Mobile users can open program in new tab for better viewing.

Open program in new tab

Day 1 13/11/2020
Room #1

Welcome message by the Organizing Committee 09:00 - 09:05

Starts at 9:00 AM China local time (GMT+8)

Welcome message by EAI Conference Manager 09:05 - 09:10

Welcome message by EAI Community Manager 09:05 - 09:10

Session 1 09:10 - 11:15

09:10 - 09:35
Prototyping an SDN Control Framework for QoS Guarantees

The centralized control capability of Software-Defined Networking (SDN) provides an excellent opportunity to enhance the Quality of Service (QoS) routing in networking environments. The end-to-end QoS-aware traffic forwarding must consider the computation latency associated with optimal path selection while reducing the controller's response time. In this paper, we propose SDN-enabled systematic mechanisms that features a queueing scheme, active link delay measurements, efficient statistic estimate of network states, and optimal path computation and selection. We implement our framework as a modular application in the Floodlight SDN controller software and conduct comprehensive experimental studies on the Global Environment for Network Innovations (GENI) testbed. Our performance evaluation demonstrates that the proposed prototype can optimize the control's latency, switching latency, and therefore find optimal end-to-end paths with the minimum delay with the reduced control overhead.
Authors: Mohamed Rahouti (Department of Computer & Information Science, Fordham University), Kaiqi Xiong (University of South Florida), Yufeng Xin (RENCI, University of North Carolina at Chapel Hill),
Hide Authors & Abstract

Show Authors & Abstract
09:35 - 09:50
Precision Improvement of Overprinting System based on Improved Laplace Edge Detection Algorithm

This paper proposes a solution to locate the crossline center in overprinting system. Crosslines of 4 different colors are printed at the same coordinate. Because of mechanical error, they don't coincide. The system uses the meas-ured deviation to calibrate. This paper obtains accurate deviation value through three steps. First, the picture collected by a CCD camera will be processed by color segregation algorithm. In this process, RGB data turn into CMYK data, and each color will be on a single picture. After that, a Laplace edge detection algorithm is improved by combining it with 2D Gauss filter. This improved Laplace edge detection algorithm has a better noise suppres-sion effect, which means it is even less likely to judge noise as the edge of a graph. Finally, target searching algorithm based on rhombus matching is used to figure out the center of crossline. The average absolute error from 2,000 simulations is 3.192 pixel, which shows that the algorithm in this pa-per has a high accuracy.
Authors: Yingbo Wang (Beijing Institute of Graphic Communication), Likun Lu (Beijing Institute Of Graphic Communication), Qingtao Zeng (Beijing Institute Of Graphic Communication), Rui Zhao (Beijing Institute Of Graphic Communication), Yang Zhang (Beijing Institute Of Graphic Communication), Fucheng You (Beijing Institute of Graphic Communication),
Hide Authors & Abstract

Show Authors & Abstract
09:50 - 10:15
Requirements for Deploying IP and ICN Network Stacks on a Common Physical Infrastructure

Deploying alternative networks such as Information-Centric Network (ICN) in a production/commercial network with real users is always challenging due to the inherent experimental nature of these novel proposals. Adopting an ICN-IP dual stack approach helped us to ameliorate many of these challenges. However, this was at the cost of introducing unpredictable emergent behavior. In this paper, we present a set of requirements that seek to impose a constraining discipline on the deployment and operational processes for the dual stack to deal with its emergent behavior in a tractable fashion. These requirements are extracted from our experience of a lab deployment followed by a field deployment with real users as part of the three-year long EU-funded RIFE project.We summarize them in a form that can be used by other practitioners in their own ICN/alternative network stack deployments and by tool developers for such deployments. These presented requirements compliment the current discussions within the Information-Centric Networking Research Group (ICNRG) of the Internet Research Task Force (IRTF).
Authors: Renan Krishna (Interdigital Europe Limited), Roger Vinas (Universitat Politecnica de Catalunya, Barcelona, Catalunya),
Hide Authors & Abstract

Show Authors & Abstract
10:15 - 10:35
A Pure Network-based Approach To Achieve Always Best Quality Video Streaming

The demand for video streaming has been more and more increasing, causing the streaming technology and the related technologies to be improved to meet the re-quirement of the best quality of experience (QoE) from various users. A lot of re-search has been focusing on studying users’ behavior or developing streaming client and/or server application. These works estimate the network state passively and are lack of a global view of the network. As a result, they meet difficulty in bandwidth competition, QoE fairness scenarios. Some works optimize routing mechanism to improve video quality and QoE. This work also proposes a pure network-based approach, however taking into account the characteristics of video streaming application, to support an always best QoE to end-users. The proposed approach leverages the advantage of SDN network to convert the explored char-acteristic of streaming application into the network configuration. The proposal has been implemented on a real testbed. The obtained results show that the pro-posed mechanism has maintained the best video quality while maximizing the bandwidth for competitor application, i.e., file download using ftp protocol.
Authors: Toan Nguyen (Hanoi University of Science and Technology), Xuan Doan (Hanoi University of Science and Technology), Hung Do (Hanoi University of Science and Technology),
Hide Authors & Abstract

Show Authors & Abstract
10:35 - 10:55
CNN-based book cover and back cover recognition and classification

As an important part of the national economy and an important supporting indus-try, the printing and publishing industry is closely related to the development of the national economy. In recent years, the massive distribution and publication of books has made the work of book storage into database more and more onerous. The maturity of deep learning technology has brought good news to the recogni-tion and classification of books. Convolutional neural network is a good tool. Convolutional neural network is a technology in deep learning, often used in computer vision, image recognition classification and other fields. The research results in the field of book recognition and classification are relatively lacking. There is no good book data set that can be used for neural network training. In this paper, we collected a large number of book data sets and we built a set of im-age classification models based on CNN to identify and classify the cover and back cover of books. Through a lot of training and testing, we have generated a set of CNN models that can effectively identify and classify the cover and back cover of books. Compared with the traditional way of manually entering books into the database, the use of neural networks makes the work more efficient and saves a lot of human resources.
Authors: Haochang Xia (Beijing Institute of Graphic Communication), Yali Qi (Beijing Institute of Graphic Communication), Qingtao Zeng (Beijing Institute of Graphic Communication), Yeli Li (Beijing Institute of Graphic Communication), Fucheng You (Beijing Institute of Graphic Communication),
Hide Authors & Abstract

Show Authors & Abstract
10:55 - 11:15
A case study of Linguistic Research Methods in the age of Computing

With the development of technologies such as computers, artificial intelligence, and big data, new research methods have emerged in the traditional social sciences. With the help of new research methods, the efficiency and accuracy of quantitative analysis can be improved, which helps to improve research efficiency and demonstrate research trends. This paper uses the data analysis software citespace as an example to analyze the current situation of domestic Tibetan-Burman language research, and graphically display the analysis results.
Authors: shao tie-jun (Research Center For Language Intelligence of China;Capital Normal University), zhou jianshe (Research Center For Language Intelligence of China;Capital Normal University), zhang wenyan (China Fire and Rescue Institute),
Hide Authors & Abstract

Show Authors & Abstract

Lunch break 11:15 - 11:45

Session 2 11:45 - 13:45

11:45 - 12:00
An analysis model of automobile running state based on neural network

A reasonable design of the operating condition curve of automobile driving characteristics is conducive to improving the credibility of the government, so it is more and more important to formulate a test condition that reflects the actual road driving conditions in China. In order to construct the model mainly by two-segment clustering, the initial clustering of the processed data is carried out by self-organizing mapping neural network, and the cluster number and clustering center are obtained to solve the problem of poor con-vergence in the K-means model in the early stage. In view of the construc-tion of the operating condition curve of the driving characteristics of light vehicles in a city, the data preprocessing, the extraction of motion fragments and the construction of the driving conditions of a car are to be provided for the driving data set of the same vehicle in a city.
Authors: Jie Yan (Beijing Institute of Graphic Communication), Xinxin Guan (National Museum of China), Qingtao Zeng (Beijing Institute of Graphic Communication), Chufeng Zhou (China Women's News), Yeli Li (Beijing Institute of Graphic Communication), Fucheng You (Beijing Institute of Graphic Communication),
Hide Authors & Abstract

Show Authors & Abstract
12:00 - 12:20
Fusing Bert and BiLSTM Model to Extract the Weaponry Entity

Weaponry entity extraction is an indispensable link in the process of con-structing a weaponry knowledge graph. In terms of entity extraction of weap-ons and equipment, a fusion model of domain BERT model and BILSTM model with embedded word vectors and word conversion rate vectors is pro-posed to identify weapons and equipment entities. First, the BERT model is used to perform pre-training tasks on massive weaponry corpus. Secondly, the Word2vec model is used to train the word vectors to provide a priori semantic information, and the word conversion rate vector is embedded to input more a priori information to the model. Finally, the hierarchical entity extractor ex-tracts entities of different categories. Experiments results show that the fusion model has strong coding ability and sufficient prior knowledge, and the F1 value on the Global Military Network corpus reaches 91.436%.
Authors: Haojie Ge (Beijing Information Science and Technology University), Xindong You (Beijing Information Science and Technology University), Jialai Tian (Beijing Information Science and Technology University), Xueqiang Lv (Beijing Information Science and Technology University),
Hide Authors & Abstract

Show Authors & Abstract
12:20 - 12:40
CIC Chinese Image Captioning Based on Image Label Information

Although image captioning technology has made great progress in recent years, the quality of Chinese image description is far from enough. In this paper, we focus on the problem of Chinese image captioning with the aim to improve the quality of Chinese image description. A novel framework for Chinese image captioning based on image label information (CIC) is proposed in this paper. Firstly, image label information is extracted by a multi-layer model with shortcut connections. Then the label information is input into the neural network with an extension of LSTM, which we coin L-LSTM for short, to generate the Chinese image descriptions. Extensive experiments are conducted on various image caption datasets such as Flickr8k-cn, Flickr30k-cn. The experimental results verify the effectiveness of the proposed framework (CIC). It obtains 27.1% and 21.2% BLEU4 average values of Flickr8k-cn and Flickr30k-cn, respectively, which outperforms the state-of-art model in Chinese image captioning domain.
Authors: Xindong You (Beijing Information Science and Technology University), Likun Lu (Beijing Institute Of Graphic Communication), Hang Zhou (Beijing Information Science and Technology University), Xueqiang Lv (Beijing Key Laboratory of Internet Culture and Digital Dissemination Research, Beijing Infor-mation Science & Technology University, Beijing, China),
Hide Authors & Abstract

Show Authors & Abstract
12:40 - 13:00
Cross-language Transfering the Patent Quality Evaluation Model Based on Active Learning Data Extension

At present, China has become a major producer of patents, and the number of patent applications has been in the first position in the world for many years. While the number of patents has increased, the quality of patents has begun to receive people's attention. At present, there is no clear evaluation method for Chinese patents. To evaluate patents manually requires a large number of relevant experts to study and compare patents in different fields, which is time-consuming and laborious. In the previous study, the author constructed an English patent quality evaluation model PQE-MT using U.S. Patents that represent patent strength. In this paper, the model will be migrat-ed to Chinese patents through transfer learning and active learning, thus min-imizing the work of manual labeling. The evaluation results show that the method in the experiment has achieved a good migration effect, Micro-F1 reaching 74%.
Authors: Jiaqi Liu (Beijing Information Science and Technology University), Xindong You (Beijing Information Science and Technology University), Zhe Wang (Beijing Information Science and Technology University), Xueqiang Lv (Beijing Information Science and Technology University),
Hide Authors & Abstract

Show Authors & Abstract
13:00 - 13:20
Construction of Unsupervised Prose Text Emotional Lexicon Based on Multidimensional Fusion

Abstract. Affective computing is an important tool for language processing and opinion mining, and emotional lexicon is the basis of emotional computing, and prose accounts for a large proportion in Chinese teaching and application in China. The construction of special emotional lexicon for prose language learning and language understanding is of great significance to the development of machine assisted human language learning and the improvement of machine deep reading comprehension. Therefore, the research on the construction of prose emotional lexicon is of great significance and value. In this paper, with the help of data collection tools, more than 27000 pieces of modern famous prose database are constructed. After preprocessing the data, denoising, deleting and selecting are completed to determine the walk set. Compared with PMI and word2vec, the accuracy of the method is improved by 16% and 14.8%, which proves that the comprehensive vector space can effectively improve the emotional vocabulary recognition of prose.Finally, 12762 prose general emotional lexicon is formed with the help of this method.
Authors: Kai Zhang (Capital Normal University), Jianshe Zhou (Capital Normal University), Su Dong (Capital Normal University),
Hide Authors & Abstract

Show Authors & Abstract
13:20 - 13:45
Algorithm Based on LL_CBF for Large Flows Identification

In order to manage large-scale network, it is very important to measure and monitor the network traffic accurately. Identifying large flows timely and accurately provide data support for network management and network security, which has important meaning. Aiming at the deficiency of high false negative rate by using traditional algorithm to detect large flows, a novel scheme called LL_CBF is presented, which uses the policies of “separation of large flow filtering and large flow identification” to improve the accuracy of traffic measurement. The algorithm is improved from four aspects: large flows handled firstly, using counting bloom filter to filtrate most small flows, using least recent used mechanism to filter small and medium flows and pre-protect large flows, and using least elimination strategy to identify large flows. The theoretical analysis and the simulation result indicates that compared with the standard LRU algorithm and LRU_BF algorithm, our algorithm can identify the large flow in the network timely and accurately, and reduce the computing resource requirements effectively.
Authors: lei bai (Capital Normal University), Jianshe Zhou (Capital Normal University), Yaning Zhang (Capital Normal University),
Hide Authors & Abstract

Show Authors & Abstract