检索结果-内蒙古大学图书馆

Auditing and Debugging Deep Learning Models via Flip Points: Individual-Level and Group-Level Analysis

Matematica 2022年第1期1卷 19-52页

作者： Yousefzadeh, Roozbeh O’Leary, Dianne P. Yale Center for Medical Informatics Yale University and VA Connecticut Healthcare System New Haven 06510 CT United States Computer Science Department and Institute for Advanced Computer Studies University of Maryland College Park 20742 MD United States

Deep learning models have been criticized for their lack of easy interpretation, which undermines confidence in their use for important applications. Nevertheless, they are consistently utilized in many applications, consequential to humans’ lives, usually because of their better performance. Therefore, there is a great need for computational methods that can explain, audit, and debug such models. Here, we use flip points to accomplish these goals for deep learning classifiers used in social applications. A trained deep learning classifier is a mathematical function that maps inputs to classes. By way of training, the function partitions its domain and assigns a class to each of the partitions. Partitions are defined by the decision boundaries which are expected to be geometrically complex. This complexity is usually what makes deep learning models powerful classifiers. Flip points are points on those boundaries and, therefore, the key to understanding and changing the functional behavior of models. We use advanced numerical optimization techniques and state-of-the-art methods in numerical linear algebra, such as rank determination and reduced-order models to compute and analyze them. The resulting insight into the decision boundaries of a deep model can clearly explain the model’s output on the individual level, via an explanation report that is understandable by non-experts. We also develop a procedure to understand and audit model behavior towards groups of people. We show that examining decision boundaries of models in certain subspaces can reveal hidden biases that are not easily detectable. Flip points can also be used as synthetic data to alter the decision boundaries of a model and improve their functional behaviors. We demonstrate our methods by investigating several models trained on standard datasets used in social applications of machine learning. We also identify the features that are most responsible for particular classifications and misclassification

关键词： Artificial neural networks and deep learning (68T07) Auditing deep learning models Interpretable machine learning Learning and adaptive systems (AMS: 68T05) Reasoning under uncertainty (AMS: 68T37)

来源：评论

学校读者我要写书评

暂无评论

Enhancing Social Media Sarcasm Detection Using Chicken Swarm Optimization and Graph Neural Networks 2

Enhancing Social Media Sarcasm Detection Using Chicken Swarm...

引用

2nd IEEE International Conference on Contemporary computing and Communications, InC4 2024

作者： Singh, Kamred Udham Singh, Niharika Chaudhary, Varun Paliwal, Deepika Singh, Teekam Dewangan, Amit kumar School of Computing Graphic Era Hill University Dehradun India IMS Engineering College Ghaziabad India Graphic Era Deemed to be University Department of Computer Science and Engineering Dehradun India G.L. Bajaj Institute of Technology & Management Greater Noida India Geetanjali Institute of Technical Studies Dabok Udaipur India Guru Ghasidas Vishwavidyalaya Department of Information Technology Bilaspur India

ISBN: (纸本)9798350383652

Understanding social media sarcasm is a difficult but crucial NLP issue. Machine learning (ML) algorithms are needed to effectively discern sarcastic undertones in text due to its contextual and linguistic complexity. Traditional ML models distinguish sarcasm from sincerity via sentiment analysis, context-specific variables, and language markers. As online communication evolves, more complex technologies are needed to understand language's subtleties. The merging of deep learning (DL) approaches with contextual embedding has greatly improved sarcasm detection systems, allowing them to analyze text more precisely and contextually. The CSOA-GNN model, which combines the Chicken Swarm Optimization Algorithm (CSOA) with Graph Neural Networks to identify sarcasm. After rigorous data pre-processing, the model embeds words using TF-IDF (Term Frequency-Inverse Document Frequency). Sarcasm's unique language traits are vectorized for ML analysis using this method. GNN is then used to classify. Its ability to map out and grasp the complicated network of contextual links and dependencies among words inside sentences improves the model's sarcasm detection accuracy. CSOA is crucial to optimizing GNN performance in this model. CSOA optimizes the GNN hyperparameters to match social media text's sarcastic expressions. This unique technique relies on GNN-based classification, CSOA-based parameter optimization, and TF-IDF word embeddings, which provide encouraging results that demonstrate their efficacy. As they increase social media sarcasm detection accuracy and advance NLP. This study introduces the CSOA-GNN paradigm, enabling more advanced, contextually aware, and nuanced language interpretation systems. It advances our understanding of the complicated world of online communication, where sarcasm, frequently laced with nuanced indications and contextual richness, is a constant challenge. This model shows that powerful optimization methods may be integrated with neural network topo

关键词： Graph neural networks

来源：评论

学校读者我要写书评

暂无评论

Coherent optical link phase noise control based on SPGD algorithm with optimal estimation 6

Coherent optical link phase noise control based on SPGD algo...

引用

6th Optics Young Scientist Summit, OYSS 2023

作者： Luo, Yihao Zhang, Jun Du, Shiyin Yan, Qiuquan Zhou, Tong Jiang, Tian College of Advanced Interdisciplinary Studies National University of Defense Technology Changsha410073 China National Innovation Institute of Defense Technology Academy of Military Sciences Pla China Beijing100071 China Institute for Quantum Information State Key Laboratory of High Performance Computing College of Computer Science and Technology National University of Defense Technology Changsha410073 China Institute of Quantum Information Science and Technology College of Science National University of Defense Technology Changsha410073 China

ISBN: (纸本)9781510672642

The Stochastic Parallel Gradient Descent (SPGD) algorithm is a widely used model-free optimization algorithm in adaptive optical systems. However, when affected by noise, the SPGD algorithm may incorrectly estimate the gradient, resulting in a significant decrease in its performance and limitations in its application in complex environments. To address this issue, this paper introduces the optimal estimation algorithm and adds a self-checking process to the SPGD algorithm, proposing a Kal-SPGD algorithm. The feasibility of the Kal-SPGD algorithm is verified through simulation and experimentation. The experimental results show that compared with the standard SPGD, the use of the Kal-SPGD algorithm can reduce the signal amplitude variance by up to 92.15%. The Kal-SPGD algorithm provides a more favorable control technique for adaptive optical systems. © COPYRIGHT SPIE. Downloading of the abstract is permitted for personal use only.

关键词： Stochastic systems

来源：评论

学校读者我要写书评

暂无评论

Polynomial-Time Classical Simulation of Noisy IQP Circuits with Constant Depth

arXiv

引用

arXiv 2024年

作者： Rajakumar, Joel Watson, James D. Liu, Yi-Kai Joint Center for Quantum Information & Computer Science National Institute of Standards & Technology University of Maryland College Park United States Department of Computer Science Institute for Advanced Computer Studies University of Maryland College Park United States United States

Sampling from the output distributions of quantum computations comprising only commuting gates, known as instantaneous quantum polynomial (IQP) computations, is believed to be intractable for classical computers, and hence this task has become a leading candidate for testing the capabilities of quantum devices. Here we demonstrate that for an arbitrary IQP circuit undergoing dephasing or depolarizing noise, whose depth is greater than a critical 1) threshold, the output distribution can be efficiently sampled by a classical computer. Unlike other simulation algorithms for quantum supremacy tasks, we do not require assumptions on the circuit’s architecture, on anti-concentration properties, nor do we require Ω(log() circuit depth. We take advantage of the fact that IQP circuits have deep sections of diagonal gates, which allows the noise to build up predictably and induce a large-scale breakdown of entanglement within the circuit. Our results suggest that quantum supremacy experiments based on IQP circuits may be more susceptible to classical simulation than previously thought. Our results also imply that fault-tolerance within the IQP framework cannot be achieved past a critical circuit depth. Furthermore, we show that the critical depth threshold of our algorithm is tight, and below this threshold there are noisy IQP circuits which are hard to sample from. Thus we demonstrate that noisy IQP circuits exhibit a phase transition in the computational complexity of sampling, as circuit depth is increased. Copyright © 2024, The Authors. All rights reserved.

关键词： Quantum computers

来源：评论

学校读者我要写书评

暂无评论

Real-Time Threat Detection and AI-Driven Predictive Security for Consumer Applications

引用

IEEE Transactions on Consumer Electronics 2025年

作者： Thayalan, Sugirtha Radhakrishnan, Niranchana Ramana, T.V. Devarajan, Ganesh Gopal Karuppiah, Marimuthu Dabel, Maryam M. Al Department of Computer Science and Engineering Tamilnadu Tiruchirappalli India Alliance University Alliance School of Advanced Computing Department of Computer Science and Engineering Anekal Karnataka Bengaluru562106 India Jain University Department of Computer Science and Engineering Bengaluru India SRM Institute of Science and Technology Delhi -NCR Campus Department of Computer Science and Engineering Faculty of Engineering and Technology Uttar Pradesh Ghaziabad Delhi -Meerut Road Modinagar201204 India Presidency University Presidency School of Computer Science and Engineering Karnataka Bengaluru560064 India University of Hafr Al Batin Department of Computer Science and Engineering College of Computer Science and Engineering Saudi Arabia

The growing use of artificial intelligence (AI) in consumer applications, especially through smart devices and the Internet of Things (IoT), presents both opportunities and security challenges. Protecting user data and privacy is increasingly critical as these technologies become part of daily life. Traditional security models struggle with the large, real-time data generated by these systems, leading to problems with accuracy, scalability, and threat detection. To address this, we propose a novel framework that combines Collaborative Federated Learning with an edge-cloud architecture for predictive security. Our approach features the Two-Stage Attention Integrated Graph-based Multi-source Spatio-Temporal Data Fusion (2S-AGMSTDF) network, which processes multiple data sources, including sensor inputs, cyber activity, and user behavior patterns. Key components include AKGCN for cyber feature embedding, a GCN-ResNet-based transformer (GRCMT) for spatial data analysis, and an Attention-Based LSTM for temporal feature extraction. By performing local computations at the edge for real-time monitoring and using Federated Averaging at the cloud level, our framework improves accuracy, scalability, and privacy. Real-world evaluations show its superior performance in anomaly detection and predictive security compared to existing methods, offering a robust AI-driven solution for consumer security. © 1975-2011 IEEE.

关键词： Federated learning

来源：评论

学校读者我要写书评

暂无评论

Learning Latent Structures in Network Games via Data-Dependent Gated-Prior Graph Variational Autoencoders 41

Learning Latent Structures in Network Games via Data-Depende...

引用

41st International Conference on Machine Learning, ICML 2024

作者： Yu, Xue Li, Muchen Leng, Yan Liao, Renjie Center for Applied Statistics School of Statistics Renmin University of China Beijing China Beijing Advanced Innovation Center for Future Blockchain and Privacy Computing China Department of Computer Science University of British Columbia VancouverBC Canada McCombs School of Business The University of Texas at Austin AustinTX United States Vector Institute for AI Canada Department of Electrical and Computer Engineering University of British Columbia VancouverBC Canada

In network games, individuals interact strategically within network environments to maximize their utilities. However, obtaining network structures is challenging. In this work, we propose an unsupervised learning model, called data-dependent gated-prior graph variational autoencoder (GPGVAE), that infers the underlying latent interaction type (strategic complement vs. substitute) among individuals and the latent network structure based on their observed actions. Specially, we propose a spectral graph neural network (GNN) based encoder to predict the interaction type and a data-dependent gated prior that models network structures conditioned on the interaction type. We further propose a Transformer based mixture of Bernoulli encoder of network structures and a GNN based decoder of game actions. We systematically study the Monte Carlo gradient estimation methods and effectively train our model in a stage-wise fashion. Extensive experiments across various synthetic and real-world network games demonstrate that our model achieves state-of-the-art performances in inferring network structures and well captures interaction types. Copyright 2024 by the author(s)

关键词： Monte Carlo methods

来源：评论

学校读者我要写书评

暂无评论

Malware Classification using LSTM-CF Framework for Disk Forensic Analysis

Malware Classification using LSTM-CF Framework for Disk Fore...

引用

Contemporary computing and Communications (InC4), IEEE International Conference on

作者： Chaithanyaka Yeddeli Thirupathi Jawahar Sundaram Kheng Cher Yeo Shujahat Ali Khan Pritika Devaraju Sellappan Faculty of Science and Technology Energy and Resources Institute Charles Darwin University Darwin Australia Department of Computer Science and Applications Christ Academy Institute for Advanced Studies Bengluru India Faculty of Science and Technology Charles Darwin University Darwin Australia School of Computing Science and Engineering VIT Bhopal University Bhopal India

ISBN: (数字)9798350383652

ISBN: (纸本)9798350383669

In the current era, digital forensic investigators need specialized tools to extract digital footprints from hard disks. The number of crimes is rising dramatically. This has resulted in an increased number of unresolved cybercrime cases including malicious software, hacking and cyberfraud. Disk forensics or disk investigation is a massive task. It takes weeks to collect the traces of the 1TB hard drive and analyze them. It is a challenging task to detect reliable evidence because of the worldwide use and advancement of digital communication technologies. By using automated tools, only predictable areas of the disk are often investigated, and as a result, latent evidence in the hidden area might be missed. In this paper, we present Chaithu's Framework (CF) that gives us a clear picture of the steps to follow and implement a machine learning algorithm which is able to predict whether the request is related to malware or not by analyzing the large world Microsoft Malware Prediction dataset collected from Kaggle. It consists of a total of 7.8 million data samples with 84 features collected in real time from Windows 7,8,9 and 10 systems. Using predictive machine learning algorithms, frauds can be detected automatically and autonomous actions can be taken to prevent them.

关键词： Analytical models Machine learning algorithms Accuracy Predictive models Malware Real-time systems Software reliability

来源：评论

学校读者我要写书评

暂无评论

Deep Learning-Based Assessment of Teacher Effectiveness in classrooms: Integrating ARIMA and LSTM Models 2

Deep Learning-Based Assessment of Teacher Effectiveness in c...

引用

2nd IEEE International Conference on Emerging Research in Computational science, ICERCS 2024

作者： Shukla, Tushar Dhar Bharadwaja, A.V. Bhattacharya, Chaitali Chary Ch, Narasimha Bhushan, P. Vinay Thulasimani, T. University Institute Physical Education & Sports Chandigarh University Punjab Mohali India Autonomous College Department of ECE Visakhapatnam India Tecnia Institute of Advanced Studies CDL New Delhi India School of Engineering & Technology Dept of CSE Telangana Hyderabad India Malla Reddy University School of Engineering Computer Science and Engineering Hyderabad India Bannari Amman Institute of Technology Department of Mathematics Erode India

ISBN: (纸本)9798331534950

Teachers are no different from everyone else when it comes to the value of helpful criticism. Because the system teach so often, teachers have plenty of chances to perfect these skills. However, improving one's abilities via unstructured exercise alone isn't enough;targeted practice, usually with the help of a coach, may be necessary for mastery. The ideal coach would provide instructors with realistic yet demanding training goals, provide them with helpful feedback, and help them find ways to advance their careers. The proposed approach incorporates preprocessing, feature selection, and training of the model. Data cleaning, conversion, and formatting are all part of data preparation, which is preparing a dataset for analysis. Data gain and principal component analysis make up feature selection. The ARIMA-LSTM model was consistently used throughout the training. With a 90.86% average accuracy, this innovative method surpasses ARIMA and LSTM. © 2024 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Towards a Sustainable Future: Analyzing Water Quality Changes for Environmental Sustainability 1

Towards a Sustainable Future: Analyzing Water Quality Change...

引用

1st IEEE International Conference on Smart Technology, ICE-SMARTec 2023

作者： Mansor, Nur Suhaili Awang, Hapini Harun, Nor Hazlyna Hassan, Mohamad Ghozali Shehu Malami, Sarkin Tudu Shamsuddin, Noraishah Institute for Advanced and Smart Digital Opportunities School of Computing Universiti Utara Malaysia Sintok Malaysia School of Computing Universiti Utara Malaysia Data Science Research Lab Sintok Malaysia School of Technology Management and Logistics Universiti Utara Malaysia Sintok Malaysia Sokoto State University Faculty of Science Department of Computer Science and Information Technology Sokoto Nigeria Ici in Smart Manufacturing Sirim Bhd Shah Alam Malaysia

ISBN: (纸本)9798350306156

The development of industrial areas is frequently thought to have a good impact on society because it helps to meet the population's requirements. However, it also negatively impacts the environment, such as river pollution. If a country experiences rapid economic growth without considering the effects of industrial development on the environment, then human well-being will be affected. In this context, advanced development is attained by a developed nation while ensuring the well-being of society. Uncontrolled and unsustainable development increases environmental pollution, adversely affecting human health. As water is essential for human life, water resources, such as rivers, must be properly managed and protected to ensure human well-being. Thus, the study conducted on the changes in the water quality of the Kim Kim River provides insights into how unintended consequences of urban and industrial development can be identified and addressed using three sets of satellite imagery data from 2015, 2019 and 2020. The study was led to identify research problems and achieve the study objectives to obtain research results. The results of image classification analyses indicate that the development area near Sungai Kim Kim increased to 50%, with 38% from industrial areas. The RMSE for the image classification is 1.51, significantly better than the minimum allowed value of 2.00. It provides an understanding of the change in river water quality because of urban and industrial development. By highlighting the importance of sustainable development and the need to balance economic growth with environmental protection, this study contributes to the development of effective policies and strategies for the sustainable development of industrial areas. © 2023 IEEE.

关键词： Water quality

来源：评论

学校读者我要写书评

暂无评论

A codesign framework for online data analysis and reduction

A codesign framework for online data analysis and reduction

引用

作者： Mehta, Kshitij Allen, Bryce Wolf, Matthew Logan, Jeremy Suchyta, Eric Singhal, Swati Choi, Jong Y. Takahashi, Keichi Huck, Kevin Yakushin, Igor Sussman, Alan Munson, Todd Foster, Ian Klasky, Scott Computer Science and Mathematics Division Oak Ridge National Laboratory Oak RidgeTN United States Mathematics and Computer Science Division Argonne National Laboratory LemontIL United States Department of Computer Science University of Maryland College ParkMD United States Division of Information Science Nara Institute of Science and Technology Nara Japan Oregon Advanced Computing Institute for Science and Society University of Oregon EugeneOR United States Department of Computer Science The University of Chicago ChicagoIL United States

science applications preparing for the exascale era are increasingly exploring in situ computations comprising of simulation-analysis-reduction pipelines coupled in-memory. Efficient composition and execution of such complex pipelines for a target platform is a codesign process that evaluates the impact and tradeoffs of various application- and system-specific parameters. In this article, we describe a toolset for automating performance studies of composed HPC applications that perform online data reduction and analysis. We describe Cheetah, a new framework for composing parametric studies on coupled applications, and Savanna, a runtime engine for orchestrating and executing campaigns of codesign experiments. This toolset facilitates understanding the impact of various factors such as process placement, synchronicity of algorithms, and storage versus compute requirements for online analysis of large data. Ultimately, we aim to create a catalog of performance results that can help scientists understand tradeoffs when designing next-generation simulations that make use of online processing techniques. We illustrate the design of Cheetah and Savanna, and present application examples that use this framework to conduct codesign studies on small clusters as well as leadership class supercomputers. © 2021 John Wiley & Sons Ltd.

关键词： Data reduction

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：