检索结果-内蒙古大学图书馆

2023 China Automation Congress, CAC 2023

作者： Wang, Lei Huang, Wei Chen, Yiping Chen, Hao Liu, Chuang Yin, Jun School of Computer Nanjing University of Posts and Telecommunications Nanjing210023 China Jiangsu Key Laboratory of Big Data Security and Intelligent Processing Nanjing210023 China School of Internet of Things Nanjing University of Posts and Telecommunications Nanjing210023 China

ISBN: (纸本)9798350303759

This paper proposes a scene segmentation method based on optimized line detection to address the difficulty of segmenting spatially independent equipment in complex substation scenes. The method first employs the RANSAC method to simulate the ground plane and extract the ground point cloud. Then, it projects the 3D point cloud scene onto a 2D plane to obtain a 2D image of the substation. Straight line segments of the power line are extracted from the image using the edge detection algorithm and an improved line detection algorithm. The straight line segments that do not belong to the power line are filtered out using the surrounding features of the 3D point corresponding to the 2D point on the line. The power line's straight line segment is extended using the region-growing algorithm. Finally, the power line clustering is completed using European clustering based on the three-dimensional coordinates of the two ends of the power line, thereby completing the scene segmentation. Experimental results show that the proposed method can effectively segment substation equipment. Compared to traditional segmentation methods, this method has an average power line identification accuracy of 14% higher. © 2023 IEEE.

关键词： Edge detection

来源：评论

学校读者我要写书评

暂无评论

Content-free Logical Modification of Large Language Model by Disentangling and Modifying Logic Representation 39

Content-free Logical Modification of Large Language Model by...

引用

39th Annual AAAI Conference on Artificial Intelligence, AAAI 2025

作者： Wu, Xin Bu, Yuqi Chen, Yifei Cai, Yi South China University of Technology Guangdong China Key Laboratory of Big Data and Intelligent Robot South China University of Technology Ministry of Education China

ISBN: (纸本)157735897X

Despite extensive training on diverse datasets and alignment with human values, large language models (LLMs) can still generate fallacious outputs. Additionally, the validity of LLM’s outputs varies significantly depending on the content. It is crucial to ensure LLMs’ logical consistency across different contexts. Drawing inspiration from cognitive psychology studies, we propose a Logic Control Framework (LCF) that disentangles LLMs’ hidden representations into separate content and logic spaces. Within the logic space, we use logically valid and invalid samples to construct distinct regions through contrastive learning. By moving logic representations to logically valid regions and fusing them with unchanged content representations, we significantly reduce logical fallacies in LLM outputs while maintaining content coherence. We demonstrate the effectiveness of LCF through experiments on conclusion generation and fallacy identification tasks, showing a significant improvement in logical validity and a reduction in fallacious outputs. © 2025, Association for the Advancement of Artificial Intelligence (***). All rights reserved.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

LDMOS-Based Doherty Power Amplifier Design in 5G Mobile Micro Base Stations

LDMOS-Based Doherty Power Amplifier Design in 5G Mobile Micr...

引用

Power Electronics, Computer Applications (ICPECA), IEEE International Conference on

作者： Mingming Gao Xinyu Wang Congying Wang Zhongxiang Zhang Shiquan Shan Yonghui Song Liaoning Key Laboratory of Radio Frequency and Big Data for Intelligent Applications College of Electronics and Information Engineering Liaoning Technicial University Huludao Liaoning China

ISBN: (数字)9798331533694

ISBN: (纸本)9798331533700

A Doherty Power Amplifier (DPA) has been designed and optimized specifically for compact mobile base station deployment, operating within a frequency range of 3.3 GHz to 3.6 GHz. The amplifier utilizes the proprietary DS10-2092 LDMOS transistors developed by the company for both the carrier and peak power amplifiers. Physical test results indicate that the DPA achieves saturated output power meeting the specified requirements, with a maximum saturated drain efficiency of 41%. The drain efficiency remains stable between 26.3% and 30.7% over a 6 dB power back-off range, demonstrating its superior performance in high-efficiency and broadband applications.

关键词： Base stations Power measurement 5G mobile communication Power amplifiers Energy efficiency Power electronics Distance measurement Transistors Power generation Optimization

来源：评论

学校读者我要写书评

暂无评论

MDPFL:A Multiple Differential Privacy Protection Method based on Federated Learning

MDPFL:A Multiple Differential Privacy Protection Method base...

引用

International Conference on Computer Supported Cooperative Work in Design

作者： Xinyu Liu Xiaolong Xu Jiangsu Key Laboratory of Big Data Security & Intelligent Processing Nanjing University of Posts and Telecommunications Nanjing China

With the continuous development of artificial intelligence technology, machine learning in distributed network systems, such as IoVflntemet of Vehicles), will inevitably lead to privacy leakage. At present, there are lots of problems in federated learning, differential privacy and other machine learning privacy protection schemes, such as high loss of availability and the need of a fully trusted third-party server. In order to solve these problems, we propose MDPFL(Multiple Differential Privacy based on Federated Learning) algorithm. The algorithm combines with the differential privacy model in each stage of federated learning to solve the problem of curious third-party data collectors obtaining users’ original data in the process of machine learning. Meanwhile, the algorithm does not adopt a decentralized machine learning scheme directly, but uses a double noise adding mode with the existence of the third-party data collectors. We use Laplacian mechanism in the central server and GRR mechanism in local clients to ensure the goal of stability of the effect of the machine ¡earning model. The algorithm is compared with FedAvg, CDP, LDP algorithm in accuracy and loss rate which based on EMNIST data sets. In different global sensitivities, the model training effect is consistent with FedAvg algorithm while comparing with LDP algorithm, the speed of model convergence is improved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

TrustUQA: A Trustful Framework for Unified Structured data Question Answering

arXiv

引用

arXiv 2024年

作者： Zhang, Wen Jin, Long Zhu, Yushan Chen, Jiaoyan Huang, Zhiwei Wang, Junjie Hua, Yin Liang, Lei Chen, Huajun Zhejiang University China University of Manchester United Kingdom Ant Group China ZJU-Ant Group Joint Lab of Knowledge Graph China Zhejiang Key Laboratory of Big Data Intelligent Computing China

Natural language question answering (QA) over structured data sources such as tables and knowledge graphs have been widely investigated, especially with Large Language Models (LLMs) in recent years. The main solutions include question to formal query parsing and retrieval-based answer generation. However, current methods of the former often suffer from weak generalization, failing to dealing with multi-types of sources, while the later is limited in trustfulness. In this paper, we propose TrustUQA, a trustful QA framework that can simultaneously support multiple types of structured data in a unified way. To this end, it adopts an LLM-friendly and unified knowledge representation method called Condition Graph (CG), and uses an LLM and demonstration-based two-level method for CG querying. For enhancement, it is also equipped with dynamic demonstration retrieval. We have evaluated TrustUQA with 5 benchmarks covering 3 types of structured data. It outperforms 2 existing unified structured data QA methods. In comparison with the baselines that are specific to one data type, it achieves state-of-the-art on 2 of the datasets. Further more, we have demonstrated the potential of our method for more general QA tasks, QA over mixed structured data and QA across structured data. The code is available at https://***/zjukg/TrustUQA. Copyright © 2024, The Authors. All rights reserved.

关键词： Structured Query Language

来源：评论

学校读者我要写书评

暂无评论

RECC: A Relationship-Enhanced Content Caching Algorithm Using Deep Reinforcement Learning 6

RECC: A Relationship-Enhanced Content Caching Algorithm Usin...

引用

6th Asian Conference on Artificial Intelligence Technology, ACAIT 2022

作者： Ren, Jiarui Zhang, Haiyan Zhou, Xiaoping Zhu, Menghan Ningxia University School of Information Engineering Yinchuan China Beijing University of Civil Engineering and Architecture Beijing Key Laboratory of Intelligent Processing for Building Big Data Beijing China

ISBN: (纸本)9781665453110

Mobile edge caching (MEC) is a promising technology to alleviate traffic congestion in the network. Current studies explored deep reinforcement learning (DRL)-based MEC methods. These methods consider the dynamics of the request size to maximize the cache hit rate. However, they usually ignored the potential request relationships among contents. Two contents with a strong relationship are usually requested sequentially. Inspired by this assumption, this paper proposes a relationship-enhanced content caching algorithm using DRL, named RECC. Our RECC infers user preferences by mining the request relationships among contents. In this work, the relationships are modeled as request sequences, and the request features are learned by using graph embedding. These features will be used as input of state in our DRL-based algorithm. We utilize the Wolpertinger architecture to solve the limitation of large discrete action space. The simulation results indicate that our RECC outperformed the traditional cache policies and state-of-the-art DRL-based method in cache hit rate. Furthermore, the proposed RECC has advantages in long-term stability in the environment where content popularity changes dynamically, and also has a higher cache hit rate when handling the requests with number changes dynamically. © 2022 IEEE.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

Recognition of Disaster Images Based on Self-Supervised Learning 7

Recognition of Disaster Images Based on Self-Supervised Lear...

引用

7th International Conference on Cloud computing and big data Analytics, ICCCBDA 2022

作者： Wang, Linyong Wu, Ruiqi Li, Cen Zou, Zhiqiang Nanjing University of Posts and Telecommunications NanJing China College of Computer Nanjing University of Posts and Telecommunications Jiangsu Key Laboratory of Big Data Security and Intelligent Processing NanJing China

ISBN: (纸本)9781665487115

Existing supervised deep learning model requires large amounts of labeled training data to learn new tasks. This is a limitation for many practical applications in disaster areas as well as in many other fields such as medical domain, where numerous labour and time are required to label datasets. According to our literature survey, there is few self-supervised learning model with high accuracy for image recognition about natural disasters. This paper proposed a self-supervised learning model for disaster image recognition based on MoCo model, called MoCo-Q, including image preprocessing, model training and transfer learning after fine-Tuning. After being trained by learning a couple of features of disaster images without artificial labels, the trained neural network is applied to downstream tasks, which achieves the recognition of a large number of disaster images at a high accuracy without the manual intervention and labeling. With the implement of the MoCo-Q, images of natural disasters can be quickly recognized without manual labeling. The work can provide scientific decision-making support for the government and it is of great significance for natural disaster assessment and disaster relief. Finally, experiments on real-world natural disaster datasets, CrisisMMD and ASONAM17, are performed, and we compared with the state-of-The-Art supervised method (ResNet50) that uses same datasets. The result presents that the accuracy of the MoCo-Q (91.56%) is 5.71% higher than that of the supervised learning model (85.85%). © 2022 IEEE.

关键词： Training Image recognition Neural networks Transfer learning Supervised learning Training data Manuals

来源：评论

学校读者我要写书评

暂无评论

SNNLog: A Log Parsing Scheme with Siamese Network and Fixed Depth Tree in Networks

SNNLog: A Log Parsing Scheme with Siamese Network and Fixed ...

引用

International Conference on Computer Supported Cooperative Work in Design

作者： Lei Sun Xiaolong Xu Jiangsu Key Laboratory of Big Data Security & Intelligent Processing Nanjing University of Posts and Telecommunications Nanjing China

Syslog records critical information of network when the system is running, and has been used to help practitioners carry out various network maintenance and operation activities. Because of abundance of syslog, automated log analysis technology is needed to complete the above activities. Usually, log parsing is an important part of log analysis. Previous log parsing methods have good achievements, but they heavily rely on well-designed regular expressions and ignore semantic information of log. To solve these problems, we propose a novel log parsing method SNNLog, which uses Siamese Network to assess the similarity between log messages, and the similarity is used for the subsequent parsing. In addition, after parsing, SNNLog merges log similar events, reducing the misjudgment rate caused by non-numeric token variables at the beginning of messages. We evaluated on five publicly accessible log datasets and the results show that compared with SOTA algorithms such as LogMine, Spell, Drain and MoLFI, SNNLog has an F1-score of 0.999 on five datasets, and parsing accuracy of four datasets is the highest.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A label-guided contrastive capsule network for few-shot text classification

引用

Neural computing and Applications 2025年 1-26页

作者： Xu, Yang Xu, Xiaolong Jiangsu Key Laboratory of Big Data Security Intelligent Processing Nanjing University of Posts and Telecommunications Nanjing210023 China School of Computer Science Nanjing University of Posts and Telecommunications Nanjing210023 China

Few-shot text classification involves transferring knowledge from a limited dataset to perform classification tasks in unseen domains. Existing metric-based meta-learning models, such as prototypical networks, have shown to be effective. However, these models fail to separate embeddings of similar classes, resulting in performance stagnation. Meanwhile, traditional text classification recently has succeeded with the support of extensive prior knowledge from large language models. However, in few-shot scenarios, it is challenging to generate discriminative sentence embeddings using pre-trained models, often leading to overfitting issues. Furthermore, random support set sampling adversely affects prototype quality, resulting in inaccurate category representations. In this paper, we propose the label-guided contrastive capsule network (LGCCN) to address these issues. LGCCN introduces label information as external knowledge for each category, comprising a label-guided contrastive clustering module (LGCCM) and a label-guided prototype enhancement module (LGPEM). LGCCM employs label information to guide a contrastive clustering process, using labels as anchors to enhance the process, yielding discriminative sentence embeddings from the pre-trained model. LGPEM iteratively generates a robust prototype with label guidance, significantly reducing the impact of random support set sampling. The proposed LGCCN was thoroughly evaluated on seven benchmark datasets from diverse application domains. Experimental results demonstrate that LGCCN outperforms the state-of-the-art (SOTA) method by 9.7% in 1-shot tasks and 1.1% in 5-shot tasks on news datasets. Additionally, on intent recognition datasets, LGCCN surpasses the SOTA with a 3.3% improvement in 1-shot tasks and a 1.4% improvement in 5-shot tasks, highlighting its superior performance in few-shot text classification. © The Author(s), under exclusive licence to Springer-Verlag London Ltd., 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

StyleMI: An Image Processing Based Method for Detecting Unauthorized Style Mimicry in Fine-tuned Diffusion Models in a More Realistic Scenario

引用

IEEE Access 2025年 13卷 97364-97375页

作者： Zhang, Xinjun Zuo, Shixin Sang, Yu Liaoning Technical University School of Electronics and Information Engineering Huludao China Liaoning Technical University Liaoning Key Laboratory of Radio Frequency and Big Data for Intelligent Applications Liaoning Huludao125105 China

The rapid development of diffusion models and model fine-tuning methods have enabled widespread applications in artistic style mimicry while also leading to significant concerns about copyright infringement. Existing approaches protect artists from unauthorized style mimicry by introducing adversarial examples, watermarks, or membership inference attacks. However, the former two are vulnerable to image pre-processing and unable to be applied to published artworks, while the latter requires access to the target model, which is typically inaccessible in real-world scenarios. To this end, the task of detecting unauthorized style mimicry without access to the target model is formulated into a novel membership inference problem. The key insight is that images generated by a diffusion model inherently carry information of its training dataset. A style comparison-based method is proposed, which consists of a LoRA-based style transfer module and a CLIP-based style extraction and comparison module. The style transfer module first learns the artist's artistic style, then removes the style of the given image to avoid its influence, and finally transfers the artist's style to the given image. The style extraction and comparison module utilizes a diffusion model to extract the content information of an image and then remove it in a latent semantic space with the pre-trained CLIP model to acquire style-related features and compare them. Experiments on Stable Diffusion demonstrate the effectiveness of the proposed method. The best result achieved a True Positive Rate of 85%, a False Positive Rate of 0%, an Attack Success Rate of 99.29%, a Precision of 100%, and an Area Under the Curve of 1. © 2013 IEEE.

关键词： Photointerpretation

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：