检索结果-内蒙古大学图书馆

SSRN 2023年

作者： Wang, Xin Tan, Ning Zhong, Zhaohui Huang, Kai School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education Sun Yat-sen University Guangzhou China

For tasks utilizing redundant manipulators, there involves motion of a number of joints when performing tracking control. In some cases, one or a few of joints might fail and then result in task failure or even damage, which means that fault tolerance is an indispensable capability for robot control systems. In this paper, we propose a novel fault-tolerant scheme to endow the redundant robots with this capability in the condition that certain joints are not working. More importantly, our scheme can achieve the fault-tolerant tracking control in a data-driven manner based on zeroing neurodyanmics when the robot kinematic model is unavailable, which is much different from existing fault-tolerant control algorithms of redundant robots that require to the kinematic model in advance. The proposed model-less fault tolerant tracking control (MLFC) scheme considers the fault-tolerant control problem of redundant manipulator as a quadratic programming (QP) problem. Then, with the help of a discrete solver proposed in this paper, the fault-tolerant control problem is transformed into a unified computing problem in the condition that without knowing the redundant manipulator's model. Extensive simulations and experiments based on a six degrees of freedom (DOF) redundant manipulator are analyzed the verification of our scheme. © 2023, The Authors. All rights reserved.

关键词： Fault tolerance

来源：评论

学校读者我要写书评

暂无评论

Advancements in Intelligent Photonics: The Role of Silicon Photonics and Phase Change Materials in High-Efficiency computing 15

Advancements in Intelligent Photonics: The Role of Silicon P...

引用

15th International Conference on Information Optics and Photonics, CIOP 2024

作者： Yun, Yiting Xu, Kai Wang, Weiquan Wei, Maoliang Li, Junying Chen, Zequn Li, Lan Lin, Hongtao The State Key Lab of Brain-Machine Intelligence Key Laboratory of Micro-Nano Electronics and Smart System of Zhejiang Province College of Information Science and Electronic Engineering Zhejiang University Zhejiang Hangzhou310027 China Hangzhou Institute for Advanced Study University of Chinese Academy of Sciences Zhejiang Hangzhou310024 China Key Laboratory of 3D Micro/Nano Fabrication and Characterization of Zhejiang Province School of Engineering Westlake University Zhejiang Hangzhou310030 China

ISBN: (数字)9781510686175

ISBN: (纸本)9781510686168

Intelligent photonics, driven by silicon photonics, is revolutionizing high-speed data processing, low-power computing, and precision sensing. Leveraging these advances, photonic chips are enabling the development of optical neural networks and nonlinear activation mapping, which are crucial for addressing the demands of large generative models. However, traditional on-chip control methods struggle with high power consumption and volatility. To overcome these challenges, phase-change materials(PCMs) offer high optical contrast and non-volatility, enhancing integration density and reducing power usage. This article discusses the performance and reversible control of PCMs and their integration with silicon photonics. By incorporating PCMs into in-memory optical computing chips, we achieved 4-bit storage and over 88% accuracy on the MNIST dataset, marking significant progress in next-generation high-performance computing. © 2024 SPIE.

关键词： Silicon photonics

来源：评论

学校读者我要写书评

暂无评论

Slowly Expanding Neural Network for Class Incremental Learning

SSRN

引用

SSRN 2023年

作者： Xu, Zhengjin Li, Xuyang Chang, Xiaobin Zheng, Wei-Shi Wang, Ruixuan School of Computer Science and Engineering Sun Yat-sen University Guangzhou China Department of Network Intelligence Pengcheng Laboratory Shenzhen China Key Laboratory of Machine Intelligence and Advanced Computing MOE Guangzhou China School of Artificial Intelligence Sun Yat-sen University Zhuhai China

Currently deep learning models often rapidly forget the knowledge of old classes when they are continually updated to learn knowledge of new classes. To alleviate such catastrophic forgetting issues, the state-of-the-art approach often freezes the learned feature extractor to preserve the old knowledge and introduces an additional network when learning new knowledge each time. The forgetting issue can be effectively handled by this method at the price of rapid model expansion. In this paper, we propose a novel continual learning framework, called SEIL, with a much Slower model Expansion rate and better Incremental Learning performance than its counterparts. Specifically, instead of introducing an entire feature extractor, the majority of network parameters are shared during continual learning, and only the remaining lightweight modules are expanded for each new task. An outof-the-distribution (OOD) technique is also applied to two auxiliary classifiers for more class-balanced predictions. The proposed method achieves new state-of-theart results on multiple benchmark datasets under the standard continual learning settings. The source code will be released publicly. © 2023, The Authors. All rights reserved.

关键词： Image classification

来源：评论

学校读者我要写书评

暂无评论

Continual Learning with Bayesian Model based on a Fixed Pre-trained Feature Extractor

arXiv

引用

arXiv 2022年

作者： Yang, Yang Cui, Zhiying Xu, Junjie Zhong, Changhong Zheng, Wei-Shi Wang, Ruixuan School of Computer Science and Engineering Sun Yat-Sen University China Key Laboratory of Machine Intelligence and Advanced Computing MOE China

Deep learning has shown its human-level performance in various applications. However, current deep learning models are characterised by catastrophic forgetting of old knowledge when learning new classes. This poses a challenge particularly in intelligent diagnosis systems where initially only training data of a limited number of diseases are available. In this case, updating the intelligent system with data of new diseases would inevitably downgrade its performance on previously learned diseases. Inspired by the process of learning new knowledge in human brains, we propose a Bayesian generative model for continual learning built on a fixed pre-trained feature extractor. In this model, knowledge of each old class can be compactly represented by a collection of statistical distributions, e.g. with Gaussian mixture models, and naturally kept from forgetting in continual learning over time. Unlike existing class-incremental learning methods, the proposed approach is not sensitive to the continual learning process and can be additionally well applied to the data-incremental learning scenario. Experiments on multiple medical and natural image classification tasks showed that the proposed approach outperforms state-of-the-art approaches which even keep some images of old classes during continual learning of new classes. © 2022, CC BY.

关键词： Bayesian networks

来源：评论

学校读者我要写书评

暂无评论

A Method of Garbage Quantity Prediction Based on Population Change 12th

A Method of Garbage Quantity Prediction Based on Population ...

引用

12th IFIP TC 12 International Conference on Intelligent Information Processing, IIP 2022

作者： Yu, Qiumei Wan, Hongjie Ma, Junchen Li, Huakang Sun, Guozi Key Laboratory of Urban Land Resources Monitoring and Simulation MNR Shenzhen China School of Computer Science NUPT Nanjing China School of Artificial Intelligence and Advanced Computing XJTLU Suzhou China

ISBN: (纸本)9783031039478

Aiming at the problem that the amount of urban waste changes due to population changes and is difficult to predict, a method for predicting the amount of waste based on urban population changes is proposed. After analyzing the correlation between the urban population data of Shanghai and the annual output of garbage from 2000 to 2019, the correlation coefficient and strong correlation data items are calculated. On this basis, judge whether the original population data items meet the conditions of the grey prediction model, and determine whether it can be modeled according to the grey model. Based on the grey theory, this paper analyzes the basic situation of population changes and future population growth trend in Shanghai. Finally, the annual production of municipal solid waste in Shanghai from 2020 to 2025 is predicted based on multiple linear regression analysis. The results show that the waste production of Shanghai has shown a slow growth trend since 2019, and will not increase significantly in the natural state in recent years, which provides a reference basis for subsequent research and analysis. © 2022, IFIP International Federation for Information Processing.

关键词： Multiple linear regression

来源：评论

学校读者我要写书评

暂无评论

The Enemy of My Enemy is My Friend: Exploring Inverse Adversaries for Improving Adversarial Training

The Enemy of My Enemy is My Friend: Exploring Inverse Advers...

引用

Conference on Computer Vision and Pattern Recognition (CVPR)

作者： Junhao Dong Seyed-Mohsen Moosavi-Dezfooli Jianhuang Lai Xiaohua Xie School of Computer Science and Engineering Sun Yat-Sen University China Imperial College London UK Guangdong Province Key Laboratory of Information Security Technology China Key Laboratory of Machine Intelligence and Advanced Computing Ministry of Education China

Although current deep learning techniques have yielded superior performance on various computer vision tasks, yet they are still vulnerable to adversarial examples. Adversarial training and its variants have been shown to be the most effective approaches to defend against adversarial examples. A particular class of these methods regularize the difference between output probabilities for an adversarial and its corresponding natural example. However, it may have a negative impact if a natural example is misclassified. To circumvent this issue, we propose a novel adversarial training scheme that encourages the model to produce similar output probabilities for an adversarial example and its “inverse adversarial” counterpart. Particularly, the counterpart is generated by maximizing the likelihood in the neighborhood of the natural example. Extensive experiments on various vision datasets and architectures demonstrate that our training method achieves state-of-the-art robustness as well as natural accuracy among robust models. Furthermore, using a universal version of inverse adversarial examples, we improve the performance of single-step adversarial training techniques at a low computational cost.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Event-Triggered Group Attitude Coordinated Control of Multi-spacecraft System with Directed Topology

Event-Triggered Group Attitude Coordinated Control of Multi-...

引用

International Conference on Guidance, Navigation and Control, ICGNC 2020

作者： Zhou, Shaolei Wang, Shuailei Liu, Wei Wang, Donglai Gao, Xiangyang Naval Aviation University Yantai Shandong264001 China CAS Key Laboratory of human-Machine Intelligence-Synergy Systems Shenzhen Institute of Advanced Technology Chinese Academy of Science Shenzhen518055 China

ISBN: (纸本)9789811581540

Event-triggered group attitude coordinated control of multi-spacecraft system with directed topology is studied. The event-triggered strategy ensures that the system communicate at discrete time instants, thus the communication burden is reduced. Based on assumptions on the topology, the Laplacian matrix of the multi-spacecraft system is a block upper triangular matrix, and the nonnegative eigenvalues of it are positive integers, which can simplify the analysis process. With variable substitution and the Laplacian matrix decomposition, the problem is turned into a second-order group consensus problem, and then turned into a stability problem of a linear system. By properly chosen parameters, the system matrix of the linear system is Hurwitz. With centralized event-triggered mechanism, the control input proposed can lead the system to group attitude coordination, and Zeno behavior is avoided. Simulation results show that the variable substitution method utilized and controller designed in this paper are effective. © 2022, The Editor(s) (if applicable) and The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Linear systems

来源：评论

学校读者我要写书评

暂无评论

Class-distribution-aware pseudo-labeling for semi-supervised multi-label learning 23

Class-distribution-aware pseudo-labeling for semi-supervised...

引用

Proceedings of the 37th International Conference on Neural Information Processing Systems

作者： Ming-Kun Xie Jia-Hao Xiao Hao-Zhe Liu Gang Niu Masashi Sugiyama Sheng-Jun Huang Nanjing University of Aeronautics and Astronautics and MIIT Key Laboratory of Pattern Analysis and Machine Intelligence Nanjing China RIKEN Center for Advanced Intelligence Project RIKEN Center for Advanced Intelligence Project and The University of Tokyo Tokyo Japan

Pseudo-labeling has emerged as a popular and effective approach for utilizing unlabeled data. However, in the context of semi-supervised multi-label learning (SSMLL), conventional pseudo-labeling methods encounter difficulties when dealing with instances associated with multiple labels and an unknown label count. These limitations often result in the introduction of false positive labels or the neglect of true positive ones. To overcome these challenges, this paper proposes a novel solution called Class-Aware Pseudo-Labeling (CAP) that performs pseudo-labeling in a class-aware manner. The proposed approach introduces a regularized learning framework incorporating class-aware thresholds, which effectively control the assignment of positive and negative pseudo-labels for each class. Notably, even with a small proportion of labeled examples, our observations demonstrate that the estimated class distribution serves as a reliable approximation. Motivated by this finding, we develop a class-distribution-aware thresholding strategy to ensure the alignment of pseudo-label distribution with the true distribution. The correctness of the estimated class distribution is theoretically verified, and a generalization error bound is provided for our proposed method. Extensive experiments on multiple benchmark datasets confirm the efficacy of CAP in addressing the challenges of SSMLL problems. The implementation is available at https://***/milkxie/SSMLL-CAP.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Finding Emotional Focus for Emotion Recognition at Sentence Level

引用

Chinese Journal of Electronics 2023年第1期22卷 99-103页

作者： Changqin Quan Fuji Ren Anhui Province Key Laboratory of Affective Computing and Advanced Intelligent Machine Hefei University of Technology Hefei China Faculty of Engineering The University of Tokushima Tokushima Japan

Emotion recognition at sentence level is one of the fundamental problems of textual emotion understanding. Based on the observation that sentence emotional focus can be expressed by some clauses in this sentence, this paper proposes to find the emotional focus for sentence emotion recognition. For the sake of breaking through the problems brought about by depending on emotion lexicons, we first recognize word emotions in a sentence based on Maximum entropy model. And then homogeneous Markov model is built for clause emotion recognition; After that, a strategy based on emotion selection is proposed for a sentence with multiple clauses, and genetic algorithm is used for clause selection by textual feature weighting. The experimental results show that, comparing with the baseline, there are 9.1% and 3.6% improvement respectively for two different evaluations. It is demonstrated that finding emotional focus by clause selection is able to improve the performance of sentence emotion recognition significantly.

关键词： Emotion recognition Speech recognition Markov processes Entropy Real-time systems Task analysis Research and development

来源：评论

学校读者我要写书评

暂无评论

Skeleton-Embedded Network for action recognition

Skeleton-Embedded Network for action recognition

引用

2023 IEEE International Conference on Real-Time computing and Robotics, RCAR 2023

作者： Luo, Li Ren, Ziliang Qin, Yong Zhang, Qieshi Gao, Xiangyang Dongguan University of Technology Department of Computer Science Dongguan52300 China Chinese Academy of Sciences Cas Key Laboratory of Human-Machine Intelligence-Synergy Systems Shenzhen Institutes of Advanced Technology Shenzhen518055 China The Chinese University of HongKong Hong Kong Hong Kong

ISBN: (纸本)9798350327182

Multimodality-based human action recognition is becoming an increasingly attractive topic as different modalities can provide complementary information. RGB and skeleton data have their pros and limitations for action recognition because they describe the action from distinctive views. Therefore, the fusion of these two modalities using their complementarity is meaningful to action recognition. However, it is difficult to completely utilize the supplementarity of RGB and skeleton data. In this paper, to take advantage of their complementary at the semantic level, A skeleton-embedded network(SE-Net) is proposed by us. The architecture is based on integrating their feature in the network not only a post-hoc fusion. With a sparse sampling strategy, which is efficient for learning using the whole action video. To examine the performance of our method, we conducted experiments on the wildly used datasets, obtaining an accuracy of 86.4% in C-sub and 88.2% in C-view on NTU-RGB+D 60. For NTU-RGB+D 120, the accuracy of 81.3% in C-sub and 83.4% in C-set is obtained. © 2023 IEEE.

关键词： Semantics

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：