检索结果-内蒙古大学图书馆

arXiv 2022年

作者： Ding, Xiaohan Chen, Honghao Zhang, Xiangyu Huang, Kaiqi Han, Jungong Ding, Guiguang Tencent AI Lab United States CRISE Institute of Automation Chinese Academy of Sciences China School of Artificial Intelligence University of Chinese Academy of Sciences China MEGVII Technology China Beijing Academy of Artificial Intelligence China Department of Computer Science The University of Sheffield United Kingdom School of Software Tsinghua University China

The well-designed structures in neural networks reflect the prior knowledge incorporated into the models. However, though different models have various priors, we are used to training them with model-agnostic optimizers such as SGD. In this paper, we propose to incorporate model-specific prior knowledge into optimizers by modifying the gradients according to a set of model-specific hyper-parameters. Such a methodology is referred to as Gradient Re-parameterization, and the optimizers are named RepOptimizers. For the extreme simplicity of model structure, we focus on a VGG-style plain model and showcase that such a simple model trained with a RepOptimizer, which is referred to as RepOpt-VGG, performs on par with or better than the recent well-designed models. From a practical perspective, RepOpt-VGG is a favorable base model because of its simple structure, high inference speed and training efficiency. Compared to Structural Re-parameterization, which adds priors into models via constructing extra training-time structures, RepOptimizers require no extra forward/backward computations and solve the problem of quantization. We hope to spark further research beyond the realms of model structure design. Code and models https://***/DingXiaoH/RepOptimizers. Copyright © 2022, The Authors. All rights reserved.

关键词： Machine learning

来源：评论

学校读者我要写书评

暂无评论

Optimal Walking Assistance Control of Lower Limb Exoskeleton Using Adaptive Learning Approach 39

Optimal Walking Assistance Control of Lower Limb Exoskeleton...

引用

39th Chinese Control Conference, CCC 2020

作者： Peng, Zhinan Luo, Rui Hu, Jiangping Kumar Ghosh, Bijoy Kiong Nguang, Sing University of Electronic Science and Technology of China School of Automation Engineering Chengdu611731 China Texas Tech University Department of Mathematics and Statistics LubbockTX79409-1042 United States University of Auckland Department of Electrical Computer and Software Engineering Auckland1010 New Zealand

ISBN: (纸本)9789881563903

In this paper, an adaptive reinforcement learning (RL) based controller is developed to solve assistance control problem for Lower Limb Exoskeleton (LLE) to aid hemiplegic individuals in walking. The communication interaction relation between both two lower-limbs and patient's unaffected leg is modelled in the context of leader-follower (LF) framework. The walking assistance control problem of LLE with patients is converted to optimal control problem. To handle the optimal control problem, a discounted cost function is designed in terms of the local tracking error, and then a policy iteration algorithm (PI) is proposed to generate an optimal control policy, followed by the convergence analysis of the presented algorithm. Further, in order to improve the adaption of the controller to different patients, on the basis of the PI algorithm, an actor-critic-based neural network (AC/NN) architecture is introduced to implement the presented control method in an online-learning fashion. Finally, simulation scenarios are established to test the effectiveness of the proposed walking assistance control approaches. © 2020 Technical Committee on Control Theory, Chinese Association of Automation.

关键词： Reinforcement learning

来源：评论

学校读者我要写书评

暂无评论

An Integrated Model to Evaluate the Transparency in Predicting Chronic Kidney Disease Using a Trio-Embedded Explainable Model

SSRN

引用

SSRN 2022年

作者： Gaur, Loveleen Biswas, Milon Bakshi, Sumedha Gupta, Pranshi Si, Tapas Mallik, Saurav Maulik, Ujjwal Amity International Business School Amity University India Bangladesh University of Business and Technology Bangladesh Ernst & Young Prospecta Software India Bankura Unnayani Institute of Engineering West Bengal Bankura India The University of Texas Health Science Center Tyler United States Department of Computer Science and Engineering Jadavpur University India

Why do clinicians doubt the applicability of artificial intelligence in the medical field? One of the primary challenges to implementation is a lack of transparency, as clinicians must be confident that the AI system can be trusted. Explainable AI has the ability to solve this problem and is a step toward AI that is trustworthy. Explainability improves learning for educational and research objectives by allowing comparisons of learned procedures with existing knowledge. This research proposes an explainable artificial intelligence (XAI) archetype utilizing shapely additive values (SHAP), local interpretable model-agnostic explanation (LIME), and DALEX to provide a qualitative knowledge of the association between the model and the parameters utilized in the model. Firstly five machine learning models (Logistic Regression, Random Forest, Gradient Boosting Machine, Support Vector Machine, and Multilayer Perceptron Classifier) are being interpreted with local models called LIME, SHAP, and DALEX and then a global model called EBM. This research also compares the performance of five classifiers with Explainable Boosting Machine (EBM), while EBM achieved the highest accuracy of 98.57%. Lastly, we use the XAI model to determine the most important factors that influence those predictions. © 2022, The Authors. All rights reserved.

关键词： Transparency

来源：评论

学校读者我要写书评

暂无评论

Multi-Resolution Localization of Individual Logs in Wooden Piles Utilizing YOLO with Tiling on Client/Server Architectures 33

Multi-Resolution Localization of Individual Logs in Wooden P...

引用

33rd European Modeling and Simulation Symposium, EMSS 2021

作者： Praschl, Christoph Auersperg-Castell, Philipp Forster-Heinlein, Brigitte Zwettler, Gerald Adam Research Group Advanced Information Systems and Technology Research and Development Department University of Applied Sciences Upper Austria Softwarepark 11 Hagenberg4232 Austria Department of Software Engineering School of Informatics Communications and Media University of Applied Sciences Upper Austria Softwarepark 11 Hagenberg4232 Austria Bluedynamics Auersperg-Castell KG Kritzing 31 Freinberg4785 Austria Faculty of Computer Science and Mathematics University of Passau Innstraße 43 Passau94032 Germany

In industrial domains with time and cost intensive manual or semi-automated inspection the demand for automation is high. Utilizing state of the art deep learning models for localization in vision-based domains such as wood log analysis, the precision can be increased while reducing the demand for manual inspection. In this paper a YOLO network is trained on wood log images to allow for detection of single wood piles in images with hundreds and thousands instances. Due to the high variability in scale and large amount of wood logs within the images, common YOLO architectures are not applicable. Thus, tiling is necessitated to implicitly form a multi-resolution image pyramid. Due to lack in training data, besides common data augmentation modelling of different seasonal and weather conditions is applied. The wood log detection process can be run on a client/server architecture to allow for both, preview and refined results. Evaluation on real-world data sets shows an log detection accuracy of 82,9% utilizing a tiny YOLO model and 94,1% with a fully connected YOLO model, respectively. © 2021 The Authors.

关键词： Deep learning

来源：评论

学校读者我要写书评

暂无评论

PPGPass: Nonintrusive and Secure Mobile Two-Factor Authentication via Wearables

PPGPass: Nonintrusive and Secure Mobile Two-Factor Authentic...

引用

IEEE Annual Joint Conference: INFOCOM, IEEE computer and Communications Societies

作者： Yetong Cao Qian Zhang Fan Li Song Yang Yu Wang School of Computer Science and Technology Beijing Institute of Technology China School of Software and BNRist Tsinghua University China Department of Computer and Information Sciences Temple University USA

ISBN: (数字)9781728164120

ISBN: (纸本)9781728164137

Mobile devices are promising to apply two-factor authentication in order to improve system security and enhance user privacy-preserving. Existing solutions usually have certain limits of requiring some form of user effort, which might seriously affect user experience and delay authentication time. In this paper, we propose PPGPass, a novel mobile two-factor authentication system, which leverages Photoplethysmography (PPG) sensors in wrist-worn wearables to extract individual characteristics of PPG signals. In order to realize both nonintrusive and secure, we design a two-stage algorithm to separate clean heartbeat signals from PPG signals contaminated by motion artifacts, which allows verifying users without intentionally staying still during the process of authentication. In addition, to deal with non-cancelable issues when biometrics are compromised, we design a repeatable and non-invertible method to generate cancelable feature templates as alternative credentials, which enables to defense against man-in-the-middle attacks and replay attacks. To the best of our knowledge, PPGPass is the first nonintrusive and secure mobile two-factor authentication based on PPG sensors in wearables. We build a prototype of PPGPass and conduct the system with comprehensive experiments involving multiple participants. PPGPass can achieve an average F1 score of 95.3%, which confirms its high effectiveness, security, and usability.

关键词： Authentication Feature extraction Biometrics (access control) Heart beat Wearable sensors Writing

来源：评论

学校读者我要写书评

暂无评论

Decomposition method-based global Mittag-Leffler synchronization for fractional-order Clifford-valued neural networks with transmission delays and impulses

引用

Neural Networks 2025年 189卷 107564页

作者： Sriraman, R. Manoj, N. Pratap, Anbalagan Department of Mathematics College of Engineering and Technology SRM Institute of Science and Technology Tamil Nadu Kattankulathur603203 India College of Mechatronics and Control Engineering Shenzhen University Shenzhen518060 China College of Computer Science and Software Engineering Shenzhen University Shenzhen518060 China

This study examines the global Mittag-Leffler synchronization (GMLS) problem for fractional-order Clifford-valued neural networks (FOCLVNNs) including transmission delays and impulses. Firstly, a novel kind of FOCLVNNs is developed that incorporates impulses and transmission delays, which can be useful for better understanding of Clifford-valued neural networks (CLVNNs) dynamics more precisely. Then, because of the non-commutative multiplication feature of the Clifford algebra, the systems of FOCLVNNs can be decomposed into multi-dimensions real-valued neural networks (RVNNs), that avoid the difficulties associated with Clifford number multiplication. By using Lyapunov functions and inequality techniques, several novel solid conditions for the GMLS of FOCLVNNs under the state feedback controller that has been formulated. In addition, this paper presents novel results that are easy to solve and new insights into GMLS of FOCLVNNs. Then, a numerical verification is given along with graphical analysis to validate the applicability of the obtained theories. Based on the considered master–slave FOCLVNNs, an image encryption algorithm for color images is provided. In addition, simulation and performance analysis are stated to prove the validity of the given algorithms. © 2025 Elsevier Ltd

关键词： Lyapunov functions

来源：评论

学校读者我要写书评

暂无评论

Automated model design and benchmarking of 3D deep learning models for COVID-19 detection with chest CT scans

arXiv

引用

arXiv 2021年

作者： He, Xin Wang, Shihao Chu, Xiaowen Shi, Shaohuai Tang, Jiangping Liu, Xin Yan, Chenggang Zhang, Jiyong Ding, Guiguang Department of Computer Science Hong Kong Baptist University Hong Kong Hong Kong Department of Computer Science and Engineering Hong Kong University of Science and Technology Hong Kong Hong Kong School of Automation Hangzhou Dianzi University Hang Zhou China School of Software Tsinghua University Beijing China

The COVID-19 pandemic has spread globally for several months. Because its transmissibility and high pathogenicity seriously threaten people’s lives, it is crucial to accurately and quickly detect COVID-19 infection. Many recent studies have shown that deep learning (DL) based solutions can help detect COVID-19 based on chest CT scans. However, most existing work focuses on 2D datasets, which may result in low quality models as the real CT scans are 3D images. Besides, the reported results span a broad spectrum on different datasets with a relatively unfair comparison. In this paper, we first use three state-of-the-art 3D models (ResNet3D101, DenseNet3D121, and MC3 18) to establish the baseline performance on the three publicly available chest CT scan datasets. Then we propose a differentiable neural architecture search (DNAS) framework to automatically search for the 3D DL models for 3D chest CT scans classification with the Gumbel Softmax technique to improve the searching efficiency. We further exploit the Class Activation Mapping (CAM) technique on our models to provide the interpretability of the results. The experimental results show that our automatically searched models (CovidNet3D) outperform the baseline human-designed models on the three datasets with tens of times smaller model size and higher accuracy. Furthermore, the results also verify that CAM can be well applied in CovidNet3D for COVID-19 datasets to provide interpretability for medical diagnosis. Code: https://***/HKBU-HPML/CovidNet3D. © 2021, CC BY.

关键词： Diagnosis

来源：评论

学校读者我要写书评

暂无评论

Privacy-Aware Key Task Scheduling in Vehicular Networks Based on Incentive Mechanism 22

Privacy-Aware Key Task Scheduling in Vehicular Networks Base...

引用

22nd IEEE International Conference on High Performance Computing and Communications, 18th IEEE International Conference on Smart City and 6th IEEE International Conference on Data Science and Systems, HPCC-SmartCity-DSS 2020

作者： Xia, Youhua Wu, Libing Jin, Jiong Zhang, Tiehua Zheng, Xi Wuhan University School of Computer Science Wuhan China School of Cyber Science and Engineering Wuhan University Wuhan China Swinburne University of Technology School of Software and Electrical Engineering Melbourne Australia Macquarie University Department of Computing Sydney Australia

ISBN: (纸本)9781728176499

The introduction of a lightweight encryption scheme in the vehicular network has improved the reliability and security of data transmission substantially among vehicles. However, it is still not guaranteed that all vehicles comply with the encryption scheme throughout the whole operation period, considering the inherent selfishness of each participant. Apart from that, the data scheduling issue is another major concern in the vehicular network due to the high-speed mobility of vehicles. To tackle these issues, a multi-objective and multi-dimensional incentive mechanism is developed in this paper to achieve privacy-Aware data scheduling for vehicles. This mechanism is designed to encourage vehicles to carry out different tasks by providing relevant incentives while maximizing the overall utility of the network. Additionally, security and privacy of data transmission between vehicles and cloud servers are realized through a data perturbation approach. Experimental evaluation shows that the proposed incentive mechanism is better than the traditional methods when it comes to maximizing the number of participants and completed key tasks. © 2020 IEEE.

关键词： Cryptography

来源：评论

学校读者我要写书评

暂无评论

A Hybrid Cross-Scale Transformer Architecture for Robust Medical Image Segmentation

A Hybrid Cross-Scale Transformer Architecture for Robust Med...

引用

IEEE International Conference on Bioinformatics and Biomedicine (BIBM)

作者： Yang Ning Shouyi Zhang Wei Zhong Peide Liu Caiming Zhang School of Computer Science and Technology Shandong Jianzhu University Jinan China School of Informatics The University of Edinburgh Edinburgh UK Department of Network Information Security Guangdong Police College Guangzhou China School of Management Science and Engineering Shandong University of Finance and Economics Jinan China School of Software Shandong University Jinan China

ISBN: (数字)9781665468190

ISBN: (纸本)9781665468206

Transformer architecture has emerged to be successful in many natural language processing tasks. However, its applications to clinical practice remain largely unexplored. In this study, we propose a Robust Cross-Scale Hybrid Transformer (RCSHT) architecture for medical image segmentation, which can effectively enhance the multi-scale feature representations while integrating local features with global dependencies. Specifically, we propose two new modules based on the self-attention mechanism: PSCM and PCCM, which perform matrix low-rank transformation on cross-scale images in spatial and channel space, respectively, to effectively enhance the discriminant ability of long-range dependencies of multi-scale images and effectively reduce the computational complexity. Meanwhile, we use spatial positional encoding to perform spatial regularization on the two attention modules, so that the model has more opportunities to obtain valuable spatial details and further enhance the spatial retention ability. Besides, by fusing outputs of the two modules, the robustness and discriminant ability of the feature representations are further improved. Note that our hybrid architecture allows transformers to be initialized as convolutional networks without pre-training. Extensive experiments conducted on our newly proposed dataset have demonstrated our RCSHT notably outperforms the state-of-the-art methods by a large margin, holding the promise to generalize well on other medical image segmentation.

关键词： Measurement Image segmentation Image coding Computed tomography computer architecture Transformers Robustness

来源：评论

学校读者我要写书评

暂无评论

Spatio-temporal graph neural network based child action recognition using data-efficient methods: A systematic analysis

引用

computer Vision and Image Understanding 2025年

作者： Sanka Mohottala Asiri Gawesha Dhrashana Karshutrirathna Pradeepa Samarasinghe Charith Abhayaratne Department of Software Engineering Faculty of Computing Sri Lanka Institute of Information Technology Sri Lanka Department of Electrical and Electronic Engineering Faculty of Engineering Sri Lanka Institute of Information Technology Sri Lanka Department of Computer Science Faculty of Computing Sri Lanka Institute of Information Technology Sri Lanka Department of Information Technology Faculty of Computing Sri Lanka Institute of Information Technology Sri Lanka School of Electrical and Electronic Engineering The University of Sheffield United Kingdom

This paper presents implementations on child activity recognition (CAR) using spatial–temporal graph neural network (ST-GNN)-based deep learning models with the skeleton modality. Prior implementations in this domain have predominantly utilized CNN, LSTM, and other methods, despite the superior performance potential of graph neural networks. To the best of our knowledge, this study is the first to use an ST-GNN model for child activity recognition employing both in-the-lab, in-the-wild, and in-the-deployment skeleton data. To overcome the challenges posed by small publicly available child action datasets, transfer learning methods such as feature extraction and fine-tuning were applied to enhance model performance. As a principal contribution, we developed an ST-GNN-based skeleton modality model that, despite using a relatively small child action dataset, achieved superior performance (94.81%) compared to implementations trained on a significantly larger (x10) adult action dataset (90.6%) for a similar subset of actions. With ST-GCN-based feature extraction and fine-tuning methods, accuracy improved by 10%–40% compared to vanilla implementations, achieving a maximum accuracy of 94.81%. Additionally, implementations with other ST-GNN models demonstrated further accuracy improvements of 15%–45% over the ST-GCN baseline. The results on activity datasets empirically demonstrate that class diversity, dataset size, and careful selection of pre-training datasets significantly enhance accuracy. In-the-wild and in-the-deployment implementations confirm the real-world applicability of above approaches, with the ST-GNN model achieving 11 FPS on streaming data. Finally, preliminary evidence on the impact of graph expressivity and graph rewiring on accuracy of small dataset-based models is provided, outlining potential directions for future research.

关键词：

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：