检索结果-内蒙古大学图书馆

Soft-GNN:towards robust graph neural networks via self-adaptive data utilization

Frontiers of computer Science 2025年第4期19卷 1-12页

作者： Yao WU Hong HUANG Yu SONG Hai JIN National Engineering Research Center for Big Data Technology and System Service Computing Technology and System LabCluster and Grid Computing LabSchool of Computer Science and TechnologyHuazhong University of Science and TechnologyWuhan 430074China College of Information and Communication National University of Defense TechnologyWuhan 430019China Department of Computer Science and Operations Research Universitéde MontréalMontreal H3C 3J7Canada

Graph neural networks(GNNs)have gained traction and have been applied to various graph-based data analysis tasks due to their high ***,a major concern is their robustness,particularly when faced with graph data that has been deliberately or accidentally polluted with *** presents a challenge in learning robust GNNs under noisy *** address this issue,we propose a novel framework called Soft-GNN,which mitigates the influence of label noise by adapting the data utilized in *** approach employs a dynamic data utilization strategy that estimates adaptive weights based on prediction deviation,local deviation,and global *** better utilizing significant training samples and reducing the impact of label noise through dynamic data selection,GNNs are trained to be more *** evaluate the performance,robustness,generality,and complexity of our model on five real-world datasets,and our experimental results demonstrate the superiority of our approach over existing methods.

关键词： graph neural networks node classification label noise robustness

来源：评论

学校读者我要写书评

暂无评论

Prior Constraints-based Reward Model Training for Aligning Large Language Models 23

引用

23rd Chinese National Conference on Computational Linguistics, CCL 2024

作者： Zhou, Hang Wang, Chenglong Hu, Yimin Xiao, Tong Zhang, Chunliang Zhu, Jingbo NLP Lab School of Computer Science and Engineering Northeastern University Shenyang China NiuTrans Research Shenyang China

ISBN: (纸本)9780000000002

Reinforcement learning with human feedback for aligning large language models (LLMs) trains a reward model typically using ranking loss with comparison pairs. However, the training procedure suffers from an inherent problem: the uncontrolled scaling of reward scores during reinforcement learning due to the lack of constraints while training the reward model. This paper proposes a Prior Constraints-based Reward Model (PCRM) training method to mitigate this problem. PCRM incorporates prior constraints—specifically, length ratio and cosine similarity between outputs of each comparison pair—during reward model training to regulate optimization magnitude and control score margins. We comprehensively evaluate PCRM by examining its rank correlation with human preferences and its effectiveness in aligning LLMs via RL. Experimental results demonstrate that PCRM significantly improves alignment performance by effectively constraining reward score scaling. As another bonus, our method is easily integrated into arbitrary rank-based alignment methods, such as direct preference optimization, and can yield consistent improvement. The code is available at https://***/wangclnlp/DeepSpeed-Chat-Extension/tree/PCRM. © 2024 China National Conference on Computational Linguistics.

关键词： Computational linguistics

来源：评论

学校读者我要写书评

暂无评论

A Novel Key Management Framework for Secure and Scalable Decentralized Identity Systems 17

A Novel Key Management Framework for Secure and Scalable Dec...

引用

17th International Conference on Security of Information and Networks, SIN 2024

作者： Yildiz, Mert Bahtiyar, Serif Istanbul Technical University Cyber Security and Privacy Research Lab SPFLab Department of Computer Engineering Istanbul Turkey

ISBN: (纸本)9798331509736

The rise of decentralized identity systems has posed significant challenges in the secure and scalable management of keys, especially in large-scale national identity programs. In this paper, we propose a new secure and scalable framework for cryptographic keys management that may be applied in a national digital identity system. The proposed framework provides a hierarchical structure for efficient key generation and isolation, while hardware security modules provide a secure environment for key storage and operations. Key wrapping is implemented to enable secure external storage of large volumes of keys. In our work, we present a comprehensive security analysis. Our analysis demonstrates the resilience of the framework against various threat vectors and its ability to address key management challenges such as complexity, scalability, security isolation, recovery and secure delegation. The proposed framework provides a promising solution for security and scalability of national-level identity systems. © 2024 IEEE.

关键词： BIP32 decentralized identity hardware security modules key management key wrapping

来源：评论

学校读者我要写书评

暂无评论

TOWARDS IMITATION LEARNING TO BRANCH FOR MIP: A HYBRID REINFORCEMENT LEARNING BASED SAMPLE AUGMENTATION APPROACH 12

TOWARDS IMITATION LEARNING TO BRANCH FOR MIP: A HYBRID REINF...

引用

12th International Conference on Learning Representations, ICLR 2024

作者： Zhang, Changwen Ouyang, Wenli Yuan, Hao Gong, Liming Sun, Yong Guo, Ziao Dong, Zhichen Yan, Junchi AI Lab Lenovo Research China Department of Computer Science and Engineering Shanghai Jiao Tong University China

Branch-and-bound (B&B) has long been favored for tackling complex Mixed Integer Programming (MIP) problems, where the choice of branching strategy plays a pivotal role. Recently, Imitation Learning (IL)-based policies have emerged as potent alternatives to traditional rule-based approaches. However, it is nontrivial to acquire high-quality training samples, and IL often converges to suboptimal variable choices for branching, restricting the overall performance. In response to these challenges, we propose a novel hybrid online and offline reinforcement learning (RL) approach to enhance the branching policy by cost-effective training sample augmentation. In the online phase, we train an online RL agent to dynamically decide the sample generation processes, drawing from either the learning-based policy or the expert policy. The objective is to strike a balance between exploration and exploitation of the sample generation process. In the offline phase, a value function is trained to fit each decision's cumulative reward and filter the samples with high cumulative returns. This dual-purpose function not only reduces training complexity but also enhances the quality of the samples. To assess the efficacy of our data augmentation mechanism, we conduct comprehensive evaluations across a range of MIP problems. The results consistently show that it excels in making superior branching decisions compared to state-of-the-art learning-based models and the open-source solver SCIP. Notably, it even often outperforms Gurobi. © 2024 12th International Conference on Learning Representations, ICLR 2024. All rights reserved.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Generative Adversarial Networks for Synthetic Jamming Attacks on UAVs 9

Generative Adversarial Networks for Synthetic Jamming Attack...

引用

9th International Conference on computer Science and engineering, UBMK 2024

作者： Sarikaya, Burcu Sönmez Bahtiyar, Şerif Istanbul Technical University Cyber Security and Privacy Research Lab SPFLab Department of Computer Engineering Istanbul Turkey

ISBN: (纸本)9798350365887

Intrusion Detection Systems (IDS) have been significant for Unmanned Aerial Vehicles (UAVs) since high connectivity is essential for such vehicles. Recently, machine learning-based defense mechanisms have contributed to detecting intrusions. Data are crucial to better results with machine learning algorithms. However, security data is not freely distributed so there is a lack of security-related data about UAVs that can be processed with machine learning. In this research, we create synthetic security data for UAVs using Generative Adversarial Networks (GANs) that can be utilized to develop machine learning-based IDS systems. We propose employing Conditional Tabular Generative Adversarial Networks (CTGANs), Gaussian copulas, and Variational AutoEncoders (VAEs) to generate syn-thetic jamming attacks from the widely-used UAV attack dataset. We experimentally evaluated the performance of each model by using many metrics, such as Kullback-Leibler decomposition. Experimental results show that the distribution of CTGAN synthetic data is similar to the distribution of real data. Finally, we created a machine learning-based IDS and applied it to each dataset to validate the success rate of intrusion detection with newly generated synthetic data. It was observed that CTGAN-generated data decreased the IDS accuracy by approximately 25%, while Gaussian copula-generated data resulted in a 20% decrease, and VAE-generated data led to a 1% reduction, all relative to the accuracy achieved with real data. These findings suggest that synthetic data generation holds significant potential for enhancing the security frameworks of UAVs. The experimental results demonstrate that the CTGAN model is particularly effective in producing synthetic data that challenges traditional intrusion detection mechanisms, thereby contributing to the development of more robust IDS capabilities. © 2024 IEEE.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

ViT-PMN: A Vision Transformer Approach for Persian Numeral Recognition 20

ViT-PMN: A Vision Transformer Approach for Persian Numeral R...

引用

20th CSI International Symposium on Artificial Intelligence and Signal Processing, AISP 2024

作者： Ardehkhani, Pegah Ardehkhani, Pouya Hooshmand, Haniyeh Sharif University of Technology Department of Industrial Engineering Tehran Iran University of Tehran Faculty of Engineering Deep Learning Research Lab Department of Computer Engineering Tehran Iran University of Tehran Faculty of Engineering Department of Computer Engineering Tehran Iran

ISBN: (纸本)9798350383942

This study focuses on the task of Persian numeral classification within image data, employing the Vision Transformer (ViT) architecture to predict numerals akin to the MNIST dataset, but adapted to the Persian script. Our approach yielded a notable validation accuracy of 0.9920, particularly noteworthy when employing a patch size of 4. Notably, the research introduces an innovative visualization aspect, showcasing the first multi-head attention linear map and its counterpart, the last one. The visualization of these attention maps provides a unique insight into the model's internal processes and highlights its proficiency in capturing intricate patterns within Persian numeral images. This work contributes to the evolving landscape of character recognition, specifically addressing the challenges posed by the Persian script, and underscores the efficacy of employing the ViT architecture for such intricate tasks. The achieved validation accuracy and the detailed visualization of attention maps mark notable milestones in the realm of Persian numeral classification, showcasing the potential of Vision Transformers in the context of script-specific optical character recognition. © 2024 IEEE.

关键词： Visualization

来源：评论

学校读者我要写书评

暂无评论

Optimization of 5G Antenna Using Machine Learning 5

Optimization of 5G Antenna Using Machine Learning

引用

5th IEEE Middle East and North Africa COMMunications Conference, MENACOMM 2025

作者： Hassan, Aline Sharanek, Aya Hamzeh, Rayan Serhal, Dina Zantout, Rached American University of Beirut Electrical and Computer Engineering Department Beirut Lebanon Rafik Hariri University Electrical and Computer Engineering Department Mechref Lebanon EFREI Paris EFREI Research Lab Paris France

ISBN: (纸本)9798331519957

This paper presents a machine learning (ML)-based approach to optimize 5G base station patch antennas operating in the 3.3-4.2 GHz range. Traditional antenna design methods are often complex and time-consuming, particularly as the demand for high-speed communication is increasing, motivating the use of ML. Due to the lack of existing data for training the ML models, a custom dataset was created to be used in training and testing of the ML models. The optimized antenna achieved a return loss less than-10 dB and a resonant frequency of 3.5 GHz. An array of the optimized antennas was also designed and analyzed for base station applications. To select the best ML model, the project involved training and testing various models. Random Forest was identified as the best model for the single antenna, while Random Tree performed best for the antenna array. © 2025 IEEE.

关键词： Slot antennas

来源：评论

学校读者我要写书评

暂无评论

Novel Approach to Image Similarity Estimation and Object Matching: Leveraging ViT Architecture and Euclidean Distance Metric 20

Novel Approach to Image Similarity Estimation and Object Mat...

引用

20th CSI International Symposium on Artificial Intelligence and Signal Processing, AISP 2024

作者： Ardehkhani, Pegah Ardehkhani, Pouya Hooshmand, Haniyeh Sharif University of Technology Department of Industrial Engineering Tehran Iran Faculty of Engineering University of Tehran Deep Learning Research Lab Department of Computer Engineering Tehran Iran Faculty of Engineering University of Tehran Department of Computer Engineering Tehran Iran

ISBN: (纸本)9798350383942

In addressing the challenge of image similarity estimation on the MNIST dataset, our research drives from conventional Siamese network methodologies by incorporating Vision Transformer (ViT) architecture. Departing from the standard MNIST dataset, we introduced a novel paired dataset tailored to enhance the capabilities of similarity estimation. The innovation lies in the utilization of ViT as the core foundation for extracting features, followed by the application of Euclidean distance metrics on the dual input. This departure from the traditional approach not only broadens the scope of image similarity assessment but also enhances the model's discriminative power. Notably, the model attains a commendable test accuracy of 97.14% with a patch size of 7, underscoring the efficacy of our proposed methodology. This work it not only adds to the changing scenery of the image similarity estimation while also emphasizing the importance of leveraging non-conventional architectures to achieve enhanced performance in this domain. © 2024 IEEE.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

Enhancing Employee Promotion Prediction with a Novel Hybrid Model Integrating Convolutional Neural Networks and Random Forest 14

Enhancing Employee Promotion Prediction with a Novel Hybrid ...

引用

14th International Conference on Information and Knowledge Technology, IKT 2023

作者： Ardehkhani, Pouya Moslemi, Seyyed Reza Hooshmand, Haniyeh University of Tehran Deep Learning Research Lab Department of Computer Engineering Faculty of Engineering Tehran Iran University of Tehran Faculty of Engineering Department of Computer Engineering Tehran Iran

ISBN: (纸本)9798350349412

In the ever-evolving landscape of human resources, the critical task of identifying employees ready for promotion remains a complex challenge. To address this issue, we propose a novel hybrid model that seamlessly integrates Convolutional Neural Networks (CNNs) with Random Forest. Through a two-step process, we initially train the CNN comprising Conv1D and Dense layers. Subsequently, we harness the extracted features from the Nth layer, merging them with the original dataset. These augmented features are then input into the Random Forest algorithm. This innovative approach has yielded remarkable results, achieving an astounding accuracy rate of 99%. This surpasses the performance of both standalone Random Forest and CNN models, as well as various other machine learning methods. The presented model not only enhances the prediction accuracy for employee promotions but also offers a powerful tool for HR managers seeking to make informed and data-driven decisions in workforce advancement, ultimately contributing to more effective and efficient talent management. © 2023 IEEE.

关键词： Personnel

来源：评论

学校读者我要写书评

暂无评论

Improving Training Stability in Variational Autoencoders Through the Integration of Score Matching Loss 14

Improving Training Stability in Variational Autoencoders Thr...

引用

14th International Conference on Information and Knowledge Technology, IKT 2023

作者： Rad, Amirreza Mokhtari Ardehkhani, Pouya Alborzi, Hormehr University of Tehran Deep Learning Research Lab Department of Computer Engineering Faculty of Engineering Tehran Iran University of Tehran Faculty of Engineering Department of Computer Engineering Tehran Iran

ISBN: (纸本)9798350349412

In this research, a Variational Autoencoder (VAE) model was developed, and the CIFAR100 dataset was employed as the primary data source. The problem addressed pertained to the instability in the training process of VAE models. To mitigate this issue, various loss expressions were explored, including the use of score matching loss independently, in conjunction with total variation loss, and in combination with reconstruction loss. The innovative approach revealed that when score matching loss was integrated either with total variation loss or when applied as a standalone loss function, the training process exhibited increased stability. This was evident through smoother loss curves and latent space visualizations that displayed characteristics akin to a normal distribution. As a consequence, this novel approach promises the potential for building more stable generative models, which can significantly enhance the overall training process in VAEs. This innovation provides a valuable contribution to the field of generative modeling, with the prospect of addressing the longstanding challenge of training stability in VAEs, thereby opening avenues for more efficient and effective model development and application. © 2023 IEEE.

关键词： Stability

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：