检索结果-内蒙古大学图书馆

12th International Conference on Learning Representations, ICLR 2024

作者： Ren, Li Chen, Chen Wang, Liqiang Hua, Kien Department of Computer Science University of Central Florida United States Center for Research in Computer Vision University of Central Florida United States

Deep Metric Learning (DML) has long attracted the attention of the machine learning community as a key objective. Existing solutions concentrate on fine-tuning the pre-trained models on conventional image datasets. As a result of the success of recent pre-trained models trained from larger-scale datasets, it is challenging to adapt the model to the DML tasks in the local data domain while retaining the previously gained knowledge. In this paper, we investigate parameter-efficient methods for fine-tuning the pre-trained model for DML tasks. In particular, we propose a novel and effective framework based on learning Visual Prompts (VPT) in the pre-trained vision Transformers (ViT). Based on the conventional proxy-based DML paradigm, we augment the proxy by incorporating the semantic information from the input image and the ViT, in which we optimize the visual prompts for each class. We demonstrate that our new approximations with semantic information are superior to representative capabilities, thereby improving metric learning performance. We conduct extensive experiments to demonstrate that our proposed framework is effective and efficient by evaluating popular DML benchmarks. In particular, we demonstrate that our fine-tuning method achieves comparable or even better performance than recent state-of-the-art full fine-tuning works of DML while tuning only a small percentage of total parameters. Code is available at https://***/Noahsark/ParameterEfficient-DML. © 2024 12th International Conference on Learning Representations, ICLR 2024. All rights reserved.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

Learning Auxiliary Representations with Inconsistency-Guided Detail Regularization for Mask-Guided Matting

引用

IEEE Transactions on Multimedia 2025年 27卷 3625-3636页

作者： Jiang, Weihao Xie, Zhaozhi Lu, Yuxiang Qi, Longjie Cai, Jingyong Uchiyama, Hiroyuki Chen, Bin Ding, Yue Lu, Hongtao Shanghai Jiao Tong University Department of Computer Science and Engineering Shanghai200240 China OPPO Japan Research Center Kanagawa2200012 Japan

Mask-guided matting networks have achieved significant improvements and have shown great potential in practical applications in recent years. However, simply learning matting representation from synthetic and lack-of-real-world-diversity matting data, these approaches tend to overfit low-level details in wrong regions, lack generalization to objects with complex structures and real-world scenes such as shadows, as well as suffer from interference of background lines or textures. To address these challenges, in this paper, we propose a novel auxiliary learning framework for mask-guided matting models, incorporating three auxiliary tasks: semantic segmentation, edge detection, and background line detection besides matting, to learn different and effective auxiliary representations from different types of data and annotations. Our framework and model introduce the following key aspects: (1) to learn real-world adaptive semantic representation for objects with diverse and complex structures under real-world scenes, we introduce extra semantic segmentation and edge detection tasks on more diverse real-world data with segmentation annotations;(2) to avoid overfitting on low-level details, we propose a module to utilize the inconsistency between learned segmentation and matting representations to regularize detail refinement;(3) we propose a novel background line detection task into our auxiliary learning framework, to suppress interference of background lines or textures. In addition, we propose a high-quality matting benchmark, Plant-Mat, to evaluate matting methods on complex structures. Extensively quantitative and qualitative results show that our approach outperforms state-of-the-art mask-guided methods. © 1999-2012 IEEE.

关键词： Semantic Segmentation

来源：评论

学校读者我要写书评

暂无评论

Early Parkinson's Disease Detection and Classification using Machine Learning and Neutrosophic Sets

引用

Neutrosophic Sets and Systems 2025年 82卷 873-886页

作者： Oudah, Hasan H. Metwaly, Ahmed A. Eassa, Mohamed Abdelhafeez, Ahmed Nagm, Ahmad M. Salama, Ahmed S. College of Arts Department of Journalism Ahl Al Bayt University Karbala Iraq Department of Computer Science Faculty of Computers and Informatics Zagazig University Zagazig44519 Egypt Computer Science Department Faculty of Information System and Computer Science October 6 University Giza12585 Egypt Applied Science Research Center Applied Science Private University Amman Jordan Department of Computer Engineering and Electronics Cairo Higher Institute for Engineering Computer Science and Management New Cairo Egypt

Optimizing therapy and rehabilitation for Parkinson's disease (PD) requires early identification and precise evaluation of the illness's course. However, there is disagreement about the best way to use gait analysis to categorize the severity of motor symptoms and identify early-stage Parkinson's disease. The precision of machine learning (ML) models in identifying early and intermediate stages of Parkinson's disease was assessed in this study. Six ML models are used in this study for the prediction of PD. Different metrics are used to evaluate ML models such as accuracy, precision, recall, f1-score, and AUC score. Then we propose a multi-criteria decision-making methodology (MCDM) to evaluate the ML models and select the best one. Two MCDM methods are used such as CRITIC method to compute the criteria weights and the TOPSIS method to rank the alternatives. These methods are used under the bipolar neutrosophic sets (BNSs) to deal with uncertainty information. The results show the support vector machine is the best ML model for the prediction of PD. © 2025, University of New Mexico. All rights reserved.

关键词： Neurodegenerative diseases

来源：评论

学校读者我要写书评

暂无评论

Deep learning applications in the Internet of Things: a review, tools, and future directions

引用

Evolutionary Intelligence 2024年第5-6期17卷 3621-3654页

作者： Raoufi, Parisa Hemmati, Atefeh Rahmani, Amir Masoud Department of Computer Engineering Central Tehran Branch Islamic Azad University Tehran Iran Department of Computer Engineering Science and Research Branch Islamic Azad University Tehran Iran Future Technology Research Center National Yunlin University of Science and Technology Yunlin64002 Taiwan

The emergence of the Internet of Things (IoT) has enabled the proliferation of interconnected devices and sensors, generating vast amounts of often complex and unstructured data. Deep learning (DL), a subfield of machine learning (ML), has shown great promise in addressing the challenges of processing and analyzing such data. Considering the increasing importance of DL and data analysis, we decided to review the articles of the last few years in this field to pave the way for researchers. In this article, we used the systematic literature review (SLR) method, and in line with that, we selected and analyzed 56 articles published from 2019 to April 2024. We first discuss the DL models used in the IoT field and clarify their specific use cases. Secondly, we outline an analysis of research areas in DL-based IoT. In addition, our research extends to the tools and simulators used to evaluate studies in the DL-based IoT domain. We also examine the DL-based IoT research datasets. Finally, our review identifies future directions and open issues in DL-based IoT. We aim to contribute to an accurate understanding of the current state, challenges, and potential breakthroughs at the intersection of DL and the IoT. © The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature 2024.

关键词： Internet of things

来源：评论

学校读者我要写书评

暂无评论

IRS-enabled NOMA communication systems:A network architecture primer with future trends and challenges

引用

Digital Communications and Networks 2024年第5期10卷 1503-1528页

作者： Haleema Sadia Ahmad Kamal Hassan Ziaul Haq Abbas Ghulam Abbas Muhammad Waqas Zhu Han Faculty of Electrical Engineering and Telecommunications and Networking(TeleCoN)Research Center Ghulam Ishaq Khan Institute of Engineering and Technology(GIKI)Topi 23640Pakistan Faculty of Computer Science and Engineering GIKITopi 23640Pakistan Department of Computer Engineering College of Information TechnologyUniversity of Bahrain32028Bahrain School of Engineering Edith Cowan UniversityPerth WA6027Australia Department of Electrical and Computer Engineering University of HoustonHoustonTX 77004USA Department of Computer Science and Engineering Kyung Hee UniversitySeoul 446-701South Korea

Non-Orthogonal Multiple Access(NOMA)has already proven to be an effective multiple access scheme for5th Generation(5G)wireless *** provides improved performance in terms of system throughput,spectral efficiency,fairness,and energy efficiency(EE).However,in conventional NOMA networks,performance degradation still exists because of the stochastic behavior of wireless *** combat this challenge,the concept of Intelligent Reflecting Surface(IRS)has risen to prominence as a low-cost intelligent solution for Beyond 5G(B5G)*** this paper,a modeling primer based on the integration of these two cutting-edge technologies,i.e.,IRS and NOMA,for B5G wireless networks is *** in-depth comparative analysis of IRS-assisted Power Domain(PD)-NOMA networks is provided through 3-fold ***,a primer is presented on the system architecture of IRS-enabled multiple-configuration PD-NOMA systems,and parallels are drawn with conventional network configurations,i.e.,conventional NOMA,Orthogonal Multiple Access(OMA),and IRS-assisted OMA *** by this,a comparative analysis of these network configurations is showcased in terms of significant performance metrics,namely,individual users'achievable rate,sum rate,ergodic rate,EE,and outage ***,for multi-antenna IRS-enabled NOMA networks,we exploit the active Beamforming(BF)technique by employing a greedy algorithm using a state-of-the-art branch-reduceand-bound(BRB)*** optimality of the BRB algorithm is presented by comparing it with benchmark BF techniques,i.e.,minimum-mean-square-error,zero-forcing-BF,and ***,we present an outlook on future envisioned NOMA networks,aided by IRSs,i.e.,with a variety of potential applications for 6G wireless *** work presents a generic performance assessment toolkit for wireless networks,focusing on IRS-assisted NOMA *** comparative analysis provides a solid foundation for the dev

关键词： Intelligent reflecting surface Non-orthogonal multiple access 6G Beamforming Sum rate Energy efficiency

来源：评论

学校读者我要写书评

暂无评论

AN EMPIRICAL ANALYSIS OF FORGETTING IN PRE-TRAINED MODELS WITH INCREMENTAL LOW-RANK UPDATES 3

AN EMPIRICAL ANALYSIS OF FORGETTING IN PRE-TRAINED MODELS WI...

引用

3rd Conference on Lifelong Learning Agents, CoLLAs 2024

作者： Soutif-Cormerais, Albin Magistri, Simone van de Weijer, Joost Bagdanov, Andrew D. Computer Vision Center Spain Department of Computer Science Universitat Autònoma de Barcelona Spain Department of Information Engineering University of Florence Italy

Broad, open source availability of large pretrained foundation models on the internet through platforms such as HuggingFace has taken the world of practical deep learning by storm. A classical pipeline for neural network training now typically consists of finetuning these pretrained network on a small target dataset instead of training from scratch. In the case of large models this can be done even on modest hardware using a low rank training technique known as Low-Rank Adaptation (LoRA). While Low Rank training has already been studied in the continual learning setting, existing works often consider storing the learned adapter along with the existing model but rarely attempt to modify the weights of the pretrained model by merging the LoRA with the existing weights after finishing the training of each task. In this article we investigate this setting and study the impact of LoRA rank on the forgetting of the pretraining foundation task and on the plasticity and forgetting of subsequent ones. We observe that this rank has an important impact on forgetting of both the pretraining and downstream tasks. We also observe that vision transformers finetuned in that way exhibit a sort of "contextual" forgetting, a behaviour that we do not observe for residual networks and that we believe has not been observed yet in previous continual learning works. © 2024 CoLLAs. All Rights Reserved.

关键词： Open systems

来源：评论

学校读者我要写书评

暂无评论

Geyser Inspired Algorithm:A New Geological-inspired Meta-heuristic for Real-parameter and Constrained engineering Optimization

引用

Journal of Bionic engineering 2024年第1期21卷 374-408页

作者： Mojtaba Ghasemi Mohsen Zare Amir Zahedi Mohammad-Amin Akbari Seyedali Mirjalili Laith Abualigah Department of Electronics and Electrical Engineering Shiraz University of TechnologyShirazIran Department of Electrical Engineering Faculty of EngineeringJahrom UniversityJahromFrasIran Department of Electrical and Computer Engineering Tarbiat Modares UniversityTehranIran Department of Electrical and Computer Engineering University ofCyprusNicosiaCyprus Centre for Artificial Intelligence Research and Optimisation Torrens University AustraliaBrisbaneQLD 4006Australia University Research and Innovation Center Obuda University1034 BudapestHungary Department of Electrical and Computer Engineering Lebanese American UniversityByblos 13-5053Lebanon Hourani Center for Applied Scientific Research Al-Ahliyya Amman UniversityAmman 19328Jordan MEU Research Unit Middle East UniversityAmman 11831Jordan Applied Science Research Center Applied Science Private UniversityAmman 11931Jordan

Over the past years,many efforts have been accomplished to achieve fast and accurate meta-heuristic algorithms to optimize a variety of real-world *** study presents a new optimization method based on an unusual geological phenomenon in nature,named Geyser inspired Algorithm(GEA).The mathematical modeling of this geological phenomenon is carried out to have a better understanding of the optimization *** efficiency and accuracy of GEA are verified using statistical examination and convergence rate comparison on numerous CEC 2005,CEC 2014,CEC 2017,and real-parameter benchmark ***,GEA has been applied to several real-parameter engineering optimization problems to evaluate its *** addition,to demonstrate the applicability and robustness of GEA,a comprehensive investigation is performed for a fair comparison with other standard optimization *** results demonstrate that GEA is noticeably prosperous in reaching the optimal solutions with a high convergence rate in comparison with other well-known nature-inspired algorithms,including ABC,BBO,PSO,and *** that the source code of the GEA is publicly available at https://***/projects/gea.

关键词： Nature-inspired algorithms Real-world and engineering optimization Mathematical modeling Geyser algorithm(GEA)

来源：评论

学校读者我要写书评

暂无评论

MSRFormer: Hybrid Scale Self-Attention and Local Fast Convolution Transformer for Facial Expression Recognition

MSRFormer: Hybrid Scale Self-Attention and Local Fast Convol...

引用

2025 IEEE International Conference on Acoustics, Speech, and Signal Processing, ICASSP 2025

作者： Shen, Zi-Qiang Tang, Yu-Yi Yan, Jun-Feng Li, Yang Zhao, Guo-Ying Hunan University of Chinese Medicine China Hunan Digital Chinese Medicine Engineering Technology Research Center China Center for Machine Vision Research Department of Computer Science and Engineering University of Oulu Finland

ISBN: (纸本)9798350368741

With the emergence of the Transformer architecture, the accuracy of deep learning within the domain of facial emotion recognition has seen further enhancement. However, Transformer comes with increased training complexity and time due to the large parameter count. Additionally, the global receptive field in Transformer's attention leads to unnecessary computations for features with limited spatial extent in image sentiment analysis. In this paper we presents a MSRFormer model, which combines Hybrid-scale self-attention and local fast convolution to address existing issues. The Hybrid-scale self-attention enables precise focus on salient regions of the image, where key features are located. A Local Fast Convolution module was integrated into the MLP head of the Transformer model, enhancing training speed and reducing parameters while maintaining feature learning. we conducted experiments on FERPlus, RAF-DB, and KDEF datasets and confirming its strong performance and broad applicability in facial expression recognition. © 2025 IEEE.

关键词： facial expression recognition hybrid scale self attention local fast convolution transformer

来源：评论

学校读者我要写书评

暂无评论

Joint View Selection, Multigroup Multicast Beamforming, and DIBR for RIS-Aided Multi-View Videos

引用

IEEE Transactions on Mobile Computing 2025年第07期24卷 6376-6393页

作者： Lee, Chi-Han Yang, De-Nian Lee, Guang-Siang Wang, Chih-Hang Liao, Wanjiun National Taiwan University Department of Electrical Engineering Taiwan Academia Sinica Institute of Information Science The Research Center for Information Technology Innovation Taiwan Academia Sinica Research Center for Information Technology Innovation Taiwan National Chung Cheng University Department of Computer Science and Information Engineering Taiwan

The rapid development of multi-view videos (MVV) transmission is an irresistible trend. Concurrently, reconfigurable intelligent surface (RIS)-assisted wireless communication has drawn significant attention. We observe that the view selection based on the base station and the view synthesis based on depth-image-based rendering (DIBR) can effectively reduce power consumption. Therefore, this paper studies the view selection and synthesis for RIS-aided MVV in multigroup multicast beamforming. To deal with this complicated scenario, we investigate a problem, named the joint View selection, Multicast beamforming, and DIBR (JVMD), to minimize the total multicast beamforming power, the view transmission operation power, and view synthesis, subject to quality-of-service (QoS), RIS phase shifts, view selection, and DIBR constraints. Unfortunately, the mathematical model is a complicated mixed discrete-continuous optimization problem. To tackle this challenging problem, we designed an algorithm, named View selection, Beamforming, RIS phase, and DIBR (VBRD) algorithm. First, we deal with the discrete optimization problem of selecting the view. VBRD uses the dual-based approximation methodology to round back a primal's integer solution. Then, in the continuous optimization problem, we apply the alternating optimization (AO) method to determine beamforming, RIS phase, and DIBR. Finally, simulation results show the performance of exploiting view synthesis for RIS-assisted wireless communication. © 2002-2012 IEEE.

关键词： Beamforming

来源：评论

学校读者我要写书评

暂无评论

Visually Encrypted Watermarking for Ordered-Dithered Clustered-Dot Halftones

引用

IEEE Transactions on Circuits and Systems for Video Technology 2023年第8期33卷 4375-4387页

作者： Guo, Jing-Ming Seshathiri, Sankarasrinivasan Department of Electrical Engineering and the Advanced Intelligent Image and Vision Technology Research Center National Taiwan University of Science and Technology Taipei10607 Taiwan

Visual encryption and show through watermarking are widely used techniques to hide secret data in halftone images. The secret watermark can be quickly revealed when the halftone images are printed in transparency and overlaid on each other. The present studies emphasize developing a show-through watermarking technique for the clustered-dot halftone types. The proposed method exploits the properties of various configurations of dither array screens constructed using distinct Gaussian filters to embed watermarks. Two approaches are proposed, i.e., adjacent dither array pairs and the dither array translations. The optimal configuration of dither array parameters is developed to obtain maximum contrast and imperceptibility. Moreover, a new edge screen dithering is proposed to eliminate the edge artifacts, resulting in smooth screen transitions. As efficient ordered dithering is adopted for the watermark embedding, the joint watermarking and halftoning can be performed without additional computations. In comparison to the existing state-of-the-art show-through watermarking techniques, the proposed method can present superior image quality, computational simplicity, and decoded watermarks can be perceived with clear contrast when the images are overlaid with each other. © 2023 IEEE.

关键词： Cryptography

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：