检索结果-内蒙古大学图书馆

48th IEEE International Conference on Acoustics, Speech and Signal Processing, ICASSP 2023

作者： Gonzalez, Philippe Sonne Alstrom, Tommy May, Tobias Technical University of Denmark Department of Health Technology Denmark Technical University of Denmark Department of Applied Mathematics and Computer Science Denmark

ISBN: (纸本)9781728163277

The performance of neural network-based speech enhancement systems is primarily influenced by the model architecture, whereas training times and computational resource utilization are primarily affected by training parameters such as the batch size. Since noisy and reverberant speech mixtures can have different duration, a batching strategy is required to handle variable size inputs during training, in particular for state-of-the-art end-to-end systems. Such strategies usually strive for a compromise between zero-padding and data randomization, and can be combined with a dynamic batch size for a more consistent amount of data in each batch. However, the effect of these strategies on resource utilization and more importantly network performance is not well documented. This paper systematically investigates the effect of different batching strategies and batch sizes on the training statistics and speech enhancement performance of a Conv-TasNet, evaluated in both matched and mismatched conditions. We find that using a small batch size during training improves performance in both conditions for all batching strategies. Moreover, using sorted or bucket batching with a dynamic batch size allows for reduced training time and GPU memory usage while achieving similar performance compared to random batching with a fixed batch size. © 2023 IEEE.

关键词： Speech enhancement

来源：评论

学校读者我要写书评

暂无评论

FreqRISE: Explaining time series using frequency masking 6

FreqRISE: Explaining time series using frequency masking

引用

6th Northern Lights Deep Learning Conference, NLDL 2025

作者： Brüsch, Thea Wickstrøm, Kristoffer Knutsen Schmidt, Mikkel N. Alstrøm, Tommy Sonne Jenssen, Robert Department of Applied Mathematics and Computer Science Technical University of Denmark Denmark Department of Physics and Technology UiT The Arctic University of Norway Norway Norwegian Computing Center Oslo Norway

Time series data is fundamentally important for many critical domains such as healthcare, finance, and climate, where explainable models are necessary for safe automated decision-making. To develop explainable artificial intelligence in these domains therefore implies explaining salient information in the time series. Current methods for obtaining saliency maps assumes localized information in the raw input space. In this paper, we argue that the salient information of a number of time series is more likely to be localized in the frequency domain. We propose FreqRISE, which uses masking-based methods to produce explanations in the frequency and time-frequency domain, and outperforms strong baselines across a number of tasks. The source code is available here: https://***/theabrusch/FreqRISE. © NLDL *** rights reserved.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Multi-Level Feature Distillation of Joint Teachers Trained on Distinct Image Datasets

Multi-Level Feature Distillation of Joint Teachers Trained o...

引用

2025 IEEE/CVF Winter Conference on Applications of computer Vision, WACV 2025

作者： Iordache, Adrian Alexe, Bogdan Ionescu, Radu Tudor University of Bucharest Department of Computer Science Bucharest Romania 'Gheorghe Mihoc - Caius Iacob' Institute of Mathematical Statistics and Applied Mathematics The Romanian Academy Bucharest Romania

ISBN: (纸本)9798331510831

We propose a novel teacher-student framework to distill knowledge from multiple teachers trained on distinct datasets. Each teacher is first trained from scratch on its own dataset. Then, the teachers are combined into a joint architecture, which fuses the features of all teachers at multiple representation levels. The joint teacher architecture is fine-tuned on samples from all datasets, thus gathering useful generic information from all data samples. Finally, we employ a multi-level feature distillation procedure to transfer the knowledge to a student model for each of the considered datasets. We conduct image classification experiments on seven benchmarks, and action recognition experiments on three benchmarks. To illustrate the power of our feature distillation procedure, the student architectures are chosen to be identical to those of the individual teachers. To demonstrate the flexibility of our approach, we combine teachers with distinct architectures. We show that our novel Multi-Level Feature Distillation (MLFD) can significantly surpass equivalent architectures that are either trained on individual datasets, or jointly trained on all datasets at once. Furthermore, we confirm that each step of the proposed training procedure is well motivated by a comprehensive ablation study. We publicly release our code at https://***/AdrianIordache/MLFD. © 2025 IEEE.

关键词： Teaching

来源：评论

学校读者我要写书评

暂无评论

Approximate Selection with Unreliable Comparisons in Optimal Expected Time 40

Approximate Selection with Unreliable Comparisons in Optimal...

引用

40th International Symposium on Theoretical Aspects of computer science, STACS 2023

作者： Huang, Shengyu Liu, Chih-Hung Rutschmann, Daniel Department of Computer Science EPFL Lausanne Switzerland Department of Electrical Engineering National Taiwan University Taipei Taiwan Department of Applied Mathematics and Computer Science Technical University of Denmark Copenhagen Denmark

ISBN: (纸本)9783959772662

Given n elements, an integer k ≤ n/2 and a parameter Ε ≥ 1/n , we study the problem of selecting an element with rank in (k − nΕ, k + nΕ] using unreliable comparisons where the outcome of each comparison is incorrect independently with a constant error probability, and multiple comparisons between the same pair of elements are independent. In this fault model, the fundamental problems of finding the minimum, selecting the k-th smallest element and sorting have been shown to require Θ(n log 1/Q), Θ(n log k/Q) and Θ(n log n/Q) comparisons, respectively, to achieve success probability 1 − Q [9]. Considering the increasing complexity of modern computing, it is of great interest to develop approximation algorithms that enable a trade-off between the solution quality and the number of comparisons. In particular, approximation algorithms would even be able to attain a sublinear number of comparisons. Very recently, Leucci and Liu [23] proved that the approximate minimum selection problem, which covers the case that k ≤ nΕ, requires expected Θ(Ε−1 log 1/Q) comparisons, but the general case, i.e., for nΕ −2 log 1/Q) comparisons to achieve success probability at least 1 − Q. For k = nΕ, the number of comparisons is O(Ε−1 log 1/Q), matching Leucci and Liu’s result [23], whereas for k = n/2 (i.e., approximating the median), the number of comparisons is O(Ε−2 log 1/Q). We also prove that even in the absence of comparison faults, any randomized algorithm with success probability at least 1 − Q performs expected Ω(min{n, k/n Ε−2 log 1/Q }) comparisons. As long as n is large enough, i.e., when n = Ω(k/n Ε−2 log 1/Q), our lower bound demonstrates the optimality of our algorithm, which covers the possible range of attaining a sublinear number of comparisons. Surprisingly, for constant Q, our algorithm performs expected O(k/n Ε−2) comparisons, matching the best possible approximation algorithm in the absence of computation faults. In contrast, for the exact selection problem, the expe

关键词： Approximation algorithms

来源：评论

学校读者我要写书评

暂无评论

Thermal-Aware Communication

引用

IEEE Transactions on Information Theory 2025年第6期71卷 4145-4155页

作者： Chee, Yeow Meng Etzion, Tuvi Immink, Kees A. Schouhamer Nguyen, Tuan Thanh Vu, Van Khu Weber, Jos H. Yaakobi, Eitan Singapore University of Technology and Design Singapore Technion — Israel Institute of Technology Department of Computer Science Haifa3200003 Israel National University of Singapore Department of Industrial Systems Engineering and Management Singapore Turing Machines Inc Rotterdam3016 DK Netherlands Singapore University of Technology and Design Science Mathematics and Technology Cluster 487372 Singapore Delft University of Technology Department of Applied Mathematics Delft2628 CD Netherlands

Temperature control is of utmost importance in transmission systems. In this paper, a binary channel model is considered in which the transmission of a one causes a temperature increase while communicating a zero causes a temperature drop. By putting constraints on the input sequences, it is guaranteed that the channel temperature will not exceed a certain pre-determined maximum. In the asymptotic regime, the capacity of such a channel is studied. For the non-asymptotic regime, fixed-length codes are presented, with the property that codewords can be freely cascaded without violating the temperature constraint. Optimization of the code size is investigated and codewords are enumerated using generating functions. © 1963-2012 IEEE.

关键词： Channel capacity

来源：评论

学校读者我要写书评

暂无评论

mLBOA:A Modified Butterfly Optimization Algorithm with Lagrange Interpolation for Global Optimization

引用

Journal of Bionic Engineering 2022年第4期19卷 1161-1176页

作者： Sushmita Sharma Sanjoy Chakraborty Apu Kumar Saha Sukanta Nama Saroj Kumar Sahoo Department of Mathematics National Institute of Technology AgartalaAgartalaTripura 799046India Department of Computer Science and Engineering Iswar Chandra Vidyasagar CollegeBeloniaTripura 799155India Department of Computer Science and Engineering National Institute of Technology AgartalaAgartalaTripura 799046India Department of Applied Mathematics Maharaja Bir Bikram UniversityAgartalaTripura 799004India

Though the Butterfly Bptimization Algorithm(BOA)has already proved its effectiveness as a robust optimization algorithm,it has certain ***,a new variant of BOA,namely mLBOA,is proposed here to improve its *** proposed algorithm employs a self-adaptive parameter setting,Lagrange interpolation formula,and a new local search strategy embedded with Levy flight search to enhance its searching ability to make a better trade-off between exploration and ***,the fragrance generation scheme of BOA is modified,which leads for exploring the domain effectively for better *** evaluate the performance,it has been applied to solve the IEEE CEC 2017 benchmark *** results have been compared to that of six state-of-the-art algorithms and five BOA ***,various statistical tests,such as the Friedman rank test,Wilcoxon rank test,convergence analysis,and complexity analysis,have been conducted to justify the rank,significance,and complexity of the proposed ***,the mLBOA has been applied to solve three real-world engineering design *** all the analyses,it has been found that the proposed mLBOA is a competitive algorithm compared to other popular state-of-the-art algorithms and BOA variants.

关键词： Butterfly optimization algorithm Lagrange interpolation Levy flight search IEEE CEC 2017 functions Engineering design problems

来源：评论

学校读者我要写书评

暂无评论

IoDseC++: authenticated key exchange protocol for cloud-enable internet of drone communication

引用

Journal of Ambient Intelligence and Humanized Computing 2023年第7期14卷 9529-9542页

作者： Amin, Ruhul Jayaswal, Sakshita Sureshkumar, Venkatasamy Rathore, Balram Jha, Ananya Abdussami, Mohammad Department of Computer Science and Engineering IIIT Naya raipur Sector 24 Chhattisgarh Naya Raipur493661 India Department of Applied Mathematics and Computational Sciences PSG College of Technology Tamilnadu Coimbatore641004 India

In recent times, sensor node or smart camera can be embedded in the drones for the collection of important data with the help of Internet Technology. Several security threats/vulnerabilities may hamper the system while collecting data. Hence, strong protection against security attacks is essential in drone based data collection system. In this work, we have considered a drone based cloud enable architecture for collecting and sharing information. The aim of this paper is to design a lightweight security protocol which will provide secure communication and mutual authentication among the entities. The proposed protocol establishes temporary session key for each session in order to make the complete system security attacks free. We have simulated the proposed protocol for measuring security strength using Scyther simulation whose results show all the claimed parameters are private during protocol run. It also ensures the correctness of mutual authentication. The informal analysis confirms strong protection on all security threats related to proposed architecture. The performance analysis has also made which shows that the protocol is lightweight and achieves better efficiency in comparison with related published works. © 2023, The Author(s), under exclusive licence to Springer-Verlag GmbH Germany, part of Springer Nature.

关键词： Drones

来源：评论

学校读者我要写书评

暂无评论

Cluster-preserving sampling algorithm for large-scale graphs

引用

science China(Information sciences) 2023年第1期66卷 60-76页

作者： Jianpeng ZHANG Hongchang CHEN Dingjiu YU Yulong PEI Yingjun DENG National Digital Switching System E&T Research Center Information Engineering University Network Systems Department of the Strategic Support Force School of Computer Science and Technology Eindhoven University of Technology Center for Applied Mathematics Tianjin University

Graph sampling is a very effective method to deal with scalability issues when analyzing largescale graphs. Lots of sampling algorithms have been proposed, and sampling qualities have been quantified using explicit properties(e.g., degree distribution) of the sample. However, the existing sampling techniques are inadequate for the current sampling task: sampling the clustering structure, which is a crucial property of the current networks. In this paper, using different expansion strategies, two novel top-leader sampling methods(i.e., TLS-e and TLS-i) are proposed to obtain representative samples, and they are capable of effectively preserving the clustering structure. The rationale behind them is to select top-leader nodes of most clusters into the sample and then heuristically incorporate peripheral nodes into the sample using specific expansion strategies. Extensive experiments are conducted to investigate how well sampling techniques preserve the clustering structure of graphs. Our empirical results show that the proposed sampling algorithms can preserve the population's clustering structure well and provide feasible solutions to sample the clustering structure from large-scale graphs.

关键词： graph sampling clustering structure top-leader nodes expansion strategies large-scale graphs

来源：评论

学校读者我要写书评

暂无评论

Optimation Cavacity Vehicle Routing Problem with K-Nearest Neighbor in Classification of Goods Ditribution Route 7

Optimation Cavacity Vehicle Routing Problem with K-Nearest N...

引用

7th IEEE International Conference of computer science and Information Technology, ICOSNIKOM 2023

作者： Novelan, Muhammad Syahputra Efendi, Syahril Sihombing, Poltak Mawengkang, Herman Universitas Sumatera Utara Doctoral Program of Computer Science Faculty of Computer Science and Information Technology Medan Indonesia Universitas Sumatera Utara Faculty of Computer Science and Information Technology Department of Computer Science Medan Indonesia Universitas Sumatera Utara Faculty of Mathematics and Natural Sciences Department of Mathematics Medan Indonesia

ISBN: (纸本)9798350360752

Industrial and distribution service problems that belong to combinatorial optimization include vehicle routing with Vehicle Routing Problem. This research builds a framework and implements it in a multi-class optimization model to reduce overfitting and misclassification results caused by unbalanced multiclassification in the influence of the number of 'nodes' on vehicle routes with machine learning. The problem of imbalance in vehicle route classification has been a challenge in the classification process and attracted the attention of a number of researchers. The purpose of the model in general is to gain an understanding of the mechanism in the problem so as to classify the imbalanced vehicle route data based on JNE delivery routes. The solution method that will be used by applying k-nearest neighbor to determine the amount of carrying capacity of the goods will then determine the delivery location point with the vehicle routing problem. So that this model can be a model in determining vehicle routes based on the capacity limit of the number of shipments of goods. © 2023 IEEE.

关键词： Machine learning

来源：评论

学校读者我要写书评

暂无评论

Bibliometric Analysis: Research Trends in Leaf Image Segmentation and Classification

Bibliometric Analysis: Research Trends in Leaf Image Segment...

引用

2024 International Conference on Artificial Intelligence and Mechatronics System, AIMS 2024

作者： Agustini, Eka Puji Gernowo, Rahmad Kurnianingsih Wibowo, Adi Warsito, Budi Doctoral Program of Information System School of Postgraduated Studies Diponegoro University Semarang Indonesia Doctoral Program of Information System Diponegoro University Semarang Indonesia Master of Applied Science Program Polytechnic Semarang Indonesia Diponegoro University Faculty of Science and Mathematics Department of Informatics Semarang Indonesia Faculty of Science and Mathematics Diponegoro Universitty Department of Statistics Semarang Indonesia

ISBN: (纸本)9798350350524

Exploring various phenomena and issues related to leaf images is paramount, particularly in segmentation and classification of such images. This study employs bibliometric analysis to delve into two overarching themes: The trends in publication and the evolution of publications, along with a keyword-based analysis. The research methodology unfolds in two stages: (1) data collection and (2) analysis. The dataset comprises 1,248 articles sourced from the Scopus database, covering the period from 1988 to 2023. The research findings unveil a noteworthy surge in publication trends, peaking at 231 documents in 2023. An in-depth examination of journal names demonstrates that studies on the segmentation and classification of leaf images are predominantly featured in computer science journals, exemplified by the publication of 589 documents. Furthermore, an analysis of frequently used keywords highlights 'Extraction' as the predominant term, employed a total of 364 times. This underscores that the research focus on leaf image segmentation and classification presents ample opportunities for researchers to delve more profoundly into the subject. © 2024 IEEE.

关键词： Image segmentation

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：