检索结果-内蒙古大学图书馆

Mathematical analysis of AMRes: unlocking enhanced recognition across audio-visual domains

International Journal of Information Technology (Singapore) 2024年 1-20页

作者： Zoughi, Toktam Deypir, Mahmood Department of Electrical and Computer Engineering Shariaty College Technical and Vocational University (TVU) Tehran Iran Faculty of Computer Engineering Shahid Sattari Aeronautical University of Science and Technology Tehran Iran

This research presents AMRes (Adaptive Windowed Convolutional Neural Network and Multiple Residual Network), a novel method that demonstrates remarkable resilience against overfitting, even in scenarios with limited training data. With meticulous design comprising up to 19 convolutional layers, it enables deep learning models to effectively extract and analyze vital information from input data while preserving it through deep layers. Through rigorous mathematical and empirical validation, the proposed method exhibits adept handling of intra- and inter-speaker variability, efficient training with limited data, and potential applicability beyond speech recognition in image-related applications, showcasing its generality. This paper establishes a solid foundation for advancements in multi-faceted recognition systems. The proposed method excels at proficiently modeling speech signals and achieving highly accurate speech recognition. Thorough evaluations utilizing well-established databases such as Switchboard, Timit, and FarsDAT for speech, and MNIST for image data underscore the method's versatility and efficacy across diverse data contexts. Comparative analysis against state-of-the-art techniques in speech recognition demonstrates the propose method's competitive and often superior performance. Notably, this method significantly reduces phoneme recognition errors by approximately 8% on notable datasets. This research provides a solid foundation for advancing the field and highlights the proposed method's potential in improving recognition systems, encouraging further exploration. © Bharati Vidyapeeth's Institute of computer Applications and Management 2024.

关键词： Audio-Visual Domains Deep learning Formal Verification Limited Training Data Mathematical Analysis Mathematical Foundations Speech recognition

来源：评论

学校读者我要写书评

暂无评论

Single-Image Driven 3D Viewpoint Training Data Augmentation for Effective Label Recognition 27th

Single-Image Driven 3D Viewpoint Training Data Augmentation ...

引用

27th International Conference on Pattern Recognition, ICPR 2024

作者： Huang, Yueh-Cheng Chen, Hsin-Yi Hung, Cheng-Jui Chuang, Jen-Hui Hwang, Jenq-Neng Department of Computer Science National Yang Ming Chiao Tung University Hsinchu Taiwan Department of Electrical and Computer Engineering University of Washington Seattle United States

ISBN: (纸本)9783031781247

Confronting the critical challenge of insufficient training data in the field of complex image recognition, this paper introduces a novel 3D viewpoint transformation technique initially tailored for label recognition. This technique can be used not only for data augmentation by generating synthetic data from any perspective but also transform photos taken from any angle into a frontal view, thereby reducing the complexity of the recognition task. Given the extensive use of wine-related applications with over 20 million users and the continuous publication of wine label datasets, we decided to focus this study on wine labels. This method enhances deep learning model performance by generating visually realistic training samples from a single real-world label image, overcoming the challenges posed by the intricate combinations of text and logos. Unlike classical Generative Adversarial Network (GAN) methods, which fall short in synthesizing such intricate content combinations and require a large amount of training data to become effective, our proposed solution leverages time-tested computer vision and image processing strategies. By using just a single monocular wine label image, we can expand our training dataset, thereby broadening the range of training samples for deep learning applications. This innovative approach to data augmentation circumvents the constraints of limited training resources. We then utilize the augmented training images through the Vision Transformer (ViT) architecture, performing one-shot recognition of existing wine labels in the training classes or future newly collected wine labels unavailable in the training. Experimental results show a significant increase in recognition accuracy over conventional 2D data augmentation techniques, indicating the potential for broader application in various labeling scenarios. © The Author(s), under exclusive license to Springer Nature Switzerland AG 2025.

关键词： Wine

来源：评论

学校读者我要写书评

暂无评论

Minimal Input Structural Modifications for Strongly Structural Controllability 63

Minimal Input Structural Modifications for Strongly Structur...

引用

63rd IEEE Conference on Decision and Control, CDC 2024

作者： Joseph, Geethu Moothedath, Shana Lin, Jiabin Delft University of Technology Faculty of Electrical Engineering Mathematics and Computer Science CD Delft2628 Netherlands Iowa State University Department of Electrical and Computer Engineering IA50011 United States

ISBN: (纸本)9798350316339

This paper studies the problem of modifying the input matrix of a structured system to make the system strongly structurally controllable. We focus on the generalized structured systems that rely on zero/nonzero/arbitrary structure, i.e., some entries of system matrices are zeros, some are nonzero, and the remaining entries can be zero or nonzero (arbitrary). We derive the feasibility conditions of the problem, and if it is feasible, we reformulate it into another equivalent problem. This new formulation leads to a greedy heuristic algorithm. However, we also show that the greedy algorithm can give arbitrarily poor solutions for some special systems. Our alternative approach is a randomized Markov chain Monte Carlo-based algorithm. Unlike the greedy algorithm, this algorithm is guaranteed to converge to an optimal solution with high probability. Finally, we numerically evaluate the algorithms on random graphs to show that the algorithms perform well. © 2024 IEEE.

关键词： Markov chains

来源：评论

学校读者我要写书评

暂无评论

Leveraging Noise Recycling in Soft Detection Decoding Using ORBGRAND

Leveraging Noise Recycling in Soft Detection Decoding Using ...

引用

2023 IEEE International Symposium on Information Theory, ISIT 2023

作者： Kizilates, Zeynep Ece Riaz, Arslan Coraluppi, Giacomo F. Medard, Muriel Duffy, Ken R. Tugce Yazicigil, Rabia Boston University Department of Electrical and Computer Engineering BostonMA United States Mit Department of Electrical Engineering and Computer Science CambridgeMA United States Northeastern University Department of Electrical and Computer Engineering Department of Mathematics BostonMA United States

ISBN: (纸本)9781665475549

For communications subject to correlated channel effects, noise recycling has recently been shown to enhance channel capacity with receiver-side-only changes. Using a taped-out chip, in a hard-detection scenario with guessing random additive noise decoding (GRAND), noise recycling has been established to both increase decoding accuracy and decrease decoding energy in single communication channels that employ interleavers. This paper presents results for the related soft-detection scenario by investigating noise recycling with an in-silicon realization of Ordered Reliability Bits Guessing Random Additive Noise Decoding (ORBGRAND). Measurements demonstrate that noise recycling leads to a significant reduction in the block error rate (BLER) and substantial improvements in latency and energy consumption by reducing the number of queries required for decoding. We also discuss dynamic lead channel selection for the soft detection scenario and show the importance of lead channel on overall decoding performance. © 2023 IEEE.

关键词： Additive noise

来源：评论

学校读者我要写书评

暂无评论

Lightweight Look-ahead Multi-context Vision Transformer

Lightweight Look-ahead Multi-context Vision Transformer

引用

2025 International Conference on Electronics, Information, and Communication, ICEIC 2025

作者： Byeon, Keunho Lim, Jeewoo Lee, Jaeung Han, Chiwon Bay, SungYun Park, Giho Kwak, Jin Tae School of Electrical Engineering Korea University Seoul Korea Republic of Sejong University Department of Computer Science and Engineering Seoul Korea Republic of

ISBN: (纸本)9798331510756

Vision Transformer (ViT) has been successfully applied to various vision tasks, outperforming convolutional neural networks due to their ability to capture global dependencies through the self-attention mechanism. However, it has been shown that ViT struggles with high model complexity and low computational efficiency, and thus it becomes less practical for deployment on edge devices. To address these issues, various studies have sought to enhance the efficiency of ViT, in particular focusing on hybrid models and self-attention mechanism. Herein, we propose a lightweight look-ahead multi-context ViT, which integrates convolutional neural networks and ViT in a hybrid, sequential design to achieve compact size and superior performance on vision tasks. The proposed model introduces a novel self-attention mechanism, which is built based upon a modified Nyström method with advanced landmark selection to provide multi-scale and multi-context information. Experimental results on image recognition dataset demonstrate the superior performance and the effectiveness of the proposed model. © 2025 IEEE.

关键词： Convolutional neural networks

来源：评论

学校读者我要写书评

暂无评论

FAST NONLINEAR VECTOR QUANTILE REGRESSION 11

FAST NONLINEAR VECTOR QUANTILE REGRESSION

引用

11th International Conference on Learning Representations, ICLR 2023

作者： Rosenberg, Aviv A. Vedula, Sanketh Romano, Yaniv Bronstein, Alex M. Department of Computer Science Technion Israel Department of Electrical and Computer Engineering Technion Israel Sibylla United Kingdom

Quantile regression (QR) is a powerful tool for estimating one or more conditional quantiles of a target variable Y given explanatory features X.A limitation of QR is that it is only defined for scalar target variables, due to the formulation of its objective function, and since the notion of quantiles has no standard definition for multivariate ***, vector quantile regression (VQR) was proposed as an extension of QR for vector-valued target variables, thanks to a meaningful generalization of the notion of quantiles to multivariate distributions via optimal *** its elegance, VQR is arguably not applicable in practice due to several limitations: (i) it assumes a linear model for the quantiles of the target Y given the features X;(ii) its exact formulation is intractable even for modestly-sized problems in terms of target dimensions, number of regressed quantile levels, or number of features, and its relaxed dual formulation may violate the monotonicity of the estimated quantiles;(iii) no fast or scalable solvers for VQR currently *** this work we fully address these limitations, namely: (i) We extend VQR to the non-linear case, showing substantial improvement over linear VQR;(ii) We propose vector monotone rearrangement, a method which ensures the quantile functions estimated by VQR are monotone functions;(iii) We provide fast, GPU-accelerated solvers for linear and nonlinear VQR which maintain a fixed memory footprint, and demonstrate that they scale to millions of samples and thousands of quantile levels;(iv) We release an optimized python package of our solvers as to widespread the use of VQR in real-world applications. © 2023 11th International Conference on Learning Representations, ICLR 2023. All rights reserved.

关键词： Vectors

来源：评论

学校读者我要写书评

暂无评论

Accelerating Mobile Edge Generation (MEG) by Constrained Learning

引用

IEEE Transactions on Cognitive Communications and Networking 2025年第3期11卷 1854-1869页

作者： Xu, Xiaoxia Liu, Yuanwei Mu, Xidong Xing, Hong Nallanathan, Arumugam Queen Mary University of London School of Electronic Engineering and Computer Science LondonE1 4NS United Kingdom The University of Hong Kong Department of Electrical and Electronic Engineering Hong Kong School of Electronics Electrical Engineering and Computer Science BelfastBT3 9DT United Kingdom The Hong Kong University of Science and Technology IoT Thrust Guangzhou Guangzhou511453 China Hong Kong University of Science and Technology Department of ECE Hong Kong Kyung Hee University Department of Electronic Engineering Gyeonggi-do Yongin-si17104 Korea Republic of

A novel mobile edge generation (MEG) framework is proposed for low-latency image generation on mobile devices. Exploiting a latent diffusion model (LDM) distributed across the edge server (ES) and the user equipment (UE), only low-dimension features need to be transmitted for creating artificial intelligence generative content (AIGC). Two novel modules, namely dynamic diffusion and feature merging, are conceived to compress the diffusion model and transmitted features, respectively. By jointly optimizing compression rates of denoising steps and feature merging, the image quality maximization problem is formulated subject to latency and energy consumption constraints. To address this problem in dynamic channel conditions, a low-complexity compression protocol is developed. First, a backbone LDM architecture is learned by offline distillation to support various compression options. Then, compression rates are predicted in online environment specific to channel and task features. To solve the resultant constrained Markov Decision Process (MDP), a constrained variational policy optimization (CVPO) based MEG algorithm, MEG-CVPO, is further developed to learn constraint-guaranteed optimization. Numerical results demonstrate that: 1) The proposed framework improves image distortions while reducing over 40% latency compared to conventional generation schemes. 2) MEG-CVPO stringently guarantee constraints and realizes a flexible trade-off between generation qualities and overheads. © 2015 IEEE.

关键词： Markov processes

来源：评论

学校读者我要写书评

暂无评论

Frequency Bandwidth Reduction of High-Frequency Intra-Body Propagation Characteristics Used for Attacker's Touch Detection in Fingerprint Authentication 12

Frequency Bandwidth Reduction of High-Frequency Intra-Body P...

引用

12th IEEE Global Conference on Consumer Electronics, GCCE 2023

作者： Tomura, Shunichi Yamada, Koki Yoshida, Takahiro Tokyo University of Science Graduate School of Engineering Department of Electrical Engineering Tokyo Japan Tokyo University of Agriculture and Technology Department of Electrical Engineering and Computer Science Tokyo Japan Tokyo University of Science Faculty of Engineering Department of Electrical Engineering Tokyo Japan

ISBN: (纸本)9798350340181

In previous studies, we have proposed the attacker's touch detection method in fingerprint authentication using high-frequency intra-body propagation characteristics to detect attacks in that an attacker holds the user's finger/wrist and presses the user's finger to a fingerprint scanner. However, the previous method used a wide range of frequency bands from 300 kHz to 10 MHz, and this caused problems in the circuit scale, cost, and size of the device in development for practical use. Therefore, we searched effective frequency bands and applied K-nearest neighbor with data augmentation in this study. As results, we achieved a 99.1 % touch detection rate with a single frequency and a 99.9 % touch detection rate with a combination of two frequencies. We achieved a significant reduction in the frequency bandwidth of the propagation characteristics. © 2023 IEEE.

关键词： Bandwidth

来源：评论

学校读者我要写书评

暂无评论

Mining Erasable Itemsets from a Product Database with Quantitative Components 24

Mining Erasable Itemsets from a Product Database with Quanti...

引用

11th Multidisciplinary International Social Networks Conference, MISNC 2024

作者： Hong, Tzung-Pei Chen, Da Huang, Wei-Ming Tsai, Yu-Chuan Department of Computer Science and Information Engineering National University of Kaohsiung Taiwan Department of Computer Science and Engineering National Sun Yat-Sen University Taiwan Department of Computer Science and Engineering Department of Electrical and Control China Steel Inc. China Library and Information Center National University of Kaohsiung Taiwan

ISBN: (纸本)9798400717550

Erasable itemset mining is one of the most well-known methods in data mining for optimizing limited materials. After mining erasable itemsets, the manager can rearrange the production plan effectively. However, in real applications, traditional mining methods neglect some critical factors, such as costs or quantities of components forming a product. This thesis thus considers mining erasable itemsets from a product database with quantitative components. An erasable profit-based itemset mining (EPIM) algorithm is proposed to mine erasable itemsets with low profits by analyzing the relationships between products and components. In experiments, a synthetic dataset was used to compare the performance in execution speed and memory usage under various threshold settings. It can also be easily generalized to find erasable itemsets satisfying the other criteria with a quantitative product database. © 2024 ACM.

关键词： Mine safety

来源：评论

学校读者我要写书评

暂无评论

Agent Selection Framework for Federated Learning in Resource-Constrained Wireless Networks

IEEE Transactions on Machine Learning in Communications and ...

引用

IEEE Transactions on Machine Learning in Communications and Networking 2024年 2卷 1265-1282页

作者： Raftopoulou, Maria Da Silva, Jose Mairton B. Litjens, Remco Vincent Poor, H. Van Mieghem, Piet Uppsala University Department of Information Technology Uppsala751 05 Sweden Delft University of Technology Faculty of Electrical Engineering Mathematics and Computer Science Delft2628 CD Netherlands Department of Networks The Hague2595 DA Netherlands Princeton University Department of Electrical and Computer Engineering PrincetonNJ08544 United States

Federated learning is an effective method to train a machine learning model without requiring to aggregate the potentially sensitive data of agents in a central server. However, the limited communication bandwidth, the hardware of the agents and a potential application-specific latency requirement impact how many and which agents can participate in the learning process at each communication round. In this paper, we propose a selection metric characterizing each agent's importance with respect to both the learning process and the resource efficiency of its wireless communication channel. Leveraging this importance metric, we formulate a general agent selection optimization problem, which can be adapted to different environments with latency or resource-oriented constraints. Considering an example wireless environment with latency constraints, the agent selection problem reduces to the 0/1 Knapsack problem, which we solve with a fully polynomial approximation. We then evaluate the agent selection policy in different scenarios, using extensive simulations for an example task of object classification of European traffic signs. The results indicate that agent selection policies which consider both learning and channel aspects provide benefits in terms of the attainable global model accuracy and/or the time needed to achieve a targeted accuracy level. However, in scenarios where agents have a limited number of data samples or where the latency requirement is very stringent, a pure learning-based agent selection policy is shown to be more beneficial during the early or late stages of the learning process. © 2023 CCBY.

关键词： Federated 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：