detail not present in previous studies, shedding light on the nature of redundant paths present in GPT-2. And our traces go beyond previous work by identifying features used to communicate between attention heads when performing IOI. © 2024, CC BY.

关键词： Signal encoding

来源：评论

学校读者我要写书评

暂无评论

Explainable Ai-Driven Financial Transaction Fraud Detection Using Machine Learning and Deep Neural Networks

SSRN

引用

SSRN 2023年

作者： Sai, Chaithanya Vamshi Das, Debashish Elmitwally, Nouh Elezaj, Ogerta Islam, Md Baharul Department of Computing & Data Science Faculty of Computing Engineering & the Built Environment Birmingham City University United Kingdom College of Data Science and Engineering American University of Malta Malta

Artificial Intelligence (AI) is revolutionizing and transforming data-driven businesses like financial services, e-commerce, and banking. AI technology and the digitization of banking are rapidly increasing the volume of financial transaction frauds, causing billions in revenue losses and unrecoverable funds. The primary business challenge faced by the finance and banking industry is understanding how their AI systems make predictions and explaining why a transaction is flagged as fraud. Explainable Artificial Intelligence (XAI) is a human-centric AI that accelerates business value through a greater understanding of predictions and makes complex black box decisions more transparent, trustable, explainable. This research involves the development of a novel classification model to classify whether a financial transaction is fraudulent or legitimate using five Machine learning algorithms and two Deep Neural Networks Algorithms like Artificial Neural Networks (ANN) and Convolutional Neural Networks (CNN). To tackle challenges like data quality issues and high-class imbalance in the dataset, several experimentations like Memory Optimisation, Hyperparameter Tuning methods have been implemented. Based on the literature, this is a novel instance of research investigating and implementing five XAI methods on the IEEE-CIS fraud detection dataset. After hyperparameter tuning, the experimental results showed that the Light GBM model performed better with 98.3% Accuracy, F1- score of 0.70, and AUC-ROC score of 0.96 compared with other models. The outcome of the research can be applied to different areas for better decision-making and generating business value through real-time predictions. © 2023, The Authors. All rights reserved.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

Attentive Feature Segmentation model for Stroke detection on MRI images

Attentive Feature Segmentation model for Stroke detection on...

引用

BIT Conference (BITCON), International

作者： S.E. Viswapriya D. Rajeswari Department of Data Science and Business Systems School of Computing College of Engineering and Technology SRM Institute of Science and Technology Chennai India

ISBN: (数字)9798331518394

ISBN: (纸本)9798331518400

Medical image segmentation plays a crucial role in treatment planning, disease diagnosis, and guiding surgical navigation. Research should focus on medical image segmentation to achieve substantial breakthroughs or fill knowledge gaps that could lead to discoveries and improvements. Deep learning (DL) techniques have recently become necessary in creating advanced models, particularly for medical image segmentation. This paper introduces a novel segmentation algorithm specifically designed for MRI images. Initially, the paper will discuss the introduction and challenges in medical image segmentation. Later, developing a novel attentive feature-based segmentation algorithm is focused and their corresponding evaluation steps are measured. The evaluation metrics are finally compared with other existing segmentation algorithms and concluded that the proposed model surpasses the cutting-edge methods by notable differences in performance, showing the adaptability and overall effectiveness of the proposed model.

关键词： Deep learning Image segmentation Adaptation models Magnetic resonance imaging Semantic segmentation Surgery Computer architecture Feature extraction Medical diagnostic imaging Testing

来源：评论

学校读者我要写书评

暂无评论

Last-X-Generation Archiving Strategy for Multi-Objective Evolutionary Algorithms 13

Last-X-Generation Archiving Strategy for Multi-Objective Evo...

引用

13th IEEE Congress on Evolutionary Computation, CEC 2024

作者： Shu, Tianye Nan, Yang Shang, Ke Ishibuchi, Hisao Southern University of Science and Technology Guangdong Provincial Key Laboratory of Brain-inspired Intelligent Computation Department of Computer Science and Engineering Shenzhen518055 China Southern University of Science and Technology Shenzhen518055 China Shenzhen University National Engineering Laboratory for Big Data System Computing Technology Shenzhen518060 China

ISBN: (纸本)9798350308365

For evolutionary multi-objective optimization algorithms (EMOAs), an external archive can be utilized for saving good solutions found throughout the evolutionary process. Recent studies showed that a solution set selected from an external archive is usually superior to the final population. That is, the incorporation of an external archive improves the performance of EMOAs. However, the computation time for maintaining an external archive is long, especially when the archive size is large. To solve this issue, a simple archiving strategy is to save all solutions generated in the last several generations. In this paper, we examine this archiving strategy for three representative EMOAs on artificial test problems (Minus-DTLZ and WFG) and real-world problems (RE). Our results show that archiving the last several generations clearly improves the performance of EMOAs without severely increasing the computation time. © 2024 IEEE.

关键词： Optimization algorithms

来源：评论

学校读者我要写书评

暂无评论

Debiasing Graph Representation Learning Based on Information Bottleneck

引用

IEEE Transactions on Neural Networks and Learning Systems 2024年第6期36卷 11092-11105页

作者： Ziyi Zhang Mingxuan Ouyang Wanyu Lin Hao Lan Lei Yang School of Software Engineering South China University of Technology Guangzhou China Department of Computing The Hong Kong Polytechnic University Hong Kong China Department of Data Science and Artificial Intelligence The Hong Kong Polytechnic University Hong Kong China Department of Computing Department of Data Science and Artificial Intelligence The Hong Kong Polytechnic University Hong Kong China Department of Computer Science and Technology Tsinghua University Beijing China

Graph representation learning has shown superior performance in numerous real-world applications, such as finance and social networks. Nevertheless, most existing works might make discriminatory predictions due to insufficient attention to fairness in their decision-making processes. This oversight has prompted a growing focus on fair representation learning. Among recent explorations on fair representation learning, prior works based on the adversarial learning usually induce unstable or counterproductive performance. To achieve fairness in a stable manner, we present the design and implementation of graph representation learning based on fairness information bottleneck (GRAFair), a new framework based on a variational graph autoencoder (VGAE). The crux of GRAFair is the conditional fairness bottleneck (CFB), where the objective is to capture the trade-off between the utility of representations and sensitive information of interest. By applying variational approximation, we can make the optimization objective tractable. Particularly, GRAFair can be trained to produce informative representations of tasks while containing little sensitive information without adversarial training. Experiments on various real-world datasets demonstrate the effectiveness of our proposed method in terms of fairness, utility, robustness, and stability.

关键词： Representation learning Adversarial machine learning Graph neural networks Optimization Electronic mail Training Predictive models Message passing Encoding Vectors

来源：评论

学校读者我要写书评

暂无评论

Phylogenetic-Informed Generative Adversarial Network for Predicting Mutations in SARS-CoV-2

Phylogenetic-Informed Generative Adversarial Network for Pre...

引用

Networking and Communications (ICNWC), International Conference on

作者： Vijay Chandar Vishnu G Nair Rajeswari D Department of Data Science and Business Systems School of Computing College of Engineering and Technology SRM Institute of Science and Technology Chennai India

ISBN: (数字)9798350365269

ISBN: (纸本)9798350365276

In the pursuit of developing more effective vaccines against viruses like SARS-CoV-2, understanding their potential mutations is crucial. In order to generate protein sequences, this study presents a unique framework that synergistically blends phylogenetic data with Generative Adversarial Networks (GANs). The proposed method combines the strengths of two models: TEMPO, a transformer-based model proficient in leveraging phylogenetic data for mutation detection in SARS-CoV-2, and MutaGAN, a GAN-based model adept in generating synthetic protein sequences. The phylogenetic data from TEMPO is used in this novel design as the input for the MutaGAN generator, replacing the conventional random noise input used in GANs. The generator may now create synthetic protein sequences that are impacted by the evolutionary past included in the phylogenetic data resulting in the creation of artificial protein sequences that are more biologically significant and lifelike. The discriminator portion of the model, similar to MutaGAN, is still in charge of determining whether or not the input sequences comprise a valid parent-child pair and uses sequences produced by the first GAN as extra negative cases. This prevented the model from going too far off its planned route. Additional investigation into this fascinating confluence of bioinformatics and AI may be possible as a result of this study’s conclusions. This approach has tremendous promise in many applications, such as the production of more powerful vaccinations, protein engineering, medication design, and improving our knowledge of biological systems, by forecasting how SARS-CoV-2 could change. A precision of 0.662 and an accuracy of around 0.671, together with a Matthews Correlation Coefficient (MCC) score of 0.313, show that the framework performs very well. These scores show that the framework performs better than a number of cutting-edge baseline techniques. The synthetic protein sequences generated by the framework not only bea

关键词： Proteins Protein engineering Biological system modeling Generative adversarial networks Transformers data models Generators

来源：评论

学校读者我要写书评

暂无评论

Pathway-Based Analysis Using SVM-RFE for Gene Selection and Classification 7th

Pathway-Based Analysis Using SVM-RFE for Gene Selection and ...

引用

7th International Conference on Artificial Intelligence and Virtual Reality, AIVR 2023

作者： Rahman, Nurazreen Afiqah A. Nasarudin, Nurul Athirah Mohamad, Mohd Saberi Artificial Intelligence and Bioinformatics Research Group Faculty of Computing Universiti Teknologi Malaysia Johor81310 Malaysia Health Data Science Lab Department of Genetics and Genomics College of Medicine and Health Science United Arab Emirates University Al Ain Abu Dhabi United Arab Emirates

ISBN: (纸本)9789819990177

The pathway-based analysis is one method for selecting and classifying genes by incorporating pathway information. Integration of pathway knowledge into microarray data significantly advances researchers in the analysis of complex diseases. Microarray data involves thousands of genes to be selected, and therefore, a suitable method to eliminate noisy and uninformative genes is needed. Selecting significant genes for a specific disease is crucial to identifying genes highly related to disease production. Previous research shows that using both pathway information and gene expression data is more significant in disease identification. Therefore, pathway-based analysis using Support Vector Machine Recursive Feature Elimination (SVM-RFE) is introduced in this research to identify significant genes associated with analyzing the targeted phenotype. The datasets involved in this research are lung cancer and gender dataset. The results from the proposed method performed better than previous work as the significant genes are selected from the highest rank of genes in the highest rank of the pathway. The performance of the proposed method was evaluated using 10-fold cross-validation in terms of accuracy. Finally, a biological validation was conducted on selected genes in the top 5 pathways based on biological literature. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2024.

关键词： Gene expression

来源：评论

学校读者我要写书评

暂无评论

Polar Region Climate Dynamics: Deep Learning and Remote Sensing Integration for Monitoring Arctic and Antarctic Changes

引用

Remote Sensing in Earth Systems sciences 2024年第4期7卷 582-595页

作者： Maniraj, Subramanian Pitchiah Rose, Jesuharan Dafni Arunachalam, Rohini Rangasamy, Kotteeswaran Patil, Vishal Ratansing Kathirvelu, Selvakumarasamy Department of Data Science and Business Systems School of Computing SRM Institute of Science and Technology Kattankulathur Campus Chennai India Department of Computer Science and Engineering St. Joseph’s Institute of Technology Chennai India Department of Computer Science and Engineering Anil Neerukonda Institute of Technology and Sciences Andhra Pradesh Visakhapatnam India Joseph’s College of Engineering Chennai 600119 India Department of CSE(AIML) Vishwakarma Institute of Technology Pune India Department of Electronics and Communication Engineering Saveetha School of Engineering (SIMATS) Chennai India

The polar regions of the Earth, specifically the Arctic and Antarctic, play a pivotal role in regulating the planet’s climate systems. These regions are highly sensitive to climate change, with observable shifts such as diminishing sea ice, retreating glaciers, and rising temperatures. Understanding these dynamics is essential for predicting broader global climate impacts. This paper presents a novel approach for monitoring and predicting these changes by integrating an attention-based transformer model with a 3D convolutional neural network (3D-CNN). The proposed model is designed to capture both the spatial features of satellite imagery and the temporal dynamics of climate data, offering enhanced predictive capabilities for key climate variables such as sea ice extent, glacier movement, and temperature variations. The dataset was split into 80% training and 20% testing with tenfold cross-validation, and the model was trained using the Adam optimizer. The results show that the proposed model achieved a remarkable accuracy of 97.43%, outperforming traditional CNN and RNN models. The model also demonstrated a precision of 98.12% for sea ice extent prediction, a 96.45% precision for glacier movement detection, and a mean squared error (MSE) of 0.35 °C for temperature predictions. These findings suggest that deep learning models, specifically the integration of transformer and 3D-CNN architectures, can significantly contribute to polar region climate monitoring, offering improved tools for scientific research and policymaking. © The Author(s), under exclusive licence to Springer Nature Switzerland AG 2024. corrected publication 2024.

关键词： 3D convolutional neural network Attention-based transformer Climate dynamics Deep learning Glacier movement Polar region Remote sensing Sea ice extent Spatial–temporal analysis Temperature prediction

来源：评论

学校读者我要写书评

暂无评论

Peak Evaluation for Liquid Chromatography-Mass Spectrometry data Using Neural Network on Statistical Quality Metrics 18

Peak Evaluation for Liquid Chromatography-Mass Spectrometry ...

引用

18th IEEE International Conference on Anti-Counterfeiting, Security, and Identification, ASID 2024

作者： Dong, Liheng Dai, Juehong Guan, Teo Bee Guo, Lei Wu, Caisheng Xu, Jingjing School of Computing and Data Science Xiamen University Malaysia Sepang43600 Malaysia Xiamen University Department of Electronic Science Xiamen361005 China Interdisciplinary Institute for Medical Engineering Fuzhou University Fuzhou350108 China School of Pharmaceutical Sciences Xiamen University Xiamen361005 China

ISBN: (纸本)9798331505639

Peak picking is a crucial step in the data analysis workflow of Liquid Chromatography-Mass Spectrometry (LCMS) in untargeted metabolomics. The process is challenging due to interference from background signals such as noise and matrix effects. To address this challenge, we first collected 5,447 ion peaks and manually scored them after alignment. Next, we calculated 12 quality metrics for each ion peak. Finally, we developed a Multilayer Perceptron (MLP) model to perform regression analysis, as well as binary and ternary classification, making it adaptable to various application scenarios. The results demonstrate that the MLP-based peak evaluation method is effective in accurately predicting peak quality. Researchers can quickly initiate their work by setting just one or two score thresholds based on their specific needs. This approach retains the flexibility to handle different datasets without the need to adjust thresholds for each peak quality metric, representing a step towards more flexible and efficient peak picking. © 2024 IEEE.

关键词： Regression analysis

来源：评论

学校读者我要写书评

暂无评论

Exploring the Impact of Herbal Remedies on Autism Spectrum Disorder Using AGLSTM Analysis

Exploring the Impact of Herbal Remedies on Autism Spectrum D...

引用

Innovative computing, Intelligent Communication and Smart Electrical Systems (ICSES), International Conference on

作者： Bagiyalakshmi N T. Veeramakali Department of Data Science and Business Systems School of Computing SRM Institute of Science and Technology Kattankulathur

Autism is a personality disorder that can cause significant emotional, intellectual, and behavioral deficits. Autism is recognized as an illness on the spectrum because, while certain kids might have mild symptoms, others may have extensive life-challenging abnormalities. However, autistic people have unique characteristics and health goals and may react differently to conventional or herbal therapy than others. Herbal medications might be a possible approach to helping autistic people's bodies and brains. However, further research is needed to determine whether herbal therapy is useful for autistic individuals. To gain insight into how people think regarding herbal medication for autism, data must be processed promptly as it is generated, which may be accomplished using sentiment assessment, which analyzes polarization in writings. This paper proposes the AGLSTM Model, a hybrid model that solves the sentiment evaluation issue by combining LSTM with agglomerative clustering. The suggested system signifies text data into numerical vectors that illustrate text features such as the frequency of words and embeddings of phrases using the TF-IDF technique and the word2vec model. Every word becomes assigned to a cluster using a method known as clustering based on how similar it is to the cluster nucleus or other texts in the region. The outcome of the experiment is a set of clusters composed of sentences with comparable emotional content. The LSTM framework can detect long-term correlations between word patterns and classify feedback as positive or negative.

关键词：

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：