检索结果-南通市图书馆

Federated Meta Reinforcement Learning for Personalized Tasks

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Tsinghua science and Technology 2024年第3期29卷 911-926页

作者： Wentao Liu Xiaolong Xu Jintao Wu Jielin Jiang School of Computer science Nanjing University of Information Science and TechnologyNanjing 210044China School of Software Nanjing University of Information Science and TechnologyNanjing 210044China

As an emerging privacy-preservation machine learning framework,Federated Learning(FL)facilitates different clients to train a shared model collaboratively through exchanging and aggregating model parameters while raw data are kept local and private.When this learning framework is applied to Deep Reinforcement Learning(DRL),the resultant Federated Reinforcement Learning(FRL)can circumvent the heavy data sampling required in conventional DRL and benefit from diversified training data,besides privacy preservation offered by FL.Existing FRL implementations presuppose that clients have compatible tasks which a single global model can cover.In practice,however,clients usually have incompatible(different but still similar)personalized tasks,which we called task shift.It may severely hinder the implementation of FRL for practical applications.In this paper,we propose a Federated Meta Reinforcement Learning(FMRL)framework by integrating Model-Agnostic Meta-Learning(MAML)and FRL.Specifically,we innovatively utilize Proximal Policy Optimization(PPO)to fulfil multi-step local training with a single round of sampling.Moreover,considering the sensitivity of learning rate selection in FRL,we reconstruct the aggregation optimizer with the Federated version of Adam(Fed-Adam)on the server side.The experiments demonstrate that,in different environments,FMRL outperforms other FL methods with high training efficiency brought by Fed-Adam.

关键词： federated learning reinforcement learning meta-learning personalization

Quantum geometric tensor and the topological characterization of the extended Su-Schrieffer-Heeger model

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Chinese Physics B 2024年第3期33卷 260-265页

作者：曾相龙赖文喜魏祎雯马余全 School of science Beijing Information Science and Technology UniversityBeijing 100192China

We investigate the quantum metric and topological Euler number in a cyclically modulated Su-Schrieffer-Heeger(SSH)model with long-range hopping terms.By computing the quantum geometry tensor,we derive exact expressions for the quantum metric and Berry curvature of the energy band electrons,and we obtain the phase diagram of the model marked by the first Chern number.Furthermore,we also obtain the topological Euler number of the energy band based on the Gauss-Bonnet theorem on the topological characterization of the closed Bloch states manifold in the first Brillouin zone.However,some regions where the Berry curvature is identically zero in the first Brillouin zone result in the degeneracy of the quantum metric,which leads to ill-defined non-integer topological Euler numbers.Nevertheless,the non-integer"Euler number"provides valuable insights and an upper bound for the absolute values of the Chern numbers.

关键词： quantum geometric tensor topological Euler number Chern number Berry curvature quantum metric Su-Schrieffer-Heeger(SSH)model

维普期刊数据库

Research on a multi‑dimensional image information fusion algorithm based on NSCT transform

在线全文

学校读者我要写书评

暂无评论

Frontiers of Optoelectronics 2024年第1期17卷 29-44页

作者： Yuxiang Su Xi Liang Danhua Cao Zhenyu Yang Yuanlong Peng Ming Zhao School of Optical and Electronic information Huazhong University of Science and TechnologyWuhan 430074China State Grid information and Telecommunication Branch Beijing 100761China

Traditional inspection cameras determine targets and detect defects by capturing images of their light intensity,but in complex environments,the accuracy of inspection may decrease.information based on polarization of light can characterize various features of a material,such as the roughness,texture,and refractive index,thus improving classification and recognition of targets.This paper uses a method based on noise template threshold matching to denoise and preprocess polarized images.It also reports on design of an image fusion algorithm,based on NSCT transform,to fuse light intensity images and polarized images.The results show that the fused image improves both subjective and objective evaluation indicators,relative to the source image,and can better preserve edge information and help to improve the accuracy of target recognition.This study provides a reference for the comprehensive application of multi-dimensional optical information in power inspection.

关键词： Power inspection Object detection Polarization imaging Image fusion Image denoising

Modifying Lyapunov exponent of chaotic map by self-cascading

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

science China Technological sciences 2024年第7期67卷 2203-2214页

作者： YI ChenLong LI ChunBiao LI YongXin XIA Ming HUA ZhongYun School of Electronic and information Engineering Nanjing University of Information Science & Technology School of Artificial Intelligence Nanjing University of Information Science & Technology School of Computer science and Technology Harbin Institute of Technology Shenzhen Graduate School

The self-cascade(SC) method is an effective technique for chaos enhancement and complexity increasing in chaos maps.Additionally, the controllable self-cascade(CSC) method allows for more accurate control of Lyapunov exponents of the discrete map. In this work, the SC and CSC systems of the original map are derived, which enhance the chaotic performance while preserving the fundamental dynamical characteristics of the original map. Higher Lyapunov exponent of chaotic sequences corresponding to higher frequency are obtained in SC and CSC systems. Meanwhile, the Lyapunov exponent could be linearly controlled with greater flexibility in the CSC system. The verification of the numerical simulation and theoretical analysis is carried out based on the platform of CH32.

关键词： self-cascade controllable self-cascade chaos enhancement chaos control

Energy consumption optimization for edge computing-supported cellular networks based on optimal transport theory

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

science China(information sciences) 2024年第2期67卷 201-218页

作者： Xiangyu LV Xiaohu GE Yi ZHONG Qiang LI Yong XIAO School of Electronic information and Communications Huazhong University of Science and Technology

With advancements in mobile communication technologies, there have been multiple user device(UD) connections to cellular networks. Because of changes in the spatial distribution of UDs and application requirements, the traditional offloading mechanism based on the nearest distance will result in heavy loads for some base stations(BSs). Because there is a high-order relationship between the energy consumption of the BS processing tasks and the number of computing tasks, the traditional offloading mechanism will result in high energy consumption, and UD’s offloading decision must be dynamically adjusted. The offloading decision of UDs is optimized based on detailed information about various parameters associated with a network from the standpoint of distribution. This information includes details regarding the spatial distribution of UDs,application requirements, and the offloading period of computing tasks. Based on optimal transport theory,an energy consumption optimization algorithm is suggested to lower the total amount of energy consumed by the offloading process of UDs’ computing tasks by reasonably planning the offloading BSs of the UDs in the networks. The simulation results show that the proposed offloading mechanism based on energy consumption optimization reduces the total energy consumption of the offloading process by 28.09% when compared to the traditional offloading mechanism, and the traffic managed by each BS is balanced.

关键词： optimal transport theory edge computing energy consumption reduction dynamic voltage frequency scaling

YOLOv5ST:A Lightweight and Fast Scene Text Detector

在线全文

同方期刊数据库

学校读者我要写书评

暂无评论

Computers, Materials & Continua 2024年第4期79卷 909-926页

作者： Yiwei Liu Yingnan Zhao Yi Chen Zheng Hu Min Xia School of Computer and science Nanjing University of Information Science and TechnologyNanjing210044China School of Automation Nanjing University of Information Science and TechnologyNanjing210044China

Scene text detection is an important task in computer vision.In this paper,we present YOLOv5 Scene Text(YOLOv5ST),an optimized architecture based on YOLOv5 v6.0 tailored for fast scene text detection.Our primary goal is to enhance inference speed without sacrificing significant detection accuracy,thereby enabling robust performance on resource-constrained devices like drones,closed-circuit television cameras,and other embedded systems.To achieve this,we propose key modifications to the network architecture to lighten the original backbone and improve feature aggregation,including replacing standard convolution with depth-wise convolution,adopting the C2 sequence module in place of C3,employing Spatial Pyramid Pooling Global(SPPG)instead of Spatial Pyramid Pooling Fast(SPPF)and integrating Bi-directional Feature Pyramid Network(BiFPN)into the neck.Experimental results demonstrate a remarkable 26%improvement in inference speed compared to the baseline,with only marginal reductions of 1.6%and 4.2%in mean average precision(mAP)at the intersection over union(IoU)thresholds of 0.5 and 0.5:0.95,respectively.Our work represents a significant advancement in scene text detection,striking a balance between speed and accuracy,making it well-suited for performance-constrained environments.

关键词： Scene text detection YOLOv5 lightweight object detection

A Cover-Independent Deep Image Hiding Method Based on Domain Attention Mechanism

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Computers, Materials & Continua 2024年第3期78卷 3001-3019页

作者： Nannan Wu Xianyi Chen James Msughter Adeke Junjie Zhao School of Computer and Software Nanjing University of Information Science and TechnologyNanjing210044China School of Electronics and information Engineering Nanjing University of Information Science and TechnologyNanjing210044China

Recently,deep image-hiding techniques have attracted considerable attention in covert communication and high-capacity information hiding.However,these approaches have some limitations.For example,a cover image lacks self-adaptability,information leakage,or weak concealment.To address these issues,this study proposes a universal and adaptable image-hiding method.First,a domain attention mechanism is designed by combining the Atrous convolution,which makes better use of the relationship between the secret image domain and the cover image domain.Second,to improve perceived human similarity,perceptual loss is incorporated into the training process.The experimental results are promising,with the proposed method achieving an average pixel discrepancy(APD)of 1.83 and a peak signal-to-noise ratio(PSNR)value of 40.72 dB between the cover and stego images,indicative of its high-quality output.Furthermore,the structural similarity index measure(SSIM)reaches 0.985 while the learned perceptual image patch similarity(LPIPS)remarkably registers at 0.0001.Moreover,self-testing and cross-experiments demonstrate the model’s adaptability and generalization in unknown hidden spaces,making it suitable for diverse computer vision tasks.

关键词： Deep image hiding attention mechanism privacy protection data security visual quality

A Review of NILM Applications with Machine Learning Approaches

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Computers, Materials & Continua 2024年第5期79卷 2971-2989页

作者： Maheesha Dhashantha Silva Qi Liu School of Electronic and information Engineering Nanjing University of Information Science and TechnologyNanjing210044China

In recent years,Non-Intrusive LoadMonitoring (NILM) has become an emerging approach that provides affordableenergy management solutions using aggregated load obtained from a single smart meter in the power grid.Furthermore, by integrating Machine Learning (ML), NILM can efficiently use electrical energy and offer less ofa burden for the energy monitoring process. However, conducted research works have limitations for real-timeimplementation due to the practical issues. This paper aims to identify the contribution of ML approaches todeveloping a reliable Energy Management (EM) solution with NILM. Firstly, phases of the NILM are discussed,along with the research works that have been conducted in the domain. Secondly, the contribution of machinelearning approaches in three aspects is discussed: Supervised learning, unsupervised learning, and hybridmodeling.It highlights the limitations in the applicability of ML approaches in the field. Then, the challenges in the realtimeimplementation are concerned with six use cases: Difficulty in recognizing multiple loads at a given time,cost of running the NILM system, lack of universal framework for appliance detection, anomaly detection andnew appliance identification, and complexity of the electricity loads and real-time demand side management.Furthermore, options for selecting an approach for an efficientNILMframework are suggested. Finally, suggestionsare provided for future research directions.

关键词： Non-intrusive load monitoring transfer learning machine learning feature extraction

Expression Complementary Disentanglement Network for Facial Expression Recognition

在线全文

维普期刊数据库

学校读者我要写书评

暂无评论

Chinese Journal of Electronics 2024年第3期33卷 742-752页

作者： Shanmin WANG Hui SHUAI Lei ZHU Qingshan LIU College of Computer science and Technology Nanjing University of Aeronautics and Astronautics College of Computer and Software Nanjing University of Information Science and Technology School of information and Control Nanjing University of Information Science and Technology

Disentangling facial expressions from other disturbing facial attributes in face images is an essential topic for facial expression recognition.Previous methods only care about facial expression disentanglement(FED)itself,ignoring the negative effects of other facial attributes.Due to the annotations on limited facial attributes,it is difficult for existing FED solutions to disentangle all disturbance from the input face.To solve this issue,we propose an expression complementary disentanglement network(ECDNet).ECDNet proposes to finish the FED task during a face reconstruction process,so as to address all facial attributes during disentanglement.Different from traditional reconstruction models,ECDNet reconstructs face images by progressively generating and combining facial appearance and matching geometry.It designs the expression incentive(EIE) and expression inhibition(EIN) mechanisms,inducing the model to characterize the disentangled expression and complementary parts precisely.Facial geometry and appearance,generated in the reconstructed process,are dealt with to represent facial expressions and complementary parts,respectively.The combination of distinctive reconstruction model,EIE,and EIN mechanisms ensures the completeness and exactness of the FED task.Experimental results on RAF-DB,AffectNet,and CAER-S datasets have proven the effectiveness and superiority of ECDNet.

关键词： Geometry Image recognition Annotations Face recognition Task analysis Image reconstruction Facial features