检索结果-内蒙古大学图书馆

Forget less,count better:a domain-incremental self-distillation learning benchmark for lifelong crowd counting

Frontiers of Information Technology & Electronic Engineering 2023年第2期24卷 187-202页

作者： Jiaqi GAO Jingqi LI Hongming SHAN Yanyun QU James ZWANG Fei-Yue WANG Junping ZHANG Shanghai Key Laboratory of Intelligent Information Processing School of Computer ScienceFudan UniversityShanghai 200433China Institute of Science and Technology for Brain-inspired Intelligence Fudan UniversityShanghai 200433China Shanghai Center for Brain Science and Brain-inspired Technology Shanghai 201210China School of Information Science and Technology Xiamen UniversityXiamen 361005China College of Information Sciences and Technology the Pennsylvania State UniversityUniversity ParkPA 16802USA State Key Laboratory of Management and Control for Complex Systems Institute of AutomationChinese Academy of SciencesBeijing 100190China

Crowd counting has important applications in public safety and pandemic control.A robust and practical crowd counting system has to be capable of continuously learning with the newly incoming domain data in real-world scenarios instead of fitting one domain ***-the-shelf methods have some drawbacks when handling multiple domains:(1)the models will achieve limited performance(even drop dramatically)among old domains after training images from new domains due to the discrepancies in intrinsic data distributions from various domains,which is called catastrophic forgetting;(2)the well-trained model in a specific domain achieves imperfect performance among other unseen domains because of domain shift;(3)it leads to linearly increasing storage overhead,either mixing all the data for training or simply training dozens of separate models for different domains when new ones are *** overcome these issues,we investigate a new crowd counting task in incremental domain training setting called lifelong crowd *** goal is to alleviate catastrophic forgetting and improve the generalization ability using a single model updated by the incremental ***,we propose a self-distillation learning framework as a benchmark(forget less,count better,or FLCB)for lifelong crowd counting,which helps the model leverage previous meaningful knowledge in a sustainable manner for better crowd counting to mitigate the forgetting when new data arrive.A new quantitative metric,normalized Backward Transfer(nBwT),is developed to evaluate the forgetting degree of the model in the lifelong learning *** experimental results demonstrate the superiority of our proposed benchmark in achieving a low catastrophic forgetting degree and strong generalization ability.

关键词： Crowd counting Knowledge distillation Lifelong learning

来源：评论

学校读者我要写书评

暂无评论

An IMU Error Correction Method Based on Multithreshold ZV Detection and SRCKF

An IMU Error Correction Method Based on Multithreshold ZV De...

引用

Cyber-Physical Social Intelligence (ICCSI), International Conference on

作者： Jialiang Luo Shichao Chen Yisheng Lv Wenqiao Sun Fan Yang Ruijie Xu The State Key Laboratory for Management and Control of Complex System State Key Laboratory of Multimodal Artificial Intelligence Systems Institute of Automation Chinese Academy of Sciences School of Information Engineering China University of Geosciences Beijing China The State Key Laboratory of Multimodal Artificial Intelligence Systems Institute of Automation Chinese Academy of Sciences The Center of National Railway Intelligent Transportation System Engineering and Technology China Academy of Railway Sciences Corporation Limited Beijing China The State Key Laboratory of Multimodal Artificial Intelligence Systems Institute of Automation Chinese Academy of Sciences Beijing China Transportation and Economics Research Institute and the Center of National Railway Intelligent Transportation System Engineering and Technology China Academy of Railway Sciences Corporation Limited China The State Key Laboratory for Management and Control of Complex System State Key Laboratory of Multimodal Artificial Intelligence Systems Institute of Automation Chinese Academy of Sciences College of Rail Transit Shan Dong JiaoTong University Jinan China Department of Aeronautical and Aviation Engineering The Hong Kong Polytechnic University Hong Kong China

With the rapid development and widespread application of Inertial Measurement Units (IMUs), IMU-based human localization has become critically important in environments lacking Global Navigation Satellite systems (GNSS) signals. Methods such as UWB/WIFI and other anchor-based localization have been employed to enhance the positioning accuracy. However, these methods are susceptible to the environmental interference and increased costs. In contrast, SINS-based localization methods eliminate reliance on the external data. However, this method can suffer from drift errors due to cumulative integration errors. Inertial navigation error correction techniques like Zero Detection Update Technology (ZUPT) provide potential solutions to address drift errors. This paper proposes an error correction method based on the multithreshold zero-velocity detection and Square Root Cubature Kalman Filter (SRCKF) updates. Compared to the single fixed threshold detection, the multi-threshold zero-velocity detection reduces the likelihood of missed zero velocity detection and false alarms, providing accurate time intervals for error correction. Experimental results demonstrate that the SRCKF-based algorithm reduces position deviation to approximately 0.3146 meters, corresponding to an error ratio of1.14%. The proposed method enables cost-effective and accurate pedestrian localization, driving advancements in the field.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Learning Representation with Q-irrelevance Abstraction for Reinforcement Learning 11

Learning Representation with Q-irrelevance Abstraction for R...

引用

11th International Conference on Intelligent control and Information Processing, ICICIP 2021

作者： Hao, Shuai Li, Luntong Liu, Minsong Zhu, Yuanheng Zhao, Dongbin Institute of Automation Chinese Academy of Sciences State Key Laboratory of Management and Control for Complex Systems Beijing China School of Software Beihang University Beijing China School of Artificial Intelligence University of Chinese Academy of Sciences Beijing China

ISBN: (纸本)9781665425155

In order to improve the performance of deep reinforcement learning (DRL) algorithm in high-dimensional observation environments, we propose a new auxiliary task to learn representations to aggregate task-relevant information of observations. Inspired by Q-irrelevance abstraction, our auxiliary task trains a deep Q-network (DQN) to predict the true Q value distribution over all discrete actions. Then we use the output of DQN to train the encoder to discriminate states with different Q values. The encoder is used as the representation of proximal policy optimization (PPO). The resulting algorithm is called as Q-irrelevance Abstraction for Reinforcement Learning (QIARL). After training, the encoder can aggregate states with similar Q value distributions together for any policy and any action. Thus the encoder can encode the important information that is relevant to reinforcement learning task. We test QIARL in four Procgen environments compare with PPO, A2C and Rainbow. The experimental results show QIARL outperforms the other three algorithms. © 2021 IEEE.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

A Simplified Variant for Graph Convolutional Network Based Knowledge Graph Completion Model 8

A Simplified Variant for Graph Convolutional Network Based K...

引用

8th International Conference on Data Science in Cyberspace, DSC 2023

作者： Wang, Yashen Li, Qiang Zhang, Yi Zhang, Xuecheng Ju, Zhuoya Sun, Jian Beijing China Scientific and Technological Innovation Center of Ari Institute of Automation Chinese Academy of Sciences The State Key Laboratory for Management and Control of Complex Systems Beijing China Cetc Academy of Electronics and Information Technology Group Co. Ltd. Beijing China Khoury College of Computer Science Northeastern University United States Scientific and Technological Innovation Center of Ari Beijing China Qi An Xin Technology Group Inc. Beijing China

ISBN: (纸本)9798350331035

Knowledge graphs (KGs) serve as useful resources for various applications including machine learning, data mining, and artificial intelligence. Knowledge Graph Completion (KGC) aims at reasoning over known (observed) facts and infer the missing (unobserved) relationships, to improve KG's coverage. Many recent works have laid stress on leveraging Graph Convolutional Networks (GCNs) for KGC task, which obtain good representation learning ability for graph-structured data, and have achieved satisfactory results. However, GCN-based researches continue to improve these lines of models by introducing an increasingly complex feature space. Especially, the hyper-parameter space of conventional GCN-based KGC model is too huge, resulting in low computational efficiency and excessive storage pressure. To overcome this problem, this paper investigates a simplified and generic variant for GCN-based KGC model, and shows that simple changes in the architecture of the underlying architecture (with a Subtraction Operator and an Addition Operator), can compte with state-of-the-art GCN-based models. Therefore, this paper proposes a easy-to-accomplish plug-in unit, which could be loaded into any current GCN-based KGC architectures, and help them maximize the maintenance of native model's original performance while significantly reducing parameter scale. © 2023 IEEE.

关键词： Graph Convolutional Network Knowledge Graph Knowledge Graph Completion Representation Learning

来源：评论

学校读者我要写书评

暂无评论

Robotic Autonomous Grasping Technique: A Survey 5

Robotic Autonomous Grasping Technique: A Survey

引用

5th Asian Conference on Artificial Intelligence Technology, ACAIT 2021

作者： Wang, Lili Zhang, Zhen Su, Jianhua Gu, Qipeng School of Artificial Intelligence University of Chinese Academy of Sciences Beijing China State Key Lab. of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences Beijing China

ISBN: (纸本)9781665426305

This paper provides a comprehensive survey of robotic autonomous grasping techniques. We summarize three key tasks: grasp detection, affordance detection, and model migration. Grasp detection determines the graspable area and grasping posture of the manipulator, so that the robot can successfully perform the grasps. The grasp detection methods based on deep learning are divided into 3DoF grasp and 6DoF grasp. The object affordances based grasping methods can further improve the robot's understanding of objects and environment, thereby improving the robot's intelligence and autonomy. Methods for object affordances detection are classified as learning-based, knowledge-based, and simulation-based. Model migration means that when the grasping model is migrated to other scenes where lightness and background changes, only little or no label data is required, so that the grasping model can be used in the target scene quickly and efficiently. This paper focuses on domain adaptation (DA) methods in model migration. © 2021 IEEE.

关键词： Surveys

来源：评论

学校读者我要写书评

暂无评论

Parallel Learning Based Foundation Model for Networked Traffic Signal control

Parallel Learning Based Foundation Model for Networked Traff...

引用

International Conference on Intelligent Transportation

作者： Chen Zhao Xingyuan Dai Yuanyuan Chen Yilun Lin Yisheng Lv Fei-Yue Wang School of Artificial Intelligence University of Chinese Academy of Sciences Beijing China State Key Laboratory for Management and Control of Complex Systems Institute of Automation Chinese Academy of Sciences Beijing China Shanghai AI Laboratory Shanghai China Macao Institute of Systems Engineering Macau University of Science and Technology Macao China

Networked Traffic Signal control (NTSC) is a fundamental component of Intelligent Transportation systems (ITS) and the broader vision of smart city development. While a plethora of intelligent strategies have been developed, the Sim2Real challenge often impedes their full realization. In response, this paper introduces the Parallel Learning-based Adaptive Network for Traffic Signal control (PLANT) as a foundation model for NTSC. We employ the Wasserstein GAN with Gradient Penalty (WGAN-GP) to generate a wide range of artificial scenarios for robust PLANT training. Further, the Transformer-based Cooperation Mechanism (TCM) is integrated as the primary learner within PLANT, facilitating effective capture of traffic dynamics and knowledge accumulation. This knowledge is readily transferable to real-world applications through meticulous fine-tuning, equipping PLANT to adapt and evolve in alignment with shifting transportation paradigms. Our empirical study on the Hangzhou road network demonstrates PLANT's superiority over both traditional and emerging DRL-based approaches, emphasizing its viability as a potential foundation model for NTSC.

关键词：

来源：评论

学校读者我要写书评

暂无评论

A Novel Iterative Adaptive Critic Design for Smart Home Energy systems With Solar Energy

A Novel Iterative Adaptive Critic Design for Smart Home Ener...

引用

Chinese automation Congress (CAC)

作者： Zehua Liao Qinglai Wei Hongyang Li School of Artificial Intelligence University of Chinese Academy of Sciences The State Key Laboratory of Management and Control for Complex Systems Beijing China

As a distributed energy storage system, smart home energy (SHE) system can be used to reduce the consumption cost of household users. Aiming at the optimal control SHE systems, a novel multi-iteration adaptive dynamic programming (MIADP) method is constructed. The present algorithm introduces three iterations, namely, exterior iteration, local iteration and interior iteration, to ensure that the iterative Q-function converges to the optimal, and thus the optimal energy storage control scheme of the SHE system with solar energy is obtained. Finally, the present MIADP algorithm is illustrated by the experimental results.

关键词： Renewable energy sources Costs Adaptive systems Heuristic algorithms Optimal control Solar energy Smart homes

来源：评论

学校读者我要写书评

暂无评论

Temporal-difference spatial sampling and aggregating graph neural network for crowd flow forecasting 1

Temporal-difference spatial sampling and aggregating graph n...

引用

1st IEEE International Conference on Digital Twins and Parallel Intelligence, DTPI 2021

作者： Bu, Xiaojie Wei, Zebing Li, Zhishuai Wang, Xiao Chen, Yuanyuan Song, Yunquan Lv, Yisheng College of Science China University of Petroleum Qingdao China Institute of Automation Chinese Academy of Sciences Beijing China School of Artificial Intelligence University of Chinese Academy of Sciences Institute of Automation Chinese Academy of Sciences Beijing China Institute of Automation Chinese Academy of Sciences State Key Laboratory of Management and Control for Complex Systems Beijing China Qingdao Academy of Intelligent Industries Qingdao China College of Science China University of Petroleum Qingdao China Institute of Automation Chinese Academy of Sciences University of Chinese Academy of Sciences State Key Laboratory of Management and Control for Complex Systems Beijing China

ISBN: (纸本)9781665433372

With the development of traffic demand management, crowd flow forecasting arouse increasing interest. In order to tackle this spatial-temporal prediction problem, GNN (Graph Neural Networks) have been recently employed to model spatial dependencies, and usually spatial graphs are constructed based on geodetic distances or direct connections between nodes. However, it might be insufficient for spatial graphs to model real dependencies because they neglect spatial-temporal correlations. In this paper, a novel Temporal-Difference Spatial Sampling and Aggregating graph neural network (TDSSA) is proposed to model spatial-temporal dependencies. Firstly, a new sub-sparse spatial-temporal graph is constructed to represent spatial-temporal relationships among different nodes, then a TDSSA block is designed to extract features by spatial sampling and aggregating, and difference is utilized in TDSSA blocks to exploit temporal trend and increase robustness. Experiments show that the proposed method outperforms baseline methods. © 2021 IEEE.

关键词： Graph neural networks

来源：评论

学校读者我要写书评

暂无评论

Learning from Crowd Labeling with Semi-crowdsourced Deep Generative Models 15th

Learning from Crowd Labeling with Semi-crowdsourced Deep Gen...

引用

15th CCF Conference on Computer Supported Cooperative Work and Social Computing, Chinese CSCW 2020

作者： Wei, Xuan Zhang, Mingyue Zeng, Daniel Dajun Antai College of Economics and Management Shanghai Jiao Tong University Shanghai China School of Business and Management Shanghai International Studies University Shanghai China State Key Laboratory of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences Beijing China Shenzhen China

ISBN: (纸本)9789811625398

Microtask crowdsourcing has become an appealing approach to collecting large-scale high-quality labeled data across a wide range of domains. As the crowd workers may be unreliable, the most fundamental question is how to aggregate the noisy annotations provided by these potentially unreliable workers. Although various factors such as worker reliability and task features are considered in the literature, they are not meaningfully combined in a unified and consistent framework. In this work, we propose a semi-crowdsourced deep generative approach called S-DARFC which combines Bayesian graphical models and deep learning techniques and unifies factors including the worker reliability, task features, task clustering structure, and semi-crowdsourcing. Graphical models are good at finding a structure that is interpretable and generalizes to new tasks easily and deep learning techniques are able to learn a flexible representation of complex high-dimensional unstructured data (e.g., task features). Extensive experiments based on six real-world tasks including text and image classification demonstrate the effectiveness of our proposed approach. © 2021, Springer Nature Singapore Pte Ltd.

关键词： Crowdsourcing

来源：评论

学校读者我要写书评

暂无评论

Learning to Manipulate Tools Using Deep Reinforcement Learning and Anchor Information

Learning to Manipulate Tools Using Deep Reinforcement Learni...

引用

2022 IEEE International Conference on Robotics and Biomimetics, ROBIO 2022

作者： Wei, Junhang Cui, Shaowei Hao, Peng Wang, Shuo The School of Future Technology University of Chinese Academy of Sciences Beijing100049 China The State Key Laboratory of Management and Control for Complex Systems Institute of Automation Chinese Academy of Sciences Beijing100190 China The University of Chinese Academy of Sciences Beijing100049 China The Center for Excellence in Brain Science and Intelligence Technology Chinese Academy of Sciences Shanghai200031 China

ISBN: (纸本)9781665481090

Endowing robots with tool manipulation skills helps them accomplish challenging tasks. While robots manipulate tools to achieve goals, the alignment of tools and targets is a noise-sensitive and contact-rich task. However, it is difficult to access the accurate pose of the tool and the target. When there is unknown noise in the observations, reinforcement learning can't be sure to perform well. In this paper, we define the easier-to-obtain accurate task-related information as anchor information and introduce a tool manipulation method based on reinforcement learning and anchor information, which can perform well when the observations include unknown noise. To evaluate the method, we build a simulated environment ToolGym, which includes four different kinds of tools and different noise sampling functions for each tool. Finally, we compare our method with baseline methods to show the effectiveness of the proposed method. © 2022 IEEE.

关键词： Reinforcement 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：