检索结果-内蒙古大学图书馆

arXiv 2025年

作者： Wang, Weizhen He, Jianping Duan, Xiaoming Department of Automation Shanghai Jiao Tong University Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai China

Policy gradient methods are one of the most successful methods for solving challenging reinforcement learning problems. However, despite their empirical successes, many SOTA policy gradient algorithms for discounted problems deviate from the theoretical policy gradient theorem due to the existence of a distribution mismatch. In this work, we analyze the impact of this mismatch on the policy gradient methods. Specifically, we first show that in the case of tabular parameterizations, the methods under the mismatch remain globally optimal. Then, we extend this analysis to more general parameterizations by leveraging the theory of biased stochastic gradient descent. Our findings offer new insights into the robustness of policy gradient methods as well as the gap between theoretical foundations and practical implementations. © 2025, CC BY.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Complementary Learning Subnetworks Towards Parameter-Efficient Class-Incremental Learning

引用

IEEE Transactions on Knowledge and Data Engineering 2025年第6期37卷 3240-3252页

作者： Li, Depeng Zeng, Zhigang Dai, Wei Suganthan, Ponnuthurai Nagaratnam Huazhong University of Science and Technology School of Artificial Intelligence and Automation Wuhan430074 China Key Laboratory of Image Processing and Intelligent Control of Education Ministry of China Wuhan430074 China China University of Mining and Technology School of Information and Control Engineering Xuzhou221116 China Qatar University Kindi Center for Computing Research College of Engineering Doha2713 Qatar

In the scenario of class-incremental learning (CIL), deep neural networks have to adapt their model parameters to non-stationary data distributions, e.g., the emergence of new classes over time. To mitigate the catastrophic forgetting phenomenon, typical CIL methods either cumulatively store exemplars of old classes for retraining model parameters from scratch or progressively expand model size as new classes arrive, which, however, compromises their practical value due to little attention paid to parameter efficiency. In this paper, we contribute a novel solution, effective control of the parameters of a well-trained model, by the synergy between two complementary learning subnetworks. Specifically, we integrate one plastic feature extractor and one analytical feed-forward classifier into a unified framework amenable to streaming data. In each CIL session, it achieves non-overwritten parameter updates in a cost-effective manner, neither revisiting old task data nor extending previously learned networks;Instead, it accommodates new tasks by attaching a tiny set of declarative parameters to its backbone, in which only one matrix per task or one vector per class is kept for knowledge retention. Experimental results on a variety of task sequences demonstrate that our method achieves competitive results against state-of-the-art CIL approaches, especially in accuracy gain, knowledge transfer, training efficiency, and task-order robustness. Furthermore, a graceful forgetting implementation on previously learned trivial tasks is empirically investigated to make its non-growing backbone (i.e., a model with limited network capacity) suffice to train on more incoming tasks. © 2025 IEEE.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

A Synthetic Inertia control Scheme for Inverter Utilizing Its DC Side Capacitor 2nd

A Synthetic Inertia Control Scheme for Inverter Utilizing It...

引用

2nd International Conference on Smart Electrical Grid and Renewable Energy, SEGRE 2024

作者： Zhang, Mingmin Luo, Ziyi Chen, Wanxing Yu, Jiaqi Guo, Wenming Li, Yong Hunan Province University Key Laboratory of Energy Storage Power System Cyber-Physical Control Changsha University Changsha China College of Electrical and Information Engineering Hunan University Changsha China

ISBN: (纸本)9789819624553

Insufficient inertia has retrained the progress toward a one hundred percent inverter-based power generation. To solve this issue, a promising way is to exploit synthetic inertia based on the power electronic inverters. Taking advantage of energy stored in the DC side capacitor, this paper proposes a synthetic inertia control scheme for inverters without a DC side battery. The basic mechanism of the synthetic inertia control is derived by comparing the working principle between a synchronous generator and a three-phase inverter. Thereafter, the control algorithm of the inverter is designed by three parts, including synthetic inertia control loop, reactive power control loop and line oscillation damping control loop. Finally, simulation results based on a simplified microgrid are presented to verify the self-grid synchronization and grid-supportive characteristics of the inverter. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Synchronous generators

来源：评论

学校读者我要写书评

暂无评论

Lightweight Single Image Super-Resolution Network Integrating CNN and Transformer 27th

Lightweight Single Image Super-Resolution Network Integratin...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Zhu, Kai Chen, Li School of Computer Science and Technology Wuhan University of Science and Technology Wuhan China Hubei Province Key Laboratory of Intelligent Information Processing and Real-time Industrial System Wuhan University of Science and Technology Wuhan China

ISBN: (纸本)9783031781247

Recently, remarkable progress has been achieved in single image super-resolution using methods based on CNN and Transformer architectures. However, existing approaches often construct a substantial number of network layers, leading to a significant increase in performance requirement and memory consumption, thereby limiting the practical deployment and usability of the models. To address this issue, we propose an Alternating CNN Transformer Block and an Integrative CNN Efficient Transformer for single image super-resolution. We enhance feature extraction efficiency by combining CNN within and between Transformer modules. In addition, we propose two novel structures: Multi-branch Gated CNN and Parallel Channel Attention, aiming to efficiently extract local spatial information and global channel information from images. Extensive experiments demonstrate that our model achieves high performance while maintaining low model complexity. The proposed model attains PSNR values of 32.32 and 30.78 for the Set5 and Manga109 benchmark datasets, respectively, at a scale factor of ×4. Compared to other lightweight super-resolution models, our proposed model outperforms them at lower computational costs. The source codes are available at https://***/kylechuuuuu/ICTSRN. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： HTTP

来源：评论

学校读者我要写书评

暂无评论

Cross-Modal Ship Grounding: Towards Large Model for Enhanced Few-Shot Learning 27th

Cross-Modal Ship Grounding: Towards Large Model for Enhanced...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Hu, Quan Chen, Li Feng, Zhida Chen, Yaojie School of Computer Science and Technology Wuhan University of Science and Technology Wuhan China Hubei Province Key Laboratory of Intelligent Information Processing and Real-Time Industrial System Wuhan University of Science and Technology Wuhan China

ISBN: (纸本)9783031781124

A growing body of research indicates that employing large models for adaptation to downstream tasks often yields remarkable performance. However, in the domain of ship detection, the potential of these large models is frequently underutilized due to domain shift issues. This paper introduces the Cross-Modal Ship Grounding (CSG) model, which leverages an efficient Cross-Modal Adapter (CMA) technology to transfer the general detection capabilities of large models to ship images, addressing domain shift with minimal training costs. To mitigate the challenges posed by complex and variable background interference, the Water-Land Separation (WLS) module is proposed to focus specifically on the water area. This module effectively addresses the issue of background target interference, thereby enhancing the model’s accuracy in complex scenes. Empirical evaluations on both private and public datasets demonstrate that the CSG model surpasses all state-of-the-art models in performance. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Zero-shot learning

来源：评论

学校读者我要写书评

暂无评论

Ocean archaea PPI prediction with pretraining models 25

Ocean archaea PPI prediction with pretraining models

引用

Proceedings of the 2025 5th International Conference on Bioinformatics and Intelligent Computing

作者： Ying Zhang Yuan Liu Xiaoyong Pan Hongbin Shen Institute of Image Processing and Pattern Recognition Shanghai Jiao Tong University Shanghai China Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai China

ISBN: (纸本)9798400712203

Protein-Protein Interaction (PPI) provides important insights into the metabolic mechanisms of different biological processes. Although PPIs in some organisms have been investigated systematically, PPIs in the ocean archaea remain largely unexplored. But such species have special investigation value since their adaptation to extreme living conditions may generate unique PPIs. In this paper, we aim to characterize and predict PPIs in ocean archaea to advance understanding of their metabolic networks. First, we collect all ocean archaea PPIs with high confidence from STRING database and analyze the PPI network features, including centrality and enrichment analysis. The functional enrichment results of the largest connecting subgraph in the PPI network show most PPIs in our constructed dataset is related to the translation and transcription processes. Then, we generate an equal number of negative PPI pairs, whose members have either different subcellular locations or GO terms. We also use the generated dataset to test the performance of three pretraining methods and their ensemble methods in the binary PPI prediction task. Our results suggest the ensemble methods could be applied to further improve models’ performance. Fine-tuned models trained on the ocean archaea dataset are expected to predict the other ocean archaea PPIs that are not included in the STRING database and get more understanding about the ocean archaea PPI universe.

关键词： Binary PPI prediction

来源：评论

学校读者我要写书评

暂无评论

GENERATIVE ADVERSARIAL NETWORKS FOR REMOTE SENSING IMAGE DEHAZING WITH COLOR FEATURE RESTORATION

引用

International Journal of Innovative Computing, information and control 2025年第2期21卷 323-338页

作者： Zhao, Liquan Qin, Yuqing Jia, Yanfei Key Laboratory of Modern Power System Simulation and Control and Renewable Energy Technology Ministry of Education Northeast Electric Power University No. 169 Changchun Road Jilin132012 China College of Electric and Information Engineering Beihua University No. 3999 Beijing Road Jilin132013 China

Remote sensing images are often affected by atmospheric factors such as haze during the acquisition process, resulting in blurring and low contrast in the collected remote sensing images. This problem impacts the image quality, thereby affecting the analysis of remote sensing images. To mitigate the impact of haze on remote sensing images, a generative adversarial network is proposed. It comprises a generative network and an adversarial network. Firstly, a novel feature extraction module is designed to enhance the capability of extracting useful information from remote sensing images. It enables the network to focus more on regions with dense haze, allowing it to extract more important information while filtering out irrelevant details. Secondly, a residual attention module is designed which can allocate different weights based on varying haze density in the feature map. This module readjusts features outputted by the encoder, facilitating better image restoration. Thirdly, a multi-scale module is also incorporated to extract feature information across various image scales. Lastly, a color feature extraction module is designed to extract color features. The novel feature extraction module, residual attention module, multi-scale module, and color feature extraction module are utilized for constructing the generative network. Besides, an adversarial network is also designed to indirectly enhance the dehazing capability of the generative network. Synthetic and real datasets are used to test six different methods for dehazing remote sensing images, respectively. The proposed method achieves higher PSNR, SSIM, and lower MSE on the synthetic remote sensing dataset. On the other hand, it achieves lower PIQE, BRISQUE, and higher MetaIQA on the real remote sensing dataset. The proposed method has best performance in dehazing remote sensing images than other methods. © 2025, Int. J. Innov. Comput. Inf. control. All rights reserved.

关键词： Generative adversarial networks

来源：评论

学校读者我要写书评

暂无评论

Investigation of the Effect of Bonding Wires Degradation on Switching Stress Waves Released During Power Cycling in Discrete IGBT

引用

IEEE Journal of Emerging and Selected Topics in Power Electronics 2025年第2期13卷 2057-2069页

作者： Geng, Xuefeng He, Yunze Tang, Longhai Chang, Shan Zhang, Jie Yang, Wenxue Yuan, Man Wang, Guangxin Li, Qiying Ping, Yang Huang, Songling Hunan University College of Electrical and Information Engineering Changsha410082 China Tsinghua University State Key Laboratory of Power System Operation and Control Beijing100084 China Hunan University Shenzhen Research Institute Shenzhen518000 China Tsinghua University Department of Electrical Engineering Beijing100084 China

Effective condition monitoring (CM) technology ensures the safe operation of insulated gate bipolar transistors (IGBTs). Most CM methods for IGBTs focus on extracting electrical, thermal, and magnetic parameters. Recent studies indicate that at the switching moment, IGBTs can emit the switching stress wave (SSW) that is detectable by the acoustic emission (AE) sensors. However, using these SSWs to monitor the health status of IGBTs, even for discrete devices with relatively simple package structures, remains challenging. This article investigates how SSWs from discrete IGBTs during power cycling change as bonding wires degrade and whether these waves are promising for monitoring the condition of internal bonding wires. In this work, a power cycle test platform and an SSW test platform for discrete IGBTs are built, and SSWs from discrete IGBTs during power cycling are analyzed. Results show that SSWs vary with the degradation of the device’s bonding wires, with changes in the time domain indicator of signal energy, reflecting SSW strength, and the frequency domain indicator of the center of gravity (COG) frequency, reflecting frequency shifts. These findings suggest the possibility for developing AE-based detection technology (AEDT) in CM of the IGBT. © 2013 IEEE.

关键词： Insulated gate bipolar transistors (IGBT)

来源：评论

学校读者我要写书评

暂无评论

Research on Medical Image Classification for Oral Cancer 3rd

Research on Medical Image Classification for Oral Cancer

引用

3rd International Conference on Machine Learning, Cloud Computing and Intelligent Mining, MLCCIM 2024

作者： Wang, Kunkun Shi, Linjun Gu, Chaochen Mo, Fan Department of Automation Shanghai Jiao Tong University Shanghai China Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai China Department of Oral Mucosal Diseases Ninth People’s Hospital Shanghai Jiaotong University School of Medicine Shanghai China Shanghai Key Laboratory of Stomatology and Shanghai Research Institution of Stomotology National Clinical Research Center of Stomatology Shanghai China

ISBN: (纸本)9789819624676

In recent years, deep learning has made significant progress in medical imaging, deepening the crossover between the medical and industrial fields. However, not all medical images are suitable for deep learning neural networks;simultaneously, some medical images are difficult to collect and preprocess. In this paper, we reconstruct the image information collected by dental clinicians over many years, normalizing it with diagnostic results as the gold standard. On this basis, we explore the applicability and potential of deep learning on complex medical images. Further, we compare the performance of multiple models on this dataset and visualize the learning outcomes, demonstrating that deep learning networks exhibit good learning capabilities on certain images. © The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd. 2025.

关键词： Contrastive Learning

来源：评论

学校读者我要写书评

暂无评论

Joint Sensing and Computation Incentive Mechanism for Mobile Crowdsensing Networks: A Multiagent Reinforcement Learning Approach

引用

IEEE Internet of Things Journal 2025年第9期12卷 13033-13046页

作者： Zhao, Nan Sun, Yiling Pei, Yiyang Niyato, Dusit Hubei University of Technology Hubei Key Laboratory for High-Efficiency Utilization of Solar Energy and Operation Control of Energy Storage System Wuhan430061 China Singapore Institute of Technology Information and Communications Technology Cluster Tampines Singapore Nanyang Technological University School of Computer Science and Engineering Jurong West Singapore

Mobile crowdsensing (MCS) is a novel sensing paradigm by utilizing mobile users (MUs) to collect data from environment. Considering the finite sensing and computing resources of MUs, it is crucial to inspire MUs to take part in crowdsensing willingly. In this study, a multiagent-deep-reinforcement-learning (DRL)-based incentive mechanism is investigated to tackle the joint data sensing and computing issues. Specifically, due to the heterogeneity of sensing tasks, multiple MCS platforms (MCPs) motivate MUs to participate in different tasks. The interaction between MCPs and MUs is modeled as a multileader-multifollower Stackelberg game with Stackelberg equilibrium proved by derivation. Moreover, the Stackelberg game is transformed as a Markov decision process (MDP) to deal with a multiagent DRL method without any prior knowledge. Due to the continuous high-dimensional action space of multiple MCPs and MUs, a multiagent double actors deep deterministic policy gradient (MA-DADDPG) algorithm is proposed to obtain the optimal sensing data size, computing resource, and incentive payment policies. Extensive simulation results illustrate the effectiveness of the proposed crowdsensing incentive mechanism. © 2014 IEEE.

关键词： Deep reinforcement learning

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：