检索结果-内蒙古大学图书馆

A Stronger Baseline for Ego-Centric Action Detection

学校读者我要写书评

暂无评论

arXiv 2021年

作者： Qing, Zhiwu Huang, Ziyuan Wang, Xiang Feng, Yutong Zhang, Shiwei Jiang, Jianwen Tang, Mingqian Gao, Changxin Ang, Marcelo H. Sang, Nong Key Laboratory of Image Processing and Intelligent Control School of Artificial Intelligence and Automation Huazhong University of Science and Technology ARC National University of Singapore Tsinghua University Alibaba Group

This technical report analyzes an egocentric video action detection method we used in the 2021 EPIC-KITCHENS-100 competition hosted in CVPR2021 Workshop. The goal of our task is to locate the start time and the end time of the action in the long untrimmed video, and predict action category. We adopt sliding window strategy to generate proposals, which can better adapt to short-duration actions. In addition, we show that classification and proposals are conflict in the same network. The separation of the two tasks boost the detection performance with high efficiency. By simply employing these strategy, we achieved 16.10% performance on the test set of EPIC-KITCHENS-100 Action Detection challenge using a single model, surpassing the baseline method by 11.7% in terms of average mAP. Copyright © 2021, The Authors. All rights reserved.

关键词： Machine learning

Global Update Guided Federated Learning

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Wu, Qilong Liu, Lin Xue, Shibei Department of Automation Shanghai Jiao Tong University Shanghai200240 China Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai200240 China Shanghai Engineering Research Center of Intelligent Control and Management Shanghai200240 China School of Electronic Information Engineering Shanghai Dianji University Shanghai201306 China

Federated learning protects data privacy and security by exchanging models instead of data. However, unbalanced data distributions among participating clients compromise the accuracy and convergence speed of federated learning algorithms. To alleviate this problem, unlike previous studies that limit the distance of updates for local models, we propose global-update-guided federated learning (FedGG), which introduces a model-cosine loss into local objective functions, so that local models can fit local data distributions under the guidance of update directions of global models. Furthermore, considering that the update direction of a global model is informative in the early stage of training, we propose adaptive loss weights based on the update distances of local models. Numerical simulations show that, compared with other advanced algorithms, FedGG has a significant improvement on model convergence accuracies and speeds. Additionally, compared with traditional fixed loss weights, adaptive loss weights enable our algorithm to be more stable and easier to implement in practice. Copyright © 2022, The Authors. All rights reserved.

关键词： Data privacy

Recent Few-Shot Object Detection Algorithms: A Survey with Performance Comparison

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Liu, Tianying Zhang, Lu Wang, Yang Guan, Jihong Fu, Yanwei Zhao, Jiajia Zhou, Shuigeng Department of Computer Science and Technology Tongji University Shanghai201804 China The Project Management Office of China National Scientific Seafloor Observatory Tongji University Shanghai200092 China Shanghai Key Lab of Intelligent Information Processing The School of Computer Science Fudan University Shanghai200438 China Shanghai Key Lab of Intelligent Information Processing The School of Data Science Fudan University Shanghai200433 China Science and Technology on Complex System Control and Intelligent Agent Cooperation Laboratory Beijing100074 China

The generic object detection (GOD) task has been successfully tackled by recent deep neural networks, trained by an avalanche of annotated training samples from some common classes. However, it is still non-trivial to generalize these object detectors to the novel long-tailed object classes, which have only few labeled training samples. To this end, the Few-Shot Object Detection (FSOD) has been topical recently, as it mimics the humans' ability of learning to learn, and intelligently transfers the learned generic object knowledge from the common heavy-tailed, to the novel long-tailed object classes. Especially, the research in this emerging field has been flourishing in recent years with various benchmarks, backbones, and methodologies proposed. To review these FSOD works, there are several insightful FSOD survey articles [58, 59, 74, 78] that systematically study and compare them as the groups of fine-tuning/transfer learning, and meta-learning methods. In contrast, we review the existing FSOD algorithms from a new perspective under a new taxonomy based on their contributions, i.e., data-oriented, model-oriented, and algorithm-oriented. Thus, a comprehensive survey with performance comparison is conducted on recent achievements of FSOD. Furthermore, we also analyze the technical challenges, the merits and demerits of these methods, and envision the future directions of FSOD. Specifically, we give an overview of FSOD, including the problem definition, common datasets, and evaluation protocols. The taxonomy is then proposed that groups FSOD methods into three types. Following this taxonomy, we provide a systematic review of the advances in FSOD. Finally, further discussions on performance, challenges, and future directions are presented. Copyright © 2022, The Authors. All rights reserved.

关键词： Object detection

Scheduling optimization of silicon single crystal production process based on adaptive improved particle swarm optimization

学校读者我要写书评

暂无评论

Scheduling optimization of silicon single crystal production...

Chinese Automation Congress (CAC)

作者： Lu Kang Ding Liu Yali Wu Guozheng Ping School of Automation and Information Engineering Xi’an University of Technology National and Local Joint Engineering Research Center of Crystal Growth Equipment and System Integration Shaanxi Key Laboratory of Complex System Control and Intelligent Information Processing Xi’an China School of Automation and Information Engineering Xi’an University of Technology National and Local Joint Engineering Research Center of Crystal Growth Xi’an University of Technology Shaanxi Key Laboratory of Complex System Control and Intelligent Information Processing Xi’an China School of Automation and Information Engineering Xi’an University of Technology Shaanxi Key Laboratory of Complex System Control and Intelligent Information Processing Xi’an China

ISBN: (纸本)9781665426480

Aiming at the potential safety hazards of silicon single crystal production enterprises in the process of multi-furnace silicon single crystal production, and considering the maximum power load requirements in actual enterprises, a silicon single crystal production process scheduling model with the goal of minimizing the maximum completion time was established. Based on this model, an adaptive improved particle swarm optimization algorithm (Adaptive GPSO) is proposed to solve it. The algorithm retains the individual optimal value and global optimal value of the particle swarm algorithm, and introduces the crossover operation and mutation operation in the genetic algorithm to improve the new individual generation mechanism. At the same time, it performs dynamic iterative calculations on the mutation probability and crossover probability to avoid the algorithm fall into the local optimum, and improve the diversity and convergence of the algorithm. By solving different scales scheduling problems of silicon single crystal production process, the experiments on real-world data shows that the feasibility and effectiveness of the proposed algorithm.

关键词： Adaptation models Heuristic algorithms Crystals Production Probability Silicon Scheduling

Fuzzy Sandwich control of Some Nonlinear Systems with a Number of Stochastically Occurring Impulses

学校读者我要写书评

暂无评论

Fuzzy Sandwich Control of Some Nonlinear Systems with a Numb...

International Conference on information Science and Technology (ICIST)

作者： Babatunde Oluwaseun Onasanya Yuming Feng Xiangguang Dai Wei Zhang School of Computer Science and Engineering Chongqing Three Gorges University Wanzhou Chongqing China Department of Mathematics University of Ibadan Ibadan Nigeria Three Gorges Institute of Artificial Intelligence Chongqing Three Gorges University Wanzhou Chongqing China Chongqing Engineering Research Center of Internet of Things and Intelligent Control Technology Chongqing Three Gorges University Wanzhou Chongqing China Key Laboratory of Intelligent Information Processing and Control Chongqing Three Gorges University Wanzhou Chongqing China

In this paper, rather than have a fixed constant matrix in the sandwich control of the nonlinear system, a fuzzy control matrix was used. In addition, simple inequalities that do not require additional computational time and skills in computer programming were used rather than the usual LMI. The result of this method is that the period T required in the control process is shorter and the control time is also reached much faster or reduced compared to the existing sandwich methods that used constant matrices. In other words, the control time of this method is shorter than the control time achieved when using constant matrices. Besides, this method also can be used to control more nonlinear chaotic systems such as Lorenz's, Rossler's and Chen's systems. Numerical simulations were used to demonstrate the effectiveness of this method.

关键词：

Differential Private Discrete Noise Adding Mechanism: Conditions, Properties and Optimization

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Qin, Shuying He, Jianping Fang, Chongrong Lam, James The Dept. of Automation Shanghai Jiao Tong University The Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai Engineering Research Center of Intelligent Control and Management Shanghai China The Dept. of Mechanical Engineering The University of Hong Kong Pokfulam Road Hong Kong

Differential privacy is a standard framework to quantify the privacy loss in the data anonymization process. To preserve differential privacy, a random noise adding mechanism is widely adopted, where the trade-off between data privacy level and data utility is of great concern. The privacy and utility properties for the continuous noise adding mechanism have been well studied. However, the related works are insufficient for the discrete random mechanism on discretely distributed data, e.g., traffic data, health records. This paper focuses on the discrete random noise adding mechanisms. We study the basic differential privacy conditions and properties for the general discrete random mechanisms, as well as the trade-off between data privacy and data utility. Specifically, we derive a sufficient and necessary condition for discrete ∊-differential privacy and a sufficient condition for discrete (∊, δ)-differential privacy, with the numerical estimation of differential privacy parameters. These conditions can be applied to analyze the differential privacy properties for the discrete noise adding mechanisms with various kinds of noises. Then, with the differential privacy guarantees, we propose an optimal discrete ∊-differential private noise adding mechanism under the utility-maximization framework, where the utility is characterized by the similarity of the statistical properties between the mechanism's input and output. For this setup, we find that the class of the discrete noise probability distributions in the optimal mechanism is Staircase-shaped. Copyright © 2022, The Authors. All rights reserved.

关键词： Probability distributions

Differentially Private Distributed Mismatch Tracking Algorithm for Constraint-Coupled Resource Allocation Problems

学校读者我要写书评

暂无评论

arXiv 2022年

作者： Wu, Wenwen Zhu, Shanying Liu, Shuai Guan, Xinping Department of Automation Shanghai Jiao Tong University Shanghai200240 China Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai200240 China Shanghai Engineering Research Center of Intelligent Control and Management Shanghai200240 China School of Control Science and Engineering Shandong University Jinan China

This paper considers privacy-concerned distributed constraint-coupled resource allocation problems over an undirected network, where each agent holds a private cost function and obtains the solution via only local communication. With privacy concerns, we mask the exchanged information with independent Laplace noise against a potential attacker with potential access to all network communications. We propose a differentially private distributed mismatch tracking algorithm (diff-DMAC) to achieve cost-optimal distribution of resources while preserving privacy. Adopting constant stepsizes, the linear convergence property of diff-DMAC in mean square is established under the standard assumptions of Lipschitz gradients and strong convexity. Moreover, it is theoretically proven that the proposed algorithm is t-differentially private. And we also show the trade-off between convergence accuracy and privacy level. Finally, a numerical example is provided for verification. Copyright © 2022, The Authors. All rights reserved.

关键词： Cost functions

Attention Residual Network for White Blood Cell Classification with WGAN Data Augmentation

学校读者我要写书评

暂无评论

Attention Residual Network for White Blood Cell Classificati...

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

作者： Meng Zhao Lingmin Jin Shenghua Teng Zuoyong Li College of Electronic and Information Engineering Shandong University of Science and Technology Qingdao China Fujian Provincial Key Laboratory of Information Processing and Intelligent Control College of Computer and Control Engineering Minjiang University Fuzhou China

ISBN: (纸本)9781728194202

In medicine, white blood cell (WBC) classification plays an important role in clinical diagnosis and treatment. Due to the similarity between classes and lack of training data, the precise classification of WBC is still challenging. To alleviate this problem, we propose an attention residual network for WBC image classification on the basis of data augmentation. Specifically, the attention residual network is composed of multiple attention residual blocks, an adaptive average pooling layer, and a full connection layer. The channel attention mechanism is introduced in each residual block to use the feature maps of WBC learned by a high layer to generate the attention map for a low layer. Each attention residual block also introduces depth separable convolution to extract the feature of WBC and decrease the training costs. The Wasserstein Generative adversarial network (WGAN) is used to create synthetic instances to enhance the size of training data. Experiments on two image datasets show the superiority of the proposed method over several state-of-the-art methods.

关键词： Training White blood cells Costs Convolution Training data Feature extraction Generators

Construction of Minecraft Virtual Reality Scene Based on Voxel Point Cloud 8th

学校读者我要写书评

暂无评论

Construction of Minecraft Virtual Reality Scene Based on Vox...

8th International Conference on Cognitive Systems and information processing, ICCSIP 2023

作者： Ji, Nuocheng Zhuang, Hanyang Yang, Ming The University of Michigan – Shanghai Jiao Tong University Joint Institute Shanghai Jiao Tong University Shanghai200240 China Global Institute of Future Technology Shanghai Jiao Tong University Shanghai200240 China Department of Automation Shanghai Jiao Tong University Shanghai200240 China Key Laboratory of System Control and Information Processing Ministry of Education of China Shanghai200240 China Shanghai Engineering Research Center of Intelligent Control and Management Shanghai200240 China

ISBN: (纸本)9789819980208

Virtual Reality is an important development direction in recent years. Minecraft, a popular sandbox game, has a great potential to become a virtual reality platform for building immersive environments. Traditional methods of building virtual reality scenes rely heavily on massive human labors and require high levels of proficiency. To address these limitations, we propose an automatic, voxelated 3D modeling method with foreground and background environment segmented that can reduce modeling costs and achieve effective application in Minecraft virtual reality scenes such as virtual campuses. The proposed method leverages the use of point cloud reconstruction using a voxelization approach. A pipeline of converting point cloud to Minecraft scene is developed. Through preprocessing, segmenting, coloring, and exporting, a Minecraft virtual reality scene can be automatically generated based on voxel point cloud. Due to the noise in point cloud may cause defect during voxelization, a denoising and smoothing method using slice scanning and voxel connectivity segmentation has been implemented. It can generate building structures with clear outlines and uniform colors. And the entire process requires low computational complexity and achieves fast generation. We conducted experiments based on this method in indoor scenes, outdoor building landmarks, and complex underground scenes, and obtained results with clear structures and smooth surfaces, saving a lot of time for subsequent detailed modeling. © 2024, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： 3D modeling