检索结果-内蒙古大学图书馆

IEEE International Conference on data Mining (ICDM)

作者： Zhiding Liu Mingyue Cheng Zhi Li Qi Liu Enhong Chen Anhui Province Key Laboratory of Big Data Analysis and Application University of Science and Technology of China Hefei China State Key Laboratory of Cognitive Intelligence Hefei China Shenzhen International Graduate School Tsinghua University Shenzhen China

Deep learning has brought significant breakthroughs in sequential recommendation (SR) for capturing dynamic user interests. A series of recent research revealed that models with more parameters usually achieve optimal performance for SR tasks, inevitably resulting in great challenges for deploying them in real systems. Following the simple assumption that light networks might already suffice for certain users, in this work, we propose CANet, a conceptually simple yet very scalable framework for assigning adaptive network architecture in an input-dependent manner to reduce unnecessary computation. The core idea of CANet is to route the input user behaviors with a light-weighted router module. Specifically, we first construct the routing space with various submodels parameterized in terms of multiple model dimensions such as the number of layers, hidden size and embedding size. To avoid extra storage overhead of the routing space, we employ a weight-slicing schema to maintain all the submodels in exactly one network. Furthermore, we leverage several solutions to solve the discrete optimization issues caused by the router module. Thanks to them, CANet could adaptively adjust its network architecture for each input in an end-to-end manner, in which the user preference can be effectively captured. To evaluate our work, we conduct extensive experiments on benchmark datasets. Experimental results show that CANet reduces computation by 55 ~ 65% while preserving the accuracy of the original model. Our codes are available at https://***/CANet.

关键词： Deep learning Adaptation models Adaptive systems Computational modeling Scalability Network architecture Routing

来源：评论

学校读者我要写书评

暂无评论

Turning Channel Noise into an Accelerator for Over-the-Air Principal Component Analysis

arXiv

引用

arXiv 2021年

作者： Zhang, Zezhong Zhu, Guangxu Wang, Rui Lau, Vincent K.N. Huang, Kaibin The University of Hong Kong Hong Kong Shenzhen Research Institute of Big Data China Southern University of Science and Technology China The The Hong Kong University of Science and Technology Hong Kong

The enormous data distributed at the network edge and ubiquitous connectivity have led to the emergence of the new paradigm of distributed machine learning and large-scale data analytics. Distributed principal component analysis (PCA) concerns finding a low-dimensional subspace that contains the most important information of high-dimensional data distributed over the network edge. The subspace is useful for distributed data compression and feature extraction. This work advocates the application of over-the-air federated learning to efficient implementation of distributed PCA in a wireless network under a data-privacy constraint, termed AirPCA. The design features the exploitation of the waveform-superposition property of a multi-access channel to realize over-the-air aggregation of local subspace updates computed and simultaneously transmitted by devices to a server, thereby reducing the multi-access latency. The original drawback of this class of techniques, namely channel-noise perturbation to uncoded analog modulated signals, is turned into a mechanism for escaping from saddle points during stochastic gradient descent (SGD) in the AirPCA algorithm. As a result, the convergence of the AirPCA algorithm is accelerated. To materialize the idea, descent speeds in different types of descent regions are analyzed mathematically using martingale theory by accounting for wireless propagation and techniques including broadband transmission, over-the-air aggregation, channel fading and noise. The results reveal the accelerating effect of noise in saddle regions and the opposite effect in other types of regions. The insight and results are applied to designing an online scheme for adapting receive signal power to the type of current descent region. Specifically, the scheme amplifies the noise effect in saddle regions by reducing signal power and applies the power savings to suppressing the effect in other regions. From experiments using real datasets, such power control is fo

关键词： Principal component analysis

来源：评论

学校读者我要写书评

暂无评论

Towards Variable-Length Textual Adversarial Attacks

arXiv

引用

arXiv 2021年

作者： Guo, Junliang Zhang, Zhirui Zhang, Linlin Xu, Linli Chen, Boxing Chen, Enhong Luo, Weihua Anhui Province Key Laboratory of Big Data Analysis and Application School of Computer Science and Technology University of Science and Technology of China China Alibaba DAMO Academy Zhejiang University China

Adversarial attacks have shown the vulnerability of machine learning models, however, it is non-trivial to conduct textual adversarial attacks on natural language processing tasks due to the discreteness of data. Most previous approaches conduct attacks with the atomic replacement operation, which usually leads to fixed-length adversarial examples and therefore limits the exploration on the decision space. In this paper, we propose variable-length textual adversarial attacks (VL-Attack) and integrate three atomic operations, namely insertion, deletion and replacement, into a unified framework, by introducing and manipulating a special blank token while attacking. In this way, our approach is able to more comprehensively find adversarial examples around the decision boundary and effectively conduct adversarial attacks. Specifically, our method drops the accuracy of IMDB classification by 96% with only editing 1.3% tokens while attacking a pre-trained BERT model. In addition, fine-tuning the victim model with generated adversarial samples can improve the robustness of the model without hurting the performance, especially for length-sensitive models. On the task of non-autoregressive machine translation, our method can achieve 33.18 BLEU score on IWSLT14 German-English translation, achieving an improvement of 1.47 over the baseline model. Copyright © 2021, The Authors. All rights reserved.

关键词： Natural language processing systems

来源：评论

学校读者我要写书评

暂无评论

Image Texture Analysis Based on Gray Level Co-Occurrence Matrix

Image Texture Analysis Based on Gray Level Co-Occurrence Mat...

引用

IEEE International Symposium on Information (IT) in Medicine and Education, ITME

作者： Chen Meilong Yu Xiehua Su Shaojun Lin Daqing Key Laboratory of Big Data and Artificial Intelligence University of Fujian Province MinNan Science and Technology University Quanzhou China State Grid Fujian Electric Power Co. Ltd. Nan'an Power Supply Company Quanzhou China Nan 'an NO.3 Primary School of Fujian Province Quanzhou China

This paper studies the algorithm of Gray Level Co-occurrence Matrix (GLCM), explains the concrete meaning of 14 texture features based on GLCM, and points out the redundancy among texture features. Through the calculation and analysis of the gray level co-occurrence matrix of texture image and the experiments of texture feature extraction, it is shown that the gray level co-occurrence matrix can reflect the features of the image, and corresponds to the features of the image described by the texture features. At the same time, there is a certain degree of redundancy among the 14 texture features of the image. In practice, we can choose several prominent texture features to classify the image according to the difference of texture features. The analysis of texture features and experimental results are of great significance to the application of image texture features.

关键词：

来源：评论

学校读者我要写书评

暂无评论

FPL+: Filtered Pseudo Label-based Unsupervised Cross-Modality Adaptation for 3D Medical Image Segmentation

arXiv

引用

arXiv 2024年

作者： Wu, Jianghao Guo, Dong Wang, Guotai Yue, Qiang Yu, Huijun Li, Kang Zhang, Shaoting The School of Mechanical and Electrical Engineering University of Electronic Science and Technology of China Chengdu611731 China Shanghai AI laboratory Shanghai200030 China The Department of Radiology West China Hospital Sichuan University Chengdu610041 China West China Biomedical Big Data Center West China Hospital Sichuan University Chengdu610041 China SenseTime Research Shanghai200233 China

Adapting a medical image segmentation model to a new domain is important for improving its cross-domain transferability, and due to the expensive annotation process, Unsupervised Domain Adaptation (UDA) is appealing where only unlabeled images are needed for the adaptation. Existing UDA methods are mainly based on image or feature alignment with adversarial training for regularization, and they are limited by insufficient supervision in the target domain. In this paper, we propose an enhanced Filtered Pseudo Label (FPL+)-based UDA method for 3D medical image segmentation. It first uses cross-domain data augmentation to translate labeled images in the source domain to a dual-domain training set consisting of a pseudo source-domain set and a pseudo target-domain set. To leverage the dual-domain augmented images to train a pseudo label generator, domain-specific batch normalization layers are used to deal with the domain shift while learning the domain-invariant structure features, generating high-quality pseudo labels for target-domain images. We then combine labeled source-domain images and target-domain images with pseudo labels to train a final segmentor, where image-level weighting based on uncertainty estimation and pixel-level weighting based on dual-domain consensus are proposed to mitigate the adverse effect of noisy pseudo labels. Experiments on three public multi-modal datasets for Vestibular Schwannoma, brain tumor and whole heart segmentation show that our method surpassed ten state-of-the-art UDA methods, and it even achieved better results than fully supervised learning in the target domain in some cases. © 2024, CC BY.

关键词： Image enhancement

来源：评论

学校读者我要写书评

暂无评论

Benefits of permutation-equivariance in auction mechanisms 22

Benefits of permutation-equivariance in auction mechanisms

引用

Proceedings of the 36th International Conference on Neural Information Processing Systems

作者： Tian Qin Fengxiang He Dingfeng Shi Wenbing Huang Dacheng Tao University of Science and Technology of China and JD Explore Academy *** Inc. JD Explore Academy *** Inc. Gaoling School of Artificial Intelligence Renmin University of China and Beijing Key Laboratory of Big Data Management and Analysis Methods

ISBN: (纸本)9781713871088

Designing an incentive-compatible auction mechanism that maximizes the auctioneer's revenue while minimizes the bidders' ex-post regret is an important yet intricate problem in economics. Remarkable progress has been achieved through learning the optimal auction mechanism by neural networks. In this paper, we consider the popular additive valuation and symmetric valuation setting; i.e., the valuation for a set of items is defined as the sum of all items' valuations in the set, and the valuation distribution is invariant when the bidders and/or the items are permutated. We prove that permutation-equivariant neural networks have significant advantages: the permutation-equivariance decreases the expected ex-post regret, improves the model generalizability, while maintains the expected revenue invariant. This implies that the permutation-equivariance helps approach the theoretically optimal dominant strategy incentive compatible condition, and reduces the required sample complexity for desired generalization. Extensive experiments fully support our theory. To our best knowledge, this is the first work towards understanding the benefits of permutation-equivariance in auction mechanisms.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A Trajectory Tracking System for Zebrafish Based on Embedded Edge Artificial Intelligence

A Trajectory Tracking System for Zebrafish Based on Embedded...

引用

International Conference on Parallel and Distributed Systems (ICPADS)

作者： Chuanhao Zang Anming Dong Jiguo Yu Key Laboratory of Computing Power Network and Information Security Ministry of Education Shandong Computer Science Center (National Supercomputer Center in Jinan) Qilu University of Technology (Shandong Academy of Sciences) Jinan China School of Computer Science and Technology Qilu University of Technology (Shandong Academy of Sciences) Jinan China Big Data Institute Qilu University of Technology (Shandong Academy of Sciences) Jinan China Shandong Provincial Key Laboratory of Computer Networks Shandong Fundamental Research Center for Computer Science Jinan China

Trajectory tracking of zebrafish is an important requirement in studying neurological disorders and developing new psychoactive drug. However, many challenges emerge for stable tracking, since zebrafish are similar in appearance, occlusion, agile, non-linear in moving and easy to swarm, all of which will lead to mistrack for multiple fish. And there is no precedent for tracking zebrafish through embedded edge artificial intelligence device. To overcome these difficulties, we present a tracking system for zebrafish based on RK3588-S. First, We construct an embedded edge AI hardware system consisting of two cameras driven by the RK3588-S, one for the front view and the other for top view of the fish. Then, we develop a 2D tracking algorithm based on YOLOv5 and the Observation-Centric Simple online and real-time tracking (OC-SORT) algorithm, which are transplanted to the RK3588-S for tracking the top and front views of the fish at the edge device. Compared with the previous methods, our method has fewer ID exchanges and highly real-time. Moreover, we apply the MQTT mechanism to establish communication links between the edge and the cloud to reliably transmit data to the cloud. The correspondence between cloud server and embedded AI is 1: N. Finally, we design a multi-view data fusion association algorithm to fuse the data of the two views in the cloud, which are further utilized to build the 3D tracklets of the zebrafish.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A binary particle swarm optimization with the hybrid s-shaped and v-shaped transfer function 13th

A binary particle swarm optimization with the hybrid s-shape...

引用

13th International Conference on Genetic and Evolutionary Computing, ICGEC 2019

作者： Jiang, Lei Liu, Jianhua Cui, Dongli Bu, Guannan Zhang, Dongyang Hu, Renyuan School of Information Science and Engineering Fujian University of Technology Fuzhou350118 China Fujian Provincial Key Laboratory of Big Data Mining and Applications Fuzhou350118 China

ISBN: (纸本)9789811533075

Binary Particle Swarm Optimization has too strong global search ability and lacks local search ability in the later stage, because it has a unreasonable transfer function. According to the analysis of the transfer function, the hybrid transfer functions of S-shaped and V-shaped function has been proposed to improve the BSPO, and an adaptive mutation method is used to obtain a new binary PSO with hybrid transfer function. The new Binary PSO has global search ability in the early stage, and has local search ability in the later stage. The experiments conducted have proved that it outperforms BPSO. © Springer Nature Singapore Pte Ltd. 2020.

关键词： Particle swarm optimization (PSO)

来源：评论

学校读者我要写书评

暂无评论

A Multi-Mesh Approach for Accurate Computation of Multi-Target Functionals in Aerodynamics Design

SSRN

引用

SSRN 2024年

作者： Hu, Guanghui Li, Ruo Wang, Jingfeng Department of Mathematics University of Macau Macao China State Key Laboratory of Internet of Things for Smart City University of Macau Macao China Zhuhai UM Science & Technology Research Institute Guangdong Province Zhuhai China CAPT LMAM and School of Mathematical Sciences Peking University Beijing100871 China Chongqing Research Institute of Big Data Peking University Chongqing401121 China

Aerodynamic optimal design is crucial for enhancing performance of aircrafts, while calculating multi-target functionals through solving dual equations with arbitrary right-hand sides remains challenging. In this paper, a novel multi-target framework of DWR-based mesh refinement is proposed and analyzed. Theoretically, an extrapolation method is generalized to expand multi-variable functionals, which guarantees the dual equations of different objective functionals can be calculated separately. Numerically, an algorithm of calculating multi-target functionals is designed based on the multi-mesh approach, which can help to obtain different dual solutions simultaneously. One feature of our framework is the algorithm is easy to implement with the help of the hierarchical geometry tree structure and the calculation avoids the Galerkin orthogonality naturally. The framework takes a balance between different targets even when they are not the same orders of magnitude. While existing approach uses a linear combination of different components in multi-target functionals for adaptation, it introduces additional coefficients for adjusting. With each component calculated under a dual-consistent scheme, this multi-mesh framework addresses challenges such as the lift-drag ratio and other kinds of multi-target functionals, ensuring smooth convergence and precise calculations of dual solutions. © 2024, The Authors. All rights reserved.

关键词： Trees (mathematics)

来源：评论

学校读者我要写书评

暂无评论

Quantifying Bubble-Induced Diffusion Resistance Through Real-Time Sam-Assisted Yolo High Density Bubble Detection Algorithm

SSRN

引用

SSRN 2025年

作者： Wang, Bowen Lv, Haoyang Wang, Xun Hao, Minghui Guay, Daniel Kirk, Donald W. Thorpe, Steven Ruan, Zhengsen Zhejiang Key Laboratory of Big data and Future E-Commerce Technology Zhejiang Gongshang University. Zhejiang Hangzhou China Surface Engineering and Electrochemistry Group Department of Material Science and Engineering University of Toronto TorontoON Canada Énergie Matériaux Télécommunications Research Centre INRS VarennesQC Canada Department of Chemical Engineering and Applied Chemistry University of Toronto TorontoON Canada

The chemistry and localized structure of materials greatly affect their efficiency in gas evolution reactions leading to cell resistance loses. This work focuses on the role of morphological changes in the nature of the bubble distribution by a comparison of two different surface structures: polished Ni and various porous Ni surfaces fabricated by dynamic hydrogen bubble templating. This study introduces a refined method for counting bubbles in gas evolution reactions, focusing on high-density, overlapping bubble flows. By incorporating the Segment Anything Model (SAM), we significantly reduced the training cost of the You Only Look Once algorithm, maintaining accuracy while improving efficiency. Surface characteristics strongly influenced the bubble distribution, with superaerophobic surfaces exhibiting smaller Sauter Mean Diameter (D32) values than polished ones. A polynomial extrapolation indicated that D32s were roughly one third of the thickness of the bubble layer at higher current densities. Building on the established connection between bubble layer thickness and diffusion resistance through electrochemical impedance spectroscopy (EIS), this study establishes a direct relationship between D32 and diffusion resistance. We have also demonstrated that the characteristics of the bubble layer are independent of the electrochemical surface area (ECSA) beneath the surface. This combined approach uses EIS for broader current coverage and imaging for real-time data, providing a powerful framework for bubble management and performance monitoring in gas evolution applications. © 2025, The Authors. All rights reserved.

关键词： Hydrogen evolution reaction

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：