检索结果-内蒙古大学图书馆

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

作者： Shaohui Peng Xing Hu Rui Zhang Ke Tang Jiaming Guo Qi Yi Ruizhi Chen Xishan Zhang Zidong Du Ling Li Qi Guo Yunji Chen SKL of Processors Institute of Computing Technology CAS and University of Chinese Academy of Sciences and Cambricon Technologies SKL of Processors Institute of Computing Technology CAS SKL of Processors Institute of Computing Technology CAS and Cambricon Technologies Department of Computer Science and Engineering Southern University of Science and Technology SKL of Processors Institute of Computing Technology CAS and Cambricon Technologies and University of Science and Technology of China University of Chinese Academy of Sciences and SKL of Computer Science Institute of Software CAS SKL of Processors Institute of Computing Technology CAS and University of Chinese Academy of Sciences

ISBN: (纸本)9781713871088

Hierarchical reinforcement learning (HRL) effectively improve agents' exploration efficiency on tasks with sparse reward, with the guide of high-quality hierarchical structures (e.g., subgoals or options). However, how to automatically discover high-quality hierarchical structures is still a great challenge. Previous HRL methods can hardly discover the hierarchical structures in complex environments due to the low exploration efficiency by exploiting the randomness-driven exploration paradigm. To address this issue, we propose CDHRL, a causality-driven hierarchical reinforcement learning framework, leveraging a causality-driven discovery instead of a randomness-driven exploration to effectively build high-quality hierarchical structures in complicated environments. The key insight is that the causalities among environment variables are naturally fit for modeling reachable subgoals and their dependencies and can perfectly guide to build high-quality hierarchical structures. The results in two complex environments, 2D-Minecraft and Eden, show that CDHRL significantly boosts exploration efficiency with the causality-driven paradigm.

关键词：

来源：评论

学校读者我要写书评

暂无评论

Towards Efficient Elastic Parallelism for Deep Learning Processor

Towards Efficient Elastic Parallelism for Deep Learning Proc...

引用

IEEE International Conference on Big Data and Cloud Computing (BdCloud)

作者： Jinyu Cheng Ruyi Qian Qinwen Shi Gaomei Hu Mengjuan Gao Qirun Huo Yuanchao Xu College of Information Engineering Capital Normal University Beijing China SKL of Computer Architecture CAS Institute of Computing Technology Beijing China

Resources underutilization is the most troublesome problem faced by almost all deep learning processors. To address this challenge, researchers have proposed various methods, especially in exploiting the parallelism of deep neural network models. However, some parallelisms have mutual effects, for example, intra-operator parallelism has adverse effects on inter-operator parallelism, resulting in a waste of resources. In order to make full use of chip resources, this paper applies the idea of elastic scheduling to the driver level of deep learning processor to achieve elastic parallelism. That is, if a task consists of multiple independent subtasks, it can be executed as long as the minimum resource requirements of the sub tasks are met. In addition, based on greedy scheduling algorithm, some improved methods are proposed, including subtask awareness, NUMA awareness, and cluster alignment awareness. Our analysis and preliminary evaluation demonstrate that, by reducing intra-operator parallelism, elastic parallelism can help improve chip utilization, and our proposed scheduling optimizations can also further improve the overall system performance.

关键词： Deep learning Program processors Scheduling algorithms System performance Pipelines Neural networks Optimization methods

来源：评论

学校读者我要写书评

暂无评论

Identification and Visualization of Zombie Enterprise Portraits-Mining Short-time Series Features from the Perspective of Image 4

Identification and Visualization of Zombie Enterprise Portra...

引用

4th IEEE International Conference on Information Systems and computer Aided Education, ICISCAE 2021

作者： Huang, Zhidong Guo, Yuxiang Cao, Di Hu, Chenrui Ding, Chenjun School of Computer Science and Technology Zhejiang University of Technology Hangzhou China School of Design and Architecture Zhejiang University of Technology Hangzhou China

ISBN: (纸本)9781665441247

Governance of zombie enterprises is an important means to ensure the healthy, sustained development of the economy. Traditional methods such as identifying zombie enterprises based on expert knowledge suffer from incomplete expert database and increasingly complex economic environment. Thus the proposed data-driven system is implemented to not only identify zombie enterprises, but also visually present the enterprise portraits. In this paper, the three-year data of 50000 enterprise is transformed into N×N×3 image-format-matrix (N×N are the number of features). Afterward, Convolutional Neural Network, namely CNN is applied and result is got in one stage instead of fitting the data of each year and voting. It is also proved that CNN can effectively mine the short-time series features of enterprises by reconstruct the data into image-format-matrix. Considering the imbalance of data, Focal-Loss is implemented as the loss function when applying CNN model to the data. Grad-CAM, a model interpretive method in the image domain, is used to explain the CNN network after the fitting is completed. It is found that the model pays too much attention to salient features. Thus Mutual Channel Loss is further implemented to make the model pay attention to those indistinguishable features. At the same time, CBAM attention module is added to pay selective attention to different characteristics of enterprises in different years. The three-year information of 15050 enterprises collected from the State Administration for Industry and Commerce of china is used as the source data. The results show that comparing with other models, our CNN model reached the state of art in the rate of misjudgment and missed judgment. © 2021 IEEE.

关键词： Time series

来源：评论

学校读者我要写书评

暂无评论

Boosting the Restoring Performance of Deduplication Data by Classifying Backup Metadata

引用

ACM/IMS Transactions on Data science 2021年第2期2卷 1–16页

作者： Yang, Ru Deng, Yuhui Zhou, Yi Huang, Ping Department of Computer Science Jinan University Guangzhou510632 China Department of Computer Science Jinan University China State Key Laboratory of Computer Architecture Institute of Computing Technology Chinese Academy of Sciences China TSYS School of Computer Science Columbus State University Doimukh Georgia College of Engineering Temple University United States

Restoring data is the main purpose of data backup in storage systems. The fragmentation issue, caused by physically scattering logically continuous data across a variety of disk locations, poses a negative impact on the restoring performance of a deduplication system. Rewriting algorithms are used to alleviate the fragmentation problem by improving the restoring speed of a deduplication system. However, rewriting methods give birth to a big sacrifice in terms of deduplication ratio, leading to a huge storage space waste. Furthermore, traditional backup approaches treat file metadata and chunk metadata as the same, which causes frequent on-disk metadata accesses. In this article, we start by analyzing storage characteristics of backup metadata. An intriguing finding shows that with 10 million files, the file metadata merely takes up approximately 340 MB. Motivated by this finding, we propose a Classified-Metadata based Restoring method (CMR) that classifies backup metadata into file metadata and chunk metadata. Because the file metadata merely takes up a meager amount of space, CMR maintains all file metadata in memory, whereas chunk metadata are aggressively prefetched to memory in a greedy manner. A deduplication system with CMR in place exhibits three salient features: (i) It avoids rewriting algorithms' additional overhead by reducing the number of disk reads in a restoring process, (ii) it increases the restoring throughput without sacrificing the deduplication ratio, and (iii) it thoroughly leverages the hardware resources to boost the restoring performance. To quantitatively evaluate the performance of CMR, we compare our CMR against two state-of-the-art approaches, namely, a history-aware rewriting method (HAR) and a context-based rewriting scheme (CAP). The experimental results show that compared to HAR and CAP, CMR reduces the restoring time by 27.2% and 29.3%, respectively. Moreover, the deduplication ratio is improved by 1.91% and 4.36%, respectively. © 2

关键词： Metadata

来源：评论

学校读者我要写书评

暂无评论

Genetic Algorithm for PI Controller Design of Grid-connected Inverter based on Multilayer Perceptron Model

Genetic Algorithm for PI Controller Design of Grid-connected...

引用

2021 IEEE Sustainable Power and Energy Conference, iSPEC 2021

作者： Yang, Qifan Dai, Ningyi University of Macau Skl of Internet of Things for Smart City Department of Electrical and Computer Engineering China Zhuhai Um Science and Technology Research Institute China

ISBN: (纸本)9781665414395

Proportional-integral (PI) controllers are the most commonly used units in the control system of power electronics converters. The design of PI controller has long been an empirical process and is usually a time-consuming task. In this paper, the raw dataset is generated from a grid-connected inverter case running on Real Time Digital Simulators (RTDS). A clean training dataset is used to train a multi-layer perceptrons (MLP) neural network, which can approximate the mapping between PI parameters and the performance of the grid-connected inverter. Based on the verified MLP model, genetic algorithm (GA) is applied to design PI parameters for improving the active power control of the grid-connected inverter. The PI parameter tuning procedure requires almost no human interference. The validity of the proposed method in designing PI parameters is verified by simulation results. The benchmark case uses PI parameters from RTDS's tutorial without optimization. The active power control performance of the grid-connected inverter is improved by using the optimized PI parameters. © 2021 IEEE.

关键词： Multilayer neural networks

来源：评论

学校读者我要写书评

暂无评论

Detection Method of Surface Defects of Electronic Components Using the Deep Network of You Only Look Once Version 5 2

Detection Method of Surface Defects of Electronic Components...

引用

2nd International Conference on Mechanical Engineering, Intelligent Manufacturing and Automation technology, MEMAT 2022

作者： Lin, Yihao Fan, Yong Wu, Shuqin Gong, Yubing Xie, Wu School of Mechanical & Electrical Engineering Guilin University of Electronic Technology Guangxi Guilin China School of Architecture and Transportation Engineering Guilin University of Electronic Technology Guangxi Guilin China School of Computer Science and Information Security Guilin University of Electronic Technology Guangxi Guilin China

ISBN: (纸本)9783800757619

The production of electronic components is becoming more and more modern and efficient. How to detect surface defects of electronic components more accurately and quickly has become a more and more important research topic. To solve this problem, this paper designs a detection method based on visual detection and image processing is proposed. The proposed method builds a target detection algorithm model based on the YOLOv5 (You Only Look Once Version 5) convolutional neural network, and combines the image target detection and recognition with the convolutional neural network to make its detection and recognition effect more accurate and efficient. This paper tested the 4 surface defects with the patch resistance SS14 to verify the reliability of the system. Through the result analysis, the loss function of the detection system is less than 0.3, and its map is above 0.96. It meets the high accuracy and high accuracy requirements of the surface defect detection of electronic components. © VDE VERLAG GMBH ∙ Berlin ∙ Offenbach.

关键词： Surface defects

来源：评论

学校读者我要写书评

暂无评论

An Identification Method for Mechanical Parts Using on Deep Learning with Single Shot-multibox Detector MobileNet 2

An Identification Method for Mechanical Parts Using on Deep ...

引用

2nd International Conference on Mechanical Engineering, Intelligent Manufacturing and Automation technology, MEMAT 2022

作者： Liu, Junfeng Fan, Yong Wu, Shuqin Liang, Zhenghao Xie, Wu Fu, Liye School of Mechanical and Electrical Engineering Guilin University of Electronic Technology Guangxi Guilin China School of Architecture and Transportation Engineering Guilin University of Electronic Technology Guangxi Guilin China School of Computer Science and Information Security Guilin University of Electronic Technology Guangxi Guilin China

ISBN: (纸本)9783800757619

SSD (Single Shot-multibox Detector)-MobileNet is widely used in embedded systems and intelligent manufacturing systems due to its lightweight structure and acceptable precision loss. In order to identify mechanical parts, images of four parts are collected, labelled and identified based on SSD-MobileNet with deep learning model. The TensorFlow frameworks are used only for training, and the trained models are applied to implement real-time detection. The results show that the model has a good recognition effect on mechanical parts with the accuracy of 0.936 and the recall rate of 0.956 under the conditions of not large-scale quantity of parts, which is useful to meet the identification and detection of parts to achieve the application purpose of parts classifications. © VDE VERLAG GMBH ∙ Berlin ∙ Offenbach.

关键词： Embedded systems

来源：评论

学校读者我要写书评

暂无评论

SplitKV: Splitting IO paths for different sized key-value items with advanced storage devices 12

SplitKV: Splitting IO paths for different sized key-value it...

引用

12th USENIX Workshop on Hot Topics in Storage and File Systems, HotStorage 2020, co-located withUSENIX ATC 2020

作者： Han, Shukai Jiang, Dejun Xiong, Jin SKL Computer Architecture ICT CAS University of Chinese Academy of Science China

Modern advanced storage devices, such as modern NVMe SSD and non-volatile memory based persistent memory (PM), provide different access features when data size varies. Existing key-value stores adopt unified IO path for all key-value items, which cannot fully exploit the advantages of different advanced storage devices. In this paper, we propose to split IO paths for different sized key-value items. We let small key-value items be written directly to PM and then migrated to SSD. Meanwhile, large key-value items are directly written to SSD. We present and discuss design choices towards challenging issues of splitting IO paths. We build SplitKV, a key-value store prototype to show the benefits of splitting IO paths. The preliminary results show SplitKV outperforms RocksDB, KVell, and NoveLSM under small-large KV mixed workloads. © HotStorage 2020 - 12th USENIX Workshop on Hot Topics in Storage and File Systems, co-located with USENIX ATC 2020. All rights reserved.

关键词： File organization

来源：评论

学校读者我要写书评

暂无评论

A Sketch-Based Generation Model for Diverse Ceramic Tile Images Using Generative Adversarial Network

引用

Intelligent Automation & Soft Computing 2023年第9期37卷 2865-2882页

作者： Jianfeng Lu Xinyi Liu Mengtao Shi Chen Cui Mahmoud Emam School of Computer Science and Technology Hangzhou Dianzi UniversityHangzhou310018China Key Laboratory of Public Security Information Application Based on Big-Data Architecture Ministry of Public SecurityZhejiang Police CollegeHangzhou310000China Faculty of Artificial Intelligence Menoufia UniversityShebin El-Koom32511Egypt

Ceramic tiles are one of the most indispensable materials for interior *** ceramic patterns can’t match the design requirements in terms of diversity and interactivity due to their natural *** this paper,we propose a sketch-based generation method for generating diverse ceramic tile images based on a hand-drawn sketches using Generative Adversarial Network(GAN).The generated tile images can be tailored to meet the specific needs of the user for the tile *** proposed method consists of four ***,a dataset of ceramic tile images with diverse distributions is created and then pre-trained based on ***,for each ceramic tile image in the dataset,the corresponding sketch image is generated and then the mapping relationship between the images is trained based on a sketch extraction network using ResNet Block and jump connection to improve the quality of the generated ***,the sketch style is redefined according to the characteristics of the ceramic tile images and then double cross-domain adversarial loss functions are employed to guide the ceramic tile generation network for fitting in the direction of the sketch style and to improve the training ***,we apply hidden space perturbation and interpolation for further enriching the output textures style and satisfying the concept of“one style with multiple faces”.We conduct the training process of the proposed generation network on 2583 ceramic tile images *** measure the generative diversity and quality,we use Frechet Inception Distance(FID)and Blind/Referenceless Image Spatial Quality Evaluator(BRISQUE)*** experimental results prove that the proposed model greatly enhances the generation results of the ceramic tile images,with FID of 32.47 and BRISQUE of 28.44.

关键词： Ceramic tile pattern design cross-domain learning deep learning GAN generative adversarial networks ResNet Block

来源：评论

学校读者我要写书评

暂无评论

Emerging Techniques for Online Learning Analytics 6th

Emerging Techniques for Online Learning Analytics

引用

6th International Conference on technology in Education, ICTE 2023

作者： Wang, Yidan Huang, Xiaoran Yu, Qijing Lai, YuXuan School of Computer Science The Open University of China Beijing China School of Architecture and Arts North China University of Technology Beijing China Academic Affairs Department The Open University of China Beijing China

ISBN: (纸本)9789819982547

The rapid growth of online learning platforms presents both challenges and opportunities for educational institutions. Online learning analytics has the potential to extract valuable insights from student learning behaviours, and such insights could offer learners flexible access to educational resources and foster lifelong learning opportunities. This paper reviews the research progress and emerging techniques in online learning analytics. We first outline the objectives of learning analytics, emphasising its potential to enhance personalised learning experiences, and examine key challenges faced in such platforms. We also review state-of-the-art techniques and explore emerging trends, natural language processing and multimodal learning analytics in particular. The paper discusses the implications of online learning analytics for educators and instructional designers, identifies open research challenges, and inspires further research and innovation to support learners in their educational journey. © 2024, The Author(s), under exclusive license to Springer Nature Singapore Pte Ltd.

关键词： Data handling

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：