检索结果-内蒙古大学图书馆

arXiv 2020年

作者： Chen, Liang Liang, Yanchun Shi, Xiaohu Zhou, You Wu, Chunguo Key Laboratory of Symbolic Computation Knowledge Engineering of Ministry of Education College of Computer Science and Technology Jilin University Changchun130012 China Zhuhai Laboratory of Key Laboratory of Symbol Computation Knowledge Engineering of Ministry of Education School of Computer Zhuhai College Jilin University Zhuhai519041 China

Time Delay Neural Network (TDNN) is a well-performing structure for deep neural network-based speaker recognition systems. In this paper we introduce a novel structure, named Crossed-Time Delay Neural Network (CTDNN) to enhance the performance of current TDNN for speaker recognition. Inspired by the multi-filters setting of convolution layers from convolution neural networks, we set multiple time delay units with different context size at the bottom layer and construct a multilayer parallel network. The proposed CTDNN gives significant improvements over original TDNN on both speaker verification and identification tasks. It outperforms in VoxCeleb1.dataset in verification experiment with a 2.6% absolute Equal Error Rate improvement. In few shots condition, CTDNN reaches 90.4% identification accuracy, which doubles the identification accuracy of original TDNN. We also compare the proposed CTDNN with another new variant of TDNN, Factorized-TDNN, which shows that our model has a 36% absolute identification accuracy improvement under few shots condition. Moreover, the proposed CTDNN can handle training with a larger batch more efficiently and hence, utilize calculation resources economically. The code of the new model is released at https://***/chenllliang/CTDNN. Copyright © 2020, The Authors. All rights reserved.

关键词： Deep neural networks

来源：评论

学校读者我要写书评

暂无评论

Task replica assignment in mobile self-organized crowdsensing

引用

International Journal of Performability engineering 2020年第1期16卷 152-162页

作者： Wei, Xiaohui Sun, Bingyi Cui, Jiaxu College of Computer Science and Technology Jilin University Changchun130012 China Key Laboratory of Symbolic Computation and Knowledge Engineering Ministry of Education Jilin University Changchun130012 China

In modern society, people carry mobile devices everywhere. However, these mobile devices often stay in an idle status, which leads to wasted resources. Thus, many researchers have sought methods to place tasks on idle mobile devices to avoid resource waste. In this paper, we propose three cooperative algorithms. The first algorithm is employed to find credible participants, which is the foundation of the latter two algorithms. The second and third algorithms are task replica assignment algorithms based on credible participants in mobile self-organized crowdsensing, and they are used in offline and online situations, respectively. The latter two algorithms adopt the greedy strategy and include constraints in assignment strategy for replicas. The experiments show that the proposed algorithms dramatically increase the probability of finding accurate results, increasing from slightly more than 0.6 for the original algorithms to nearly 0.98, even though the proposed algorithms have slightly longer average execution times. © 2020 Totem Publisher, Inc.

关键词： Credit value Crowdsensing Mobile self-organized network Opportunistic sensing Task replica assignment

来源：评论

学校读者我要写书评

暂无评论

Understanding the Runtime Overheads of Deep Learning Inference on Edge Devices

Understanding the Runtime Overheads of Deep Learning Inferen...

引用

IEEE International Conference on Big Data and Cloud Computing (BdCloud)

作者： Xiu Ma Guangli Li Lei Liu Huaxiao Liu Xiaobing Feng College of Computer Science and Technology Jilin University China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University China SKL of Computer Architecture Institute of Computing Technology Chinese Academy of Sciences China University of Chinese Academy of Sciences China

With the growing ubiquity of the Internet of Things, in-the-edge inference of deep neural network models has been a major driver for promoting the widespread use of intelligent applications. As model inference characteristics are crucial for optimizing and deploying deep neural networks on hardware platforms, many studies focus on analyzing the performance of neural networks such as latency, accuracy, throughput, and energy consumption. However, few existing works have ever discussed the runtime overheads hidden in neural network inference, despite the overheads are non-negligible for edge applications. The lack of in-depth analysis of the overheads hinders the understanding of how hardware designs and model structures impact on-device inference performance. In this paper, we characterize the runtime overheads of deep learning inference on representative edge devices by leveraging state-of-the-art neural network models, performing a systematical analysis from the perspective of end-to-end performance, hardware platforms, memory bandwidth, and neural network model structures. Based on experimental results, the crucial insights are offered to facilitate the design and configure of resource-efficient networks and pick appropriate models on the specific platform, which provides a comprehensive view of runtime overheads of in-the- edge neural network inference for architects and developers.

关键词： Deep learning Performance evaluation Analytical models Runtime Roads Neural networks Throughput

来源：评论

学校读者我要写书评

暂无评论

A new SAT Encoding Scheme for Exactly-one Constraints 5

A new SAT Encoding Scheme for Exactly-one Constraints

引用

5th Annual International Conference on Network and Information Systems for Computers, ICNISC 201.

作者： Cai, Jiatong Su, Yating Yang, Xixi Min, Jionghao Lai, Yong College of Software Jilin University Changchun China Key Laboratory of Symbolic Computation and Knowledge Engineering Ministry of Education Jilin University Changchun China

Exactly-one constraints have comprehensive applications for the fields of artificial intelligence and operations research. For many encoded SAT problems generated by the existing encoding schemes of exactly-one constraints, the state-of-the-art knowledge compilers cannot complete compilation. In this paper, we propose a new encoding scheme of exactly-one constraints. We introduce two-dimensional auxiliary variables (represented as a matrix) to denote the constraint that exactly one of some variables can be assigned as true. The clauses generated by our scheme is significantly less than those generated by three other existing encoding schemes. The experimental results on the exact cover problems show that the encoded CNF formulas generated by our scheme requires less compilation time, compared with the other three coding schemes. © Published under licence by IOP Publishing Ltd.

关键词： Operations research

来源：评论

学校读者我要写书评

暂无评论

Neural architecture search based on the cartesian genetic programming

arXiv

引用

arXiv 2021年

作者： Wu, Xuan Zhang, Xiuyi Jia, Linhan Chen, Liang Liang, Yanchun Zhou, You Wu, Chunguo Key Laboratory of Symbolic Computation Knowledge Engineering of Ministry of Education College of Computer Science and Technology Jilin University Changchun130012 China College of Software Jilin University Changchun130012 China School of Computer Science Zhuhai College of Science and Technology Zhuhai519041 China

Neural architecture search (NAS) is a hot topic in the field of automated machine learning and outperforms humans in designing neural architectures on quite a few machine learning tasks. Motivated by the natural representation form of neural networks by Cartesian Genetic Programming (CGP), we propose an evolutionary approach of NAS based on CGP, called CGPNAS, to solve sentence classification task. To evolve the architectures under the framework of CGP, the operations such as convolution are identified as the types of function nodes of CGP, and the evolutionary operations are designed based on Evolutionary Strategy. The experimental results show that the searched architectures are comparable with the performance of human-designed architectures. We verify the ability of domain transfer of our evolved architectures and the transfer experimental results show that the accuracy deterioration is lower than 2-5%. Finally, the ablation study identifies the Attention function as the single key function node and the linear transformations along could keep the accuracy similar with the full evolved architectures, which is worthy of investigation in the future. Copyright © 2021. The Authors. All rights reserved.

关键词： Genetic algorithms

来源：评论

学校读者我要写书评

暂无评论

An unmanned aerial vehicle pose estimation system guided by desired shot 2

An unmanned aerial vehicle pose estimation system guided by ...

引用

2nd IEEE International Conference on Automation, Electronics and Electrical engineering, AUTEEE 201.

作者： Dou, Xinglei Ding, Qijian Du, Qiao Yang, Min Huo, Yuzhou Wang, Xin College of Software Engineering Jilin University Changchun China Key Laboratory of Symbolic Computation and Knowledge Engineer Ministry of Education Jilin University Changchun China

ISBN: (纸本)9781728150291

Unmanned aerial vehicles (UAVs) are widely used in aerial photography nowadays for their strong maneuverability, good image quality and high cost performance, while they have limited battery capacity and difficulty in operation. It is difficult to reach the ideal pose and get the desired shot by manual manipulation in a short time. To solve these problems, an UAV pose estimation system guided by desired shot is constructed. Either a sequence of images or a video is taken as input. The user interacts with the system to express the desired shot. The 3D-reconstruction is done by Structure from Motion (SfM). Then the Perspective-n-Point (PnP) problem is solved to estimate the ideal pose of the UAV. The experimental results proves that the system is effective and valid. © 201. IEEE.

关键词： Aerial photography

来源：评论

学校读者我要写书评

暂无评论

Dual-proxies Contrast-focused Loss in Domain Generalization

引用

Expert Systems with Applications 2025年

作者： Shihao Liu Pingping Liu Qiuzhan Zhou College of Computer Science and Technology Jilin University Changchun 130012 China Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education Jilin University Changchun 130012 China School of Mechanical Science and Engineering Jilin University Changchun 130025 China College of Communication Engineering Jilin University Changchun 130012 China

Domain generalization aims to train a model that can directly generalize to unseen domains after learning from different source domains. Due to the significant distribution discrepancy between training domains and the unseen target domain, model performance can be significantly affected. To address the issue, one current trend is to develop contrastive learning frameworks that achieve clustering effects by aligning features in the embedding space. Among existing solutions under contrastive learning frameworks, contrastive-based methods interact with intra-class and inter-class samples indiscriminately with equal intensity. Such indiscriminate interaction can be insufficient for effectively clustering domains with large gaps due to the influence of domain discrepancy. In contrast, metric learning frameworks, as traditional approaches for embedding space optimization, widely implement constructing class proxies as their central technique. However, when applied in domain generalization, proxy-based methods tend to neglect the rich semantic relationships between samples and fail to account for the domain-specific distributional relationships among classes. To resolve these limitations, we integrate contrastive-based methods with proxy-based methods, proposing a dual-proxy contrastive learning framework. Focusing on sample relations, we pay special attention to same-class cross-domain samples and assign greater interaction intensity to reduce feature discrepancy across domains. For class relations, we build two groups of centroid proxies: random proxy groups that adjust class-internal distributions along model optimization and mean proxy groups that represent statistical information to calibrate class-external distributions. Ultimately, we integrate the explorations from both the sample and class perspectives. We evaluate our method on five benchmark datasets, including PACS, VLCS, OfficeHome, TerraIncognita, and DomainNet, and the results validate its effectiveness with

关键词：

来源：评论

学校读者我要写书评

暂无评论

Image retrieval algorithm based on response value center weighted convolution feature

引用

Jilin Daxue Xuebao (Gongxueban)/Journal of Jilin University (engineering and Technology Edition) 2022年第11期52卷 2669-2675页

作者： Li, Xiao-Ning Zhao, Hong-Wei Zhang, Dan-Yang Zhang, Yuan College of Computer Science and Technology Jilin University Changchun130012 China College of Computer Science and Technology Changchun Normal University Changchun130032 China Key Laboratory of Symbolic Computation and Knowledge Engineering Ministry of Education Jilin University Changchun130012 China

A novel cross-entropy-based loss is proposed，which is applied to a classical convolutional neural network to obtain a better embedding space. A convolutional feature aggregation algorithm based on the weighting of the center of the response value is designed to process the three-dimensional convolutional features obtained by the neural network. The algorithm obtains the spatial weighting coefficient of the feature map by calculating the current position response value and the Gaussian center. The algorithm reduces the loss of information when the 3D convolution feature map is reduced to a 1. image feature descriptor and realizes the enhancement of the target area. Finally，the obtained image feature descriptors are used for retrieval tasks. On the CUB-200-201. dataset，the effectiveness of the loss function and the response value center weighting algorithm are respectively verified by ablation experiments. Compared with the current retrieval methods，higher precision and recall rates have been obtained on the four datasets of Paris6k，Oxford5k，CUB-200-201. and CARS1.6. © 2022 Editorial Board of Jilin University. All rights reserved.

关键词： Convolution

来源：评论

学校读者我要写书评

暂无评论

GEOM-GCN: GEOMETRIC GRAPH CONVOLUTIONAL NETWORKS 8

GEOM-GCN: GEOMETRIC GRAPH CONVOLUTIONAL NETWORKS

引用

8th International Conference on Learning Representations, ICLR 2020

作者： Pei, Hongbin Wei, Bingzhe Chang, Kevin Chen-Chuan Lei, Yu Yang, Bo College of Computer Science and Technology Jilin University China Department of Electrical and Computer Engineering University of Illinois Urbana-Champaign United States Department of Computer Science University of Illinois Urbana-Champaign United States Department of Computing Hong Kong Polytechnic University Hong Kong Key Laboratory of Symbolic Computation and Knowledge Engineering of Ministry of Education China

Message-passing neural networks (MPNNs) have been successfully applied to representation learning on graphs in a variety of real-world applications. However, two fundamental weaknesses of MPNNs' aggregators limit their ability to represent graph-structured data: losing the structural information of nodes in neighborhoods and lacking the ability to capture long-range dependencies in disassortative graphs. Few studies have noticed the weaknesses from different perspectives. From the observations on classical neural network and network geometry, we propose a novel geometric aggregation scheme for graph neural networks to overcome the two weaknesses. The behind basic idea is the aggregation on a graph can benefit from a continuous space underlying the graph. The proposed aggregation scheme is permutation-invariant and consists of three modules, node embedding, structural neighborhood, and bi-level aggregation. We also present an implementation of the scheme in graph convolutional networks, termed Geom-GCN, to perform transductive learning on graphs. Experimental results show the proposed Geom-GCN achieved state-of-the-art performance on a wide range of open datasets of graphs. © 2020 8th International Conference on Learning Representations, ICLR 2020. All rights reserved.

关键词： Message passing

来源：评论

学校读者我要写书评

暂无评论

knowledge Compilation Methods Based on the Clausal Relevance and Extension Rule

引用

Chinese Journal of Electronics 2018年第5期27卷 1037-1042页

作者： NIU Dangdang LIU Lei LYU Shuai College of Computer Science and Technology Jilin University Key Laboratory of Symbolic Computation and Knowledge Engineering (Jilin University) Ministry of Education

We introduce the concepts of Relevancematrix(RM) and Relevance-set(RS). And we construct the association between RM and the knowledge compilation(KC) methods based on Extension rule(ER). Based on the basic parameters of RS and the relationship between RM and the KC methods based on ER, we design two efficient heuristics, called M2S(maximum sum of elements in RS and sum of literals in RS) and MNE(minimum number of maximum terms not extended by RS). Both of above heuristics intend to find the minimum set of maximum terms which cannot be extended by RS. Furthermore,we apply M2S and MNE on KCER. M2S KCER(KCER with M2S) and MNE KCER(KCER with MNE) are designed and implemented based on M2S and MNE, respectively. Experimentally, for the SAT instances with random lengths of clauses, M2S KCER and MNE KCER can improve the efficiency and quality of KCER sharply, and they are two best KC algorithms of EPCCL(each pair contains complementary literal) theory in all KC algorithms based on KCER.

关键词： Extension rule knowledge compilation Clausal relevance Heuristic strategy

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：