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

Open program in new tab

Day 1 16/10/2020
Room #1

Organizing Committee Welcome 08:40 - 08:45

Beijing Time Zone

Welcome Message by EAI Conference Manager 08:45 - 08:55

Welcome Message by EAI Community Manager 08:55 - 09:00

Keynote Speech by Prof. Fumiyuki Adachi 09:00 - 10:00

Coffee Break 10:00 - 10:30

30 minutes

Keynote Speech by Prof. Deke Guo 10:30 - 11:30

Lunch Break 11:30 - 13:30

Session 1: Collaborative Applications for Network and E-Commerce 13:30 - 15:20

13:30 - 13:45
Towards Accurate Search for E-Commerce in Steel Industry: A Knowledge-Graph-Based Approach

Mature artificial intelligence (AI) makes human life more and more convenient. However, in some application fields, it is impossible to achieve the satisfactory results only depending on the traditional AI algorithm. Specifically, in order to avoid the limitations of traditional searching strategies in e-commerce field related to steel, such as the inability to analyzing long technical sentences, we propose a collaborative decision making method in this field, through the combination of deep learning algorithms and expert systems. Firstly, we construct a knowledge graph (KG) on the basis of steel commodity data and expert data-base, and then train a model to accurately extract steel entities from long technical sentences, while using an advanced bidirectional encoder representation from transformers (BERT), a bidirectional long short-term memory (Bi-LSTM), and a conditional random field (CRF) approach. Finally, we develop an intelligent searching system for e-commence in steel industry, with the help of the designed KG and entity extraction model, while improving the searching performance and user experience in such system.
Authors: Maojian Chen (University of Science and Technology Beijing), Hailun Shen (Ouyeel Co., Ltd), Ziyang Huang (Ouyeel Co., Ltd), Xiong Luo (University of Science and Technology Beijing), Junluo Yin (University of Science and Technology Beijing),
Hide Authors & Abstract

Show Authors & Abstract
13:45 - 14:05
WSN Coverage Optimization Based on Two-Stage PSO

Wireless Sensor Networks(WSN) coverage perception is an important basis for communication between the cyber world and the physical world in Cyber-Physical Systems(CPS). To address the coverage redundancy, hole caused by initial random deployment and the energy constraint in redeployment, this paper proposes a multi-objective twostage particle swarm optimization algorithm(MTPSO) based on coverage rate and moving distance deviation to improve coverage efficiency. This algorithm establishes a multi-objective optimization model for above problems, and determines the candidate deployment scheme by reducing its local convergence probability through improved inertia weight, and then introduces virtual force mechanism to adjust the relative position between nodes. This paper mainly analyzes the influence of different initial deployment category and mobile nodes proportion on multi-objective optimization performance, and gives the corresponding algorithm implement. Simulation experiments show that compared with MVFA, SPSO and OPSO algorithms, MTPSO algorithm has a better redeployment coverage performance, which fully demonstrates its effectiveness.
Authors: Wei Qi (East China University of Science and Technology), Huiqun Yu (Dept of Computer Science and Engineering,East China University of Science and Technology), Guisheng Fan (Dept of Computer Science and Engineering,East China University of Science and Technology), Liang Chen (East China University of Science and Technology), Xinxiu Wen (East China University of Science and Technology),
Hide Authors & Abstract

Show Authors & Abstract
14:05 - 14:25
A covert ultrasonic phone-to-phone communication scheme

With the increasing penetration of mobile phones, modern life is closely related to mobile phones. The common wireless communication scheme on mobile phones are based on electromagnetic wave transmission. However, the attenuation rate of electromagnetic waves in the air medium is very low, and the detectable range of electromagnetic waves is relatively far, which causes wireless communication a risk of information leakage. To achieve a covert channel for short-distance mobile phone to mobile phone communication, we design a scheme based on ultrasonic wave transmission. The scheme uses chirp signal BFSK modulation and convolutional coding, and Cepstrum method is applied to compensate the effect of RIR. Through experiments, error rate of the ultrasonic communication system designed for mobile phone can be within 10−5 in 1-meter range.
Authors: Liming Shi (Xi'an Jiaotong-Liverpool University), Limin Yu (Xi'an Jiaotong-Liverpool University), Kaizhu Huang (Xi'an Jiaotong-Liverpool University), Xu Zhu (University of Liverpool), Zhi Wang (Zhejiang University), Xiaofei Li (Westlake University), Wenwu Wang (University of Surrey), Xinheng Wang (Xi'an Jiaotong-Liverpool University),
Hide Authors & Abstract

Show Authors & Abstract
14:25 - 14:40
An Efficient and Truthful Online Incentive Mechanism for a Social Crowdsensing Network

Crowdsening plays an important role in spatiotemporal data collection by leveraging ubiquitous smart devices equipped with sensors. Considering rational and strategic device users, designing a truthful incentive mechanism is a crucial issue. Moreover, another key challenge is that there may not exist adequate participating users in reality. To encourage more users to participate, the social relationship among them can be leveraged, as users may be significantly influenced by their social friends. In this paper, we assume recruited users to diffuse uncompleted sensing tasks to their friends, and propose an efficient and truthful online incentive mechanism for a such social crowdsensing network. Specially, we model the time-varying social influence of a user by extending two metrics of node centrality used in social networks. In order to maximize the accumulated social welfare achieved by the network, we design a user selection algorithm and a payment determination algorithm respectively, in which payments given to participants not only depend on data qualities but also related with social influences. We theoretically prove that our mechanism achieves properties of computational efficiency, individual rationality, and truthfulness. Extensive simulations are conducted, and the results show the superiority of our mechanism.
Authors: Lu Fang (Shanghai University), Tong Liu (Shanghai University), Honghao Gao (Shanghai University), Chenhong Cao (Shanghai University), weimin li (Shanghai University), Weiqin Tong (Shanghai University),
Hide Authors & Abstract

Show Authors & Abstract
14:40 - 15:00
Toward Sliding Time Window of Low Watermark to Detect Delayed Stream Arrival

Some emergency events such as time skew between input streams, operator-induced disorder, and network delay might cause stream processing system produce unbounded out-of-order data streams. Recent work on this issue focuses on explicit punctuation or heartbeats to handle faults and stragglers (outlier data). Most parallel and distributed models on stream processing, such as Google MillWheel and Apache Flink, require hot replication, logging, and upstream backup in an expensive manner. But these frameworks ignore straggler processing. Some latest frameworks such as Google MillWheel and Apache Flink only process disorder on an operator level, but only point-in-time and fixed window of low watermarks are discussed. Therefore, we propose a new sliding time window of low watermark to detect delayed stream arrival. Contributions of our methods conclude as adaptive low watermark, distinguishing stragglers from late data, and dynamic rectification of low watermark. The experiments show that our method is better in tolerating more late data to detect stragglers accurately.
Authors: Xiaoqian Zhang (University of Jinan), Kun Ma (University of Jinan),
Hide Authors & Abstract

Show Authors & Abstract
15:00 - 15:20
CPNSA: Cascade Prediction With Network Structure Attention

Online social medias provide convenient platforms for information spread, which makes the social network structure plays important role on online information spread. Although online social network structure can be obtained easily, few researches use network structure information in the cascade of the resharing prediction task. In this paper, we propose a cascade prediction method (named by CPNSA) involves the network structure information into cascade prediction of resharing task. The method is based on the recurrent neural network, and we introduce a network structure attention to incorporates the network structure information into cascade representation. In order to fuse network structure information with cascading time series data, we use network embedding method to get the representations of nodes from the network structure firstly. Then we use the attention mechanism to capture the structural dependency for cascade prediction of resharing. Experiments are conducted on both synthetic and real-world datasets, and the results show that our approach can effectively improve the performance of the cascade prediction of resharing.
Authors: chaochao liu (Tianjin University), Wenjun Wang (Tianjin University), Pengfei Jiao (Tianjin University), Yueheng Sun (Tianjin University), xiaoming Li (TianJin University), Xue Chen (Tianjin University),
Hide Authors & Abstract

Show Authors & Abstract

Coffee Break 15:20 - 15:55

35 minutes

Session 4: Artificial Intelligence 15:55 - 17:50

15:55 - 16:15
Identification of Sequential Feature for Volcanic Ash Cloud Using FNN-LSTM Collaborative Computing

Collaborative computing performs quickly and accurately the task via combining the multimedia, multi-methods, and multi-clients. Analyzing of traditional feed-foward neural network (FNN), long short term memory (LSTM) neural net-works and remote sensing data, this paper proposes a new identification method of sequential feature based on FNN-LSTM collaborative calculation in the vol-canic ash cloud monitoring. In this method, combining remote sensing data, the FNN network is used firstly to construct the identification model of volcanic ash cloud. Next, the LSTM network is used to identify the sequential feature of dy-namic changes in volcanic ash cloud based on the text data of the volcanic ash re-port. And then the simulation and true volcanic ash cloud case is performed and analyzed. The experimental results show that: 1) the proposed method is high in training accuracy with 76.54% and testing accuracy with 77%, respectively, and has obvious advantages for small-scale data volumes; 2) the total accuracy and RMS of the simulation analysis reached 79.05% and 0.0149, respectively, it veri-fied the feasibility and effectiveness in the prediction of spatiotemporal evolution; 3) the anti-noise property and the image segmentation effect is good, the obtained sequential feature of the volcanic ash cloud are closer to the actual diffusion. It can provide a reference for sequential feature extraction and dynamic monitoring of volcanic ash cloud in complex environments.
Authors: lan liu, chengfan li (Shanghai Nniversity, China), xiankun sun (Shanghai University of Engineering Science), Jiangang Shi (Shanghai Shang Da Hai Run Information System Co),
Hide Authors & Abstract

Show Authors & Abstract
16:15 - 16:35
Automated Detection of Standard Image Planes in 3D Echocardiographic Images

Background: During the diagnosis and analysis of complex congenital heart malformation, it is time-consuming, as well as tedious, for doctors to search for standard image planes by hand from among the huge amounts of patients' three-dimensional(3D) ultrasound heart images. To relieve the laborious manual searching task for echocardiographers, especially for non-physicians, this paper focuses on the auto-detection of five standard image planes suggested by experts in the 3D echocardiographic images. Methods: The four-chamber(4C) image plane is first auto-detected by template matching, and then the other standard image planes are obtained according to their spatial relation with the 4C image plane. Results: We have tested our methods on 28 normal and 22 abnormal datasets, and the error rates are 7.1% and 13.6%, respectively. Conclusion: With low computational complexity and simple operation, the method of auto-detection of standard planes in 3D echocardiographic images shows encouraging prospects of application.
Authors: Wei Peng (School of Computer Science and Software Engineering, East China Normal University,), XiaoPing Liu (East China Normal University), Lanping Wu (Shanghai Jiaotong University),
Hide Authors & Abstract

Show Authors & Abstract
16:35 - 16:50
Hybrid CF on Modeling Feature Importance with Joint Denoising AutoEncoder and SVD++

Abstract. AutoEncoder is an unsupervised learning approach that can maps inputs to useful intermediate features,which can be used to build recommendation. Intermediate features of different entities obtained by AutoEncoder may have different weight for predicting users behavior. However,existing research typically uses a uniform weight on intermediate features to make a fast learning algorithm,this general approach may lead to the limited performance of the model. In this paper,we proposes a novel approach by using SGD to dynamically learn the intermediate features importance,which can integrate the intermediate features into matrix factorization framework seamlessly. In the previous works,the entities intermediate features learned by AutoEncoder are modeled as a whole. On this basis,we proposes to use attention parameters in entity intermediate feature to dynamically learn the intermediate features importance and build fine-grained model. By learning unique attention unit for each entity intermediate feature,the entities intermediate features are integrated into the matrix factorization framework better. Extensive experiments conducted over two real-world datasets demonstrate our proposed approach outperforms the compared models.
Authors: Qing Yang (Guangxi Key Laboratory of Automatic Detection Technology and Instrument, Guilin University of Electronic Technology, Guilin 541004, China), Heyong Li (Guangxi Key Laboratory of Automatic Detection Technology and Instrument, Guilin University of Electronic Technology, Guilin 541004, China), Ya Zhou (Guangxi Key Laboratory of Trusted Software, Guilin University of Electronic Technology. Guilin 541004, China), Jingwei Zhang (Guangxi Key Laboratory of Trusted Software, Guilin University of Electronic Technology. Guilin 541004, China), Stelios Fuentes (Leicester University, UK),
Hide Authors & Abstract

Show Authors & Abstract
16:50 - 17:10
Sentiment Analysis of Film Reviews Based on Deep Learning Model Collaborated with Content Credibility Filtering

Sentiment analysis of film reviews is the basis of obtaining the opinions of movie viewers. It has an important influence on movie public opinion control and stimulating potential viewers. Due to the natural openness and randomness of social media, there may exist a considerable amount of useless or false information in film review comments, making it challenging to analyze the credibility of the comments. This paper proposes a fine-grained sentiment analysis method based on the key-viewpoint sentences of Chinese film reviews, where a deep learning model is used to classify the fine-grained emotions in film reviews. Based on the analysis results, a method for calculating the credibility of review comments is proposed. Under the credibility criteria, corpus screened through the overall sentiment classification can obtain 9% improvement on accuracy than the original corpus, which verifies the validity of the credibility algorithm. The higher quality corpus achieved the credibility algorithm is benefit for improving the accuracy of the sentiment classification.
Authors: Xindong You (Beijing Key Laboratory of Internet Culture and Digital Dissemination Research, Beijing Infor-mation Science & Technology University, Beijing, China), Xueqiang Lv (Beijing Key Laboratory of Internet Culture and Digital Dissemination Research, Beijing Infor-mation Science & Technology University, Beijing, China), Shangqqian Zhang (Beijing Key Laboratory of Internet Culture and Digital Dissemination Research, Beijing Infor-mation Science & Technology University, Beijing, China), Dawei Sun (China University of Geosciences), Shang Gao (School of Information Technology, Deakin University, Victoria 3216, Australia),
Hide Authors & Abstract

Show Authors & Abstract
17:10 - 17:30
Essence Computation Oriented Multi-semantic Analysis Crossing Multi-modal DIKW Graphs

In recent years, with the rise of big data cloud computing and edge computing, privacy protection has become an increasingly important issue. When protecting the privacy of target resources, the content often has multiple semantics. In this work, we focus on modeling multi-semantic content based on data graphs, information graphs, and knowledge graphs, and analyze the causes of multi-semantics, discuss solutions to eliminate multi-semantics, and provide corresponding algorithms . At the same time, a corresponding transformation paradigm is proposed for the cross-modal transformation of different types of resources in the process of eliminating multiple semantics.
Authors: Yucong Duan (Hainan University), Shijing Hu (Peking University),
Hide Authors & Abstract

Show Authors & Abstract
17:30 - 17:50
Distributed Reinforcement Learning with States Feature Encoding and States Stacking in Continuous Action Space

The practical application of reinforcement learning agents is often bottlenecked by the duration of training time. To accelerate training, practitioners often turn to distributed reinforcement learning architectures to parallelize and accelerate the training process. This work, we utilize the distributed reinforcement learning architecture to deal with continuous control tasks. The Importance Weighted Actor-Learner Architectures(IMPALA) decouples the acting and learning process to reduce queuing time. IMPALA attains higher scores on the new DMLab-30 set and the Atari-57 set because of its high performance, good scalability, and high efficiency. We extend IMPALA on the continuous control tasks with three changes. We encoder states into low dimensional data to establish an action distribution function that the agents have the ability to exploit and explore. A queue buffer is used to store a mini-batch data and discard them after training. In order to make the agent take appropriate action in the continuous control environment, we stack the past three steps states that attempt to make the robot moves smoothly. Finally, experiments are carried out on Mujoco tasks. The results show that our work is better than other distributed reinforcement learning algorithms.
Authors: Tianqi Xu (National Innovation Institute of Defense Technology), Dianxi Shi (Artificial Intelligence Research Center (AIRC), National Innovation Institute of Defense Technology (NIIDT), Beijing 100166, China), Zhiyuan Wang (Artificial Intelligence Research Center (AIRC), National Innovation Institute of Defense Technology (NIIDT), Beijing 100166, China), Xucan Chen (National Innovation Institute of Defense Technology (NIIDT)), Yaowen Zhang (National Innovation Institute of Defense Technology),
Hide Authors & Abstract

Show Authors & Abstract
Room #2

Session 2: Workshop 13:30 - 15:20

13:30 - 13:45
Speech2Stroke: Generate Chinese Character Strokes Directly from Speech

Chinese character is composed of spatial arrangement of strokes. A portion of these strokes combine to form phonetic component, which provides a clue to the pronunciation of the entire character, the others combine to form signific component, which indicates semantic level information for speech context. How closely the connection between the internal strokes of Chinese characters and speech? In this paper, we propose Speech2Stroke, a end-to-end model that exploits the phonetic and morphologic level information of pictographic words. Specifically, we generate strokes directly from the speech by Speech2Stroke. The performance of Speech2Stroke is evaluated by the specific stroke error rate(SER). The SER of the optimal model can achieve 20.61%. Through the experiments and analysis, we show that our model has the ability to capture the alignment between the internal structures of pictographic and audio.
Authors: Yinhui Zhang (Xi'an Jiaotong University), Wei Xi (Xi'an Jiaotong University), Sitao Men (Xi'an Jiaotong University), Zhao Yang (Xi'an Jiaotong University), Rui Jiang (Xi'an Jiaotong University), Yuxin Yang (Xi'an Jiaotong University), Jizhong Zhao (Xi'an Jiaotong University),
Hide Authors & Abstract

Show Authors & Abstract
13:45 - 14:05
TAB: CSI Lossless Compression for MU-MIMO Network

Multi-user MIMO (MU-MIMO) is an important technology to improve data transmission efficiency for future network, such as 5G and WiFi 6, due to its ability of enabling multi-users’ concurrent transmissions. To achieve concurrent diversity gains, MU-MIMO network resource allocation relies on the feedback of Channel State Information (CSI) from multiple clients. CSI feedbacks from large user population, however, heavily degrade the throughput of a MU-MIMO network. Pursuing smart CSI feedback, we present a CSI timeliness-aware balanced mechanism, named TAB . It is a novel MU-MIMO protocol to eliminate unnecessary feedback overhead and improve CSI utilization within channel coherence time. TAB is fully compatible with the WiFi 5/6 standard and most state-of-the-art CSI feedback strategies, and is easy to be deployed on existing WiFi systems. Our software-radio based implementation and testbed experimentation demonstrate that TAB substantially improves the throughput of both downlink and uplink MU-MIMO network by 1.5× at least.
Authors: Qigui Xu (Xi’an Jiaotong University), Wei Xi (Xi'an Jiaotong University), Lubing Han (Xi'an Jiaotong University), Kun Zhao (Xi'an Jiaotong University),
Hide Authors & Abstract

Show Authors & Abstract
14:05 - 14:25
Towards Mobility-Aware Dynamic Service Migration in Mobile Edge Computing

Mobile edge computing is beneficial to reduce service re- sponse time by pushing cloud functionalities to the network edge. However, it is necessary to consider whether to conduct service migration to ensure the quality of service as users migrate to new locations. It is challenging to make migration decisions optimally due to the mobility of the users. To address this issue, we propose a mobility-aware dynamic service migration scheme for mobile edge computing. In order to predict a mobile user s movement behavior in terms of boundary crossing probability, we use a new approach for modeling user mobility and formulate the service migration problem as a Markov Decision Process (MDP). This policy can effectively weigh the relationship between delay and migration costs. Our methods capture general cost models and provide a mathematical framework to design optimal service migration policies. Experimental evaluation based on real-world mobility traces of Beijing taxis show superior performance of the proposed solution.
Authors: Fangzheng Liu (China University of Petroleum, Beijing), Bofeng Lv (China University of Petroleum, Beijing), Jiwei Huang (China University of Petroleum, Beijing), Sikandar Ali (China University of Petroleum, Beijing),
Hide Authors & Abstract

Show Authors & Abstract
14:25 - 14:40
Research on Debugging Interaction of IoT Devices Based on Visible Light Communication

Wireless sensor networks are usually deployed in the places where users are dif-ficult to reach and maintain. Once a failure of sensor nodes occurs, debugging in-teraction is necessary depending on the wireless network infrastructure. Howev-er, some failures are very brought up by wireless communication malfunctions. The debugging interactions based on the traditional wireless networks become invalid. In addition, due to the low cost hardware structure of the sensor node, it is hard to provide an extra interaction method. To this end, we find out the poten-tial capacity of various visible light components equipped with general sensor nodes and smartphones, and implemented a duplexing debugging interaction sys-tem (DIS) based on visible light communication (VLC). In order to improve transmission rate of debugging codes, we also presented a CPU binary instruc-tion compression method for source coding and overlap dual-header pulse inter-val modulation (ODH-PIM) for channel coding. In the instruction compression method, we reuse op code and employ bit-mask to compress operand. Moreover, we designed the ODH-PIM by adding a half-strength pulse in DH-PIM. Exper-iment results show that our method achieve 84.11% average compression rate, and reduce transmission time by 10.71% compared to the dual-header pulse in-terval modulation (DH-PIM).
Authors: Jiefan Qiu (Zhejiang University of Technology), Chenglin Li (Zhejiang University of Technology), Yuanchu Yin (Zhejiang University of Technology), Mingsheng Cao (University of Electronic Science and Technology of China),
Hide Authors & Abstract

Show Authors & Abstract
14:40 - 15:00
Attacking the Dialogue System at Smart Home

Intelligent dialogue systems are widely applied in smart home systems, and the security of the such systems deserve concern. In this paper, we design a threatening scenario of dialogue systems in smart home. A trojan robot is disguised as one part of the whole system, but generates dialogue adversarial examples to attack the normal robots according to the information of users. To achieve the goal in such scenario, the responding speed, the correctness of the grammar and the consistency of semantic is necessary. Based on these requirements, we propose a novel method named Attention weight Estimation(APE) to allocate the keys words in a dialogue, and substitute these words with synonyms in real time. We perform our experiments on popular classification dataset in DNN model, and the result shows that APE effectively attack the system with low responding time and high success rate.
Authors: Erqiang Deng (1.School of Information and Software Engineering, University of Electronic Science and Technology of China), Zhen Qin (University of Electronic Science and Technology of China), Meng Li (School of Information and Software Engineering, University of Electronic Science and Technology of China), Yi Ding (School of Information and Software Engineering, University of Electronic Science and Technology of China), Zhiguang Qin (School of Information and Software Engineering, University of Electronic Science and Technology of China),
Hide Authors & Abstract

Show Authors & Abstract
15:00 - 15:20
Boosting the performance of object detection CNNs with context-based anomaly detection

In this paper, we employ anomaly detection methods to enhance the ability of object detectors to consider the context for their detections. This has numerous potential applications from boosting the performance of standard object detectors, to the preliminary validation of annotation quality, and even for robotic exploration and object search. We build our method on autoencoder networks for detecting anomalies, where we do not try to filter incoming data based on anomality score as is usual, but instead, we focus on the individual features of the data representing an actual scene. We show that one can teach autoencoders about the contextual relationship of objects in images, i.e. the likelihood of co-detecting classes in the same scene. This can then be used to identify detections that do and do not fit with the rest of the current observations in the scene. We show that the use of this information yields better results than using a traditional thresholding when deciding if weaker detections are actually observed or not. The experiments performed not only show that our method significantly improves the performance of CNN object detectors, but that it can be used as an efficient tool to discover incorrectly-annotated images.
Authors: Jan Blaha (Czech Technical University in Prague), George Broughton (Czech Technical University in Prague), Tomáš Krajník (Czech Technical University in Prague),
Hide Authors & Abstract

Show Authors & Abstract

Coffee Break 15:20 - 15:55

35 minutes

Session 5: Workshop 15:55 - 17:50

15:55 - 16:15
An Efficient Approach for Parameter Learning of Bayesian Network with Multiple Latent Variables by Neural Network and P-EM

Bayesian network with multiple latent variables (BNML) is used to solve many realistic problems with unobservable features, such as disease diagnosis and preference modeling. Efficiency is the bottleneck of parameter learning of BNML from data with missing values due to the large amount of intermediate results by the classic EM based parameter learning. To overcome the efficiency bottleneck for parameter learning of BNML, we propose the clustering and P-EM based methods to improve the performance of parameter learning. Firstly, we give a clustering framework to reduce the size of parameters in an unsupervised manner by incorporating the structural information of BNML into Mixture of Generative Adversarial Network (MGAN) with Recurrent Neural Network (RNN). Then, we propose the P-EM (Parabolic acceleration of the EM algorithm) based algorithm to make parameter learning converge fast, in which geometry knowledge is adopted to obtain approximate parameters. Experiments are conducted to demonstrate the efficiency and effectiveness of our proposed methods.
Authors: Kaiyu Song (School of Information Science and Engineering, Yunnan University, Kunming, 650500, China), Kun Yue (School of Information Science and Engineering, Yunnan University, Kunming, 650500, China), Xinran Wu (School of Information Science and Engineering, Yunnan University, Kunming, 650500, China), Jia Hao (School of Information Science and Engineering, Yunnan University, Kunming, 650500, China),
Hide Authors & Abstract

Show Authors & Abstract
16:15 - 16:30
Cooperative Pollution Source Exploration and Cleanup with a Bio-inspired Swarm Robot Aggregation

Using robots for exploration of extreme and hazardous environments has the potential to significantly improve human safety. For example, robotic solutions can be deployed to find the source of a chemical leakage and clean the contaminated area. This paper demonstrates a proof-of-concept bio-inspired exploration method using a swarm robotic system based on a combination of two bio-inspired behaviors: aggregation, and pheromone tracking. The main idea of the work presented is to follow pheromone trails to find the source of a chemical leakage and then carry out a decontamination task by aggregating at the critical zone. Using experiments conducted by a simulated model of a Mona robot, the effects of population size and robot speed on the ability of the swarm was evaluated in a decontamination task. The results indicate the feasibility of deploying robotic swarms in an exploration and cleaning task in an extreme environment.
Authors: Arash Sadeghi Amjadi (Msc student of Mechanical Engineering at Middle East Technical University), Mohsen Raoufi (Middle East Technical University, Turkey), Ali Emre Turgut (Asst. Prof. Dr. at Middle East Technical University, Turkey), George Broughton (Czech Technical University in Prague), Tom Krajnik (Czech Technical University), Farshad Arvin (School of Electrical and Electronic Engineering, The University of Manchester),
Hide Authors & Abstract

Show Authors & Abstract
16:30 - 16:50
DECS: Collaborative Edge-Edge Data Storage Service for Edge Computing

With the development of IoT and 5G, data are generated by the numerous smart end devices at each moment. Simultaneously, as the improvement of the hardware's performance, computing and storage are partly transferred to the edge of the Internet. However, the core cloud and massive data centers are still responsible for management and coordination. In more and more local-area and small-scale scenarios such as a parking lot, an office building, or a college campus, these scenarios also need the edge nodes to offload computing and storage tasks. Moreover, in order to decrease costs and be lightweight, these scenarios need to decouple with the core cloud partly. In this paper, we proposed a collaborative edge-edge data storage service called DECS for edge computing in local-area scenarios. DECS can make the edge nodes collaborate with others. Such as trade-off to pick the most appropriate edge node to offload storage or computing tasks. DECS can also replicate data or generate forwarding rules in advance by predicting data's popularity proactively. In this paper, we evaluated DECS at two real scenarios compared with state-of-the-art research. The experiment results proved that DECS was more suitable for the local-area edge cluster. Which lowered the access latency, saved the total bandwidth, and improved the resource utilization of the whole edge cluster.
Authors: Fuxiao Zhou (Shanghai Jiao Tong University), Haopeng Chen (Shanghai Jiao Tong University),
Hide Authors & Abstract

Show Authors & Abstract
16:50 - 17:10
DCT: A Deep Collaborative Filtering Approach Based on Content-Text Fused for Recommender Systems

Recommender systems commonly make recommendations by means of user-item interaction ratings. One of the basic methodologies of recommendation is collaborative filtering, which exploits users’ and items’ latent space features to make predictions of personalized ranking list for an individual user. However, most of existing collaborative filtering approaches only employ explicit interaction ratings data to predict user preferences, and neglect the necessary of exploiting implicit feedback data and auxiliary information in promoting the performance for recommendation. In this paper, we raise a novel model of recommendation on the basis of neural networks architecture. Concretely, the model exploits both of interaction data and content text information as input and adopts two parallel neural networks to learn the la-tent feature representations of users and items for a better performance. We utilize three kind of real-world data to make extensive evaluations on the model. The experimental results reveal that the method we proposed dramatically outperforms the state-of-the-art methodologies and achieves expressively improvement in performance.
Authors: Zhiqiao Zhang (Chongqing University), Junhao Wen (Chongqing University), Jianing Zhou (Chongqing University),
Hide Authors & Abstract

Show Authors & Abstract
17:10 - 17:30
Real-time Self-defense Approach based on Customized Netlink Connection for Industrial Linux-based Devices

With the deep integration of IT (Information Technology) and OT (Operational Technology), various Linux operating systems have been successfully applied in critical industrial devices, such as Linux-based IIoT (Industrial Internet of Things) controllers or gateways, and the vulnerabilities of these systems may be-come a new breakthrough for the organized and high-intensity attacks. In order to prevent malwares from corrupting or disabling industrial Linux-based devices, this paper proposes a novel real-time self-defense approach, which can be easily developed without redesigning the basic software and hardware platform. By es-tablishing the customized Netlink connection between kernel mode and user mode, this approach can monitor all application processes, and block each new malicious application process, which cannot conform to the trusted white-listing rules. All experimental results show that the proposed approach has a compara-tive advantage to effectively detect and prevent the malware-related attacks, and provides a self-defense function for industrial Linux-based devices, which meets their availability due to the millisecond resolution.
Authors: Ming Wan (School of Information, Liaoning University), Jiawei Li (School of Information, Liaoning University), Jiangyuan Yao (School of Computer Science & Cyperspace Security, Hainan University),
Hide Authors & Abstract

Show Authors & Abstract
17:30 - 17:50
Budget Constraint Task Allocation for Mobile Crowd Sensing with Hybrid Participant

In the Mobile crowdsensing (MCS) system, the task allocation problem is a crucial problem. In this paper, we focus on the task allocation problem for hybrid participants with a budget-constrained in the MCS system. There are two types of participants: the mobile participants and static participants. Moblie participants have low cost, large numbers, and flexibility. However, most of the sensing data they submitted are of low quality. On the other hand, static participants, such as city cameras, roadside infrastructure, have high-quality sensing data. Despite the benefit of high quality, static participants have less coverage and high cost. Given a budget, the problem is how to assign the task to the two types of participants, such that the social welfare is maximized. To solve the problem, we propose a reverse auction-based task allocation method to select winning bids round by round. Then, a Shapley value-based online algorithm is proposed to ensure the task finished. Moreover, we consider the different types of participants to have a different probability to finish tasks. We exploit the semi-Markov model to calculate the probability that participants finished tasks. We prove that the proposed task allocation method has truthfulness, individual rationality, and computational efficiency. We conduct extensive experiments to evaluate the performance of our system, and the evaluation results demonstrate the remarkable effect of the proposed task allocation method.
Authors: Xin Wang (Wuhan University of Science and Technology), Peng Li (Wuhan University of Science and Technology), Junlei Xiao (Wuhan University of Science and Technology),
Hide Authors & Abstract

Show Authors & Abstract
Room #3

Session 3: Cloud & Edge Computing 13:30 - 15:25

13:30 - 13:50
EFMLP: A Novel Model for Web Service QoS Prediction

With the emergence of service-oriented architecture, quality of service (QoS) has become a crucial factor in describing the non-functional characteristics of Web services. Therefore, how to predict QoS values for Web services has become a hot research topic. In the real world, the user only requests limited Web services, the QoS record of Web services are sparsity. In this paper, we propose an ap-proach named factorization machine and multi-layer perceptron model based on embedding technology (EFMLP) to solve the problem of sparsity and high di-mension. First, the input data will be sent to embedding layer to reduce the data dimension and keep the information. Then, the feature vector which has been em-bedded will send to the factorization machine, after that, the first-order weights and second-order weights of the factorization machine are used as the initial weights of the first layer of the multi-layer perceptron. Then we train the multi-layer perceptron for adjusting the weights. Finally, an experiment using 1974675 pieces of data from an open data set to validate the model, and the result shows that our EFMLP model can predict QoS value accurately on the client side.
Authors: kailing ye, Huiqun Yu (Dept of Computer Science and Engineering,East China University of Science and Technology),
Hide Authors & Abstract

Show Authors & Abstract
13:50 - 14:10
The Design and Implementation of Secure Distributed Image Classification Model Training System for Heterogenous Edge Computing

Deep learning provides many new and efficient solutions for edge computing. We study training image classification models on edge devices in this paper. Although there have been some researches on deep learning in edge computing. Most of them did not consider the impact of the limited service capabilities of edge devices, the problem of straggler and the insecurity of training data on the system. We design a new distributed computing system to train image classification models on edge devices. To be more specific, we vectorize the convolutional neural network(CNN) to transform it to a lot of matrix multiplications. These matrix multiplications are arbitrarily cut into many smaller matrix multiplications suitable for computing on edge devices. Besides, our system utilizes codes to ensure the stability and security of distributed matrix multiplications on edge devices. In the performance evaluation, we test the performance of matrix multiplications and a CNN model training in our system with uncoded and coded strategies. The evaluation results show that the system with code strategies perform better than with uncode strategies on the edge devices having the problem of straggler. In summary, we design a secure distributed image classification model training system for heterogenous edge computing.
Authors: Cong Cheng, Huan Dai (Suzhou University of Science and Technology), Lingzhi Li (Soochow University), Jin Wang (Soochow University), Fei Gu (Soochow University),
Hide Authors & Abstract

Show Authors & Abstract
14:10 - 14:30
HIM: A Systematic Model to Evaluate the Health of Platform-based Service Ecosystems

With the vigorous development of the platform-based service ecosystem represented by e-commerce, service recommendation is used as a personalized matching method. There exist some service recommendation strategies that mainly focus on high popularity services and ignore non-popular ones. This will not only lead to oligopoly, but also be detrimental to the health of the platform-based service ecosystem. In addition, the health evaluation indicators for this kind of ecosystems are mostly qualitative and single. In view of the above phenomenon and based on the system view of balance and health, a health index model (HIM) is proposed to measure the health from two aspects quantitatively: stability and sustainability. Specifically, the model includes the system activity and organizational structure reflecting stability, as well as the productivity and vitality reflecting sustainability, which helps to illustrate the health status of the platform-based service ecosystem from the perspective of multi-dimensional integration. Additionally, this paper analyzes the factors affecting the health of this ecosystem based on HIM. In this work, a platform-based service ecosystem simulation model is constructed by using the computational experiment method to verify the effectiveness of HIM. The simulation results show that the HIM can reasonably measure the health of such ecosystems, which has guiding significance for the overall management and sound development of e-commerce platforms.
Authors: Yiran Feng (Tianjin University/School of Computer Software), Zhiyong Feng (Tianjin University/School of Computer Software), Xiao Xue (Tianjin University/School of Computer Software), Shizhan Chen (Tianjing University/ School of Computer Science and Technology),
Hide Authors & Abstract

Show Authors & Abstract
14:30 - 14:50
SETE: A Trans-boundary Evolution Model of Service Ecosystem Based on Diversity Measurement

Trans-boundary and integration are important characteristics of the development of modern service industry. With the development of Internet technology, trans-boundary cooperation between domains constantly emerges, which drives the development of service ecosystem. Currently, there is a lack of an appropriate model for analyzing the impact of trans-boundary services on the entire service ecosystem. In this paper, we propose a service ecosystem trans-boundary evolution model (SETE). It analyzes the interactions between user needs and services, and focuses on the mechanism of trans-boundary services to promote the evolution of the service ecosystem. At the same time, we develop a diversity measurement algorithm for service ecosystems based on the theory of biodiversity in ecology. Based on these, a computational experimental system is established. It simulates the trans-boundary evolution mechanism of the service ecosystem and shows the characteristics of each stage of the service ecosystem evolution. At last, we verify the effectiveness of the SETE model through actual cases (the Alibaba Group). The results show that the SETE model can provide new ideas for the study of the trans-boundary evolution of the service ecosystem, and provide decision support for the development direction of the modern service industry.
Authors: Tong Gao (Tianjin University/College of Intelligence and Computing), Zhiyong Feng (Tianjin University/College of Intelligence and Computing), Shizhan Chen (Tianjin University/College of Intelligence and Computing), Xiao Xue (Tianjin University/College of Intelligence and Computing),
Hide Authors & Abstract

Show Authors & Abstract
14:50 - 15:10
A DNN Inference Acceleration Algorithm in Heterogeneous Edge Computing: Joint Task Allocation and Model Partition

Edge intelligence, as a new computing paradigm, aims to allocate Artificial Intelligence (AI)-based tasks partly on the edge to execute for reducing latency, consuming energy and improving privacy. As the most important technique of AI, Deep Neural Networks (DNN) has been widely used in various fields. And for those DNN based tasks, a new computing scheme named DNN model partition can further reduce the execution time. This computing scheme partition the DNN task into two parts, one will be executed on the terminal devices and the other will be executed on edge servers. However, in a complex edge computing system, it is difficult to coordinate DNN model partition and task allocation. In this work, we study this problem in the heterogeneous edge computing system. We first establish the mathematical model of adaptive DNN model partition and task offloading. The mathematical model contains a large number of binary variables, and the solution space will be too large to be solved directly in a multi-task scenario. Then we use dynamic programming and greedy strategy to reduce the solution space under the premise of a good solution, and propose our offline algorithm named GSPI. Then considering the actual situation, we subsequently proposed the online algorithm. Through our experiments and simulations, we proved that our proposed GSPI algorithm can reduce the system time cost by at least 32% and the online algorithm can reduce the system time cost by at least 24%.
Authors: Lei Shi (Hefei University of Technology), Zhigang Xu (Hefei University of Technology), Yi Shi (Intelligent Automation Inc.), Yuqi Fan (Hefei University of Technology), Xu Ding (Hefei University of Technology), Yabo Sun (Hefei University of Technology),
Hide Authors & Abstract

Show Authors & Abstract
15:10 - 15:25
A Novel Probabilistic-Performance-Aware and Evolutionary Game-theoretic Approach to Task Offloading in the Hybrid Cloud-Edge Environment

The mobile edge computing (MEC) paradigm provides a promising solution to solve the resource-insufficiency problem in mobile terminals by offloading computation-intensive and delay-sensitive tasks to nearby edge nodes. However, pure edge resources can be limited and insufficient for computational-intensive applications raised by multiple users, which calls for a hybrid architecture with a centralized cloud service and multiple edge nodes and smart resource management strategies in such hybrid environment. The problem is however challenging due to the distributed nature and intrinsic dynamicness of the environment. Existing researches in this direction mainly see that edge servers are with constant performance and consider the offloading decision-making as a static optimization problem. In this paper, instead, we consider that geographically distributed edge servers are with time-varying performance and introduce a dynamic offloading strategy based on a probabilistic evolutionary game-theoretic framework. To validate our proposed framework, we conduct experimental case studies based on a real-world dataset of cloud edge resource locations and show that our proposed approach outperforms traditional ones in terms of multiple metrics.
Authors: Ying Lei (Chongqing University), Wanbo Zheng (Kunming University of Science and Technology), Yunni Xia (Chongqing University), Yong Ma (Jiangxi Normal University), Qing Xia (Chongqing Key Laboratory of Smart Electronics Reliability Technology),
Hide Authors & Abstract

Show Authors & Abstract

Coffee Break 15:25 - 16:00

35 minutes

Session 6: Cloud & Edge Computing; Workshop 16:00 - 17:55

16:00 - 16:25
Location-Aware MEC Service Migration for Mobile User Reallocation in Crowded Scenes

Mobile edge computing(MEC) paradgim is evolving as an increasingly popular means for developing and deploying smart-city-oriented applications. MEC servers can receive a great deal of requests from equipments of highly mobile users,especially in crowded scenes,e.g., city's central business district (CBD) and school areas.It thus remains a great challenge for appropriate scheduling and managing strategies to avoid hotspot, guarantee load-fairness among MEC servers, and maintain high resource utilization at the same time. To address this challenge, we propose a coalitional-game-based and location-aware approach to MEC Service migration for mobile user reallocation in crowded scenes. Our proposed method includes multiple steps: 1)dividing MEC servers into multiple coalitions according to their inter-euclidean distance by using a modified $k$-means clustering method;2)discovering hotspots in every coalition area and scheduling services based on their corresponding cooperations; 3)migrating services to appropriate edge servers to achieve load-fairness among coalition members by using a migration budget mechanism; 4) transferring workloads to nearby coalitions by backbone network in case of extrahigh workloads. Experimental results based on a real-world mobile trajectory dataset for crowded scenes, and an urban-edge-server-position dataset demonstrate that our method outperforms existing approaches in terms of load-fairness,migration times,and energy consumption of migrations.
Authors: Xuan Xiao (chongqing university), Yunni Xia (Chongqing University), Yong Ma (Jiangsu Normal University), Yin Li (Institute of Software Application Technology, Guangzhou & Chinese Academy of Sciences), Chunxu Jiang (Chongqing Key Laboratory of Smart Electronics Reliability Technology), Xingli Zhong (CISDI R &D co. Ltd),
Hide Authors & Abstract

Show Authors & Abstract
16:25 - 16:45
A New Collaborative Scheduling Mechanism Based on Grading Mapping for Resource Balance in Distributed Object Cloud Storage System

An algorithmic mapping of storage locations brings high storage efficiency to the storage system and also unbalances resource storage due to loss of efficient scheduling between storage nodes, which makes the system prone to crash at low usage. This paper uses the Ceph storage system as a research sample to analyze these issues and proposes a grading mapping adaptive storage resource collaborative optimization mechanism. This approach graded both the storage device and the storage content and introduced random factors and influence factors as two-factors to quantify the grading mapping relationship between the two of them. This relation coordinates the storage systems' performance and reliability. In addition, a collaborative storage algorithm is proposed in order to realize balanced storage efficiency and control data migration. The experimental results show that in comparison with the inherent mechanism in the traditional Ceph system, the proposed cooperative storage adaptation mechanism exhibits a more substantial, stabilizing effect on data balance and has the ability to control data migration.
Authors: yu lu, ningjiang chen (-), wenjuan pu (-), ruifeng wang (-),
Hide Authors & Abstract

Show Authors & Abstract
16:45 - 17:05
User Perspective Discovery Method Based on Deep Learning in Cloud Service Community

The rapid development of cloud computing has promoted the coordinated integration of resources in various industries. In order to facilitate users’ selection and invocation, more and more individuals and organizations have moved local application resources into the cloud service communities in the form of web services. In recent years, more and more people are interested in the emotional attitudes reflected in consumer reviews, but the sentiment analysis using the deep learning method to achieve evaluation of API (Application Programming Interface) services has received little attention. In order to explore the effective information of user's point of view data in the cloud service community, we propose an approach to analyze the user's opinion data using deep learning. We design three deep learning models of Long Short-Term Memory (LSTM), Bi-directional Long Short-Term Memory (Bi-LSTM) and Gated Recurrent Unit (GRU). The result shows that the accuracy rate and recall rate of Bi-LSTM model is higher than the LSTM and GRU. Finally, we evaluate the performance of the three deep learning models, and choose the optimal Bi-LSTM model as the model used by the cloud service community in the future. According to the parameter comparison experiment of Bi-LSTM model, we obtained the optimal tuning of the model, and the model achieved the accuracy of 89.68%.
Authors: lei yu (Inner mongolia university), yaoyao wen (Inner mongolia university), shanshan liang (inner mongolia university),
Hide Authors & Abstract

Show Authors & Abstract
17:05 - 17:25
Bidding Strategy Based on Adaptive Differential Evolution Algorithm for Dynamic Pricing IaaS Instances

In recent years, with the development of cloud computing technology and the improvement of infrastructure performance, cloud computing has devel-oped rapidly. In order to meet the diverse needs of users and to maximize the revenue of cloud computing service providers, cloud providers have launched auction-type instances like Amazon Spot instances in the AWS cloud. For dynamic pricing cloud instances, how to select appropriate instance or in-stance group among multiple instances and make reasonable bids to optimize its own costs is a great challenge. This paper models the dynamic pricing in-stance pricing and multi-instance combination problem as a constrained op-timization problem. Then we introduce the basic differential algorithm and proposes an adaptive differential evolution algorithm to optimize the combi-nation of price bidding based on the optimal cost and the use of instances. Finally, we use real dynamic pricing instance price data released by Amazon cloud to verify the optimization strategy. The experimental results show that the adaptive differential evolution algorithm has a better optimization effect on short-term task requirements and long-term task requirements. The adap-tive differential evolution algorithm has an average optimization cost of 12.48% over the maximum bidding strategy pair and an average optimization cost of 3.53% over the basic differential evolution algorithm.
Authors: Dawei Kong (School of software, Shandong University), Guangze Liu (Pennsylvania State University), Li Pan (School of software, Shandong University), Shijun Liu (School of software, Shandong University),
Hide Authors & Abstract

Show Authors & Abstract
17:25 - 17:40
SBiNE: Signed Bipartite Network Embedding

This work develops a representation learning method for signed bipartite networks. Recent years, embedding nodes of a given network into a low dimensional space has attracted much interest due to it can be widely applied in link prediction, clustering, and anomalous detection. Most existing network embedding methods mainly focus on homogeneous networks with only positive edges and single node type. However, negative edges are more valuable than positive edges in certain analysis tasks. Even though the work on signed network representation learning distinguishes between positive and negative edges, it does not consider the difference in node types. However, bipartite network representation learning which considers two types of vertices do not tell link signs. In order to solve this problem, we further consider the link sign on the basis of the bipartite network to conduct signed bipartite network analysis. In this paper, we propose a simple deep learning framework SBiNE, short for signed bipartite network embedding, which both preserves the first-order(i.e., observed links) and second-order proximity(i.e., unobserved links but have similar sign context), and then by optimizing the objective function, experiments on three datasets show that our proposed framework SBiNE is competitive in link sign prediction task.
Authors: Youwen Zhang (School of Computer Science and Engineering, Anhui University), Wei Li (School of Computer Science and Engineering, Anhui University), Dengcheng Yan (School of Computer Science and Engineering, Anhui University), Yiwen Zhang (School of Computer Science and Engineering, Anhui University), Qiang He (School of Software and Electrical Engineering, Swinburne University of Technology),
Hide Authors & Abstract

Show Authors & Abstract
17:40 - 17:55
Usable and Secure Pairing based on Handshake for Wrist-Worn Smart Devices on Different Users

Wrist-worn smart devices are being used to share various sensitive personal information in various fields such as social, medical, sports, etc. Secure pair-ing establishing a trusted channel between the involved devices is a prerequi-site to ensure data transmission security. Handshake has been proposed to realize secure pairing between devices worn by different users without pre-shared knowledge, the participation of third parties or complex user interac-tions. However, existing schemes cannot meet the practical requirement in terms of time delay and security. In this paper, we proposed a user friendly handshake based secure pairing scheme, which utilizes the handshake accel-eration data. Specifically, we quantify the features through random threshold values, to shorten the handshake time required for guaranteeing the length of the negotiated key. Besides, we propose an optimal feature selection algo-rithm that improves the success rate and security of the system. Experiments are performed on our scheme, which show that the proposed scheme is ro-bust and secure. Users only need to take a few seconds to perform simple operations, and the devices can automatically pair securely.
Authors: Xiaohan Huang (Xidian University), Guichuan Zhao (Xidian University), Qi Jiang (Xidian University), Youliang Tian (Guizhou University), Xindi Ma (Xidian Unversity), Jianfeng Ma (Xidian University),
Hide Authors & Abstract

Show Authors & Abstract
Day 2 17/10/2020
Room #1

Session 7: Classification and Recommendation 08:30 - 10:25

08:30 - 08:50
SC-GAT: Web Services Classification Based on Graph Attention Network

The classification of Web services with high similarity is conducive to the promotion for service management and service discovery. With the increasing number of Web services, how to accurately and efficiently classify the Web services becomes a challenging task. Through the existing methods achieve significant results in service classification via integrating the structure information of service network with the content features of service node, without discriminating the importance of neighbor services in the service network on the service node needed to be classified. To solve this problem, we propose a Web services classification method based on graph attention network. Firstly, according to the composition and shared annotation relationship of Web services, it applies the description documents, tags of Web services and the call relationship between Mashups and services to build a service relationship network. Then, the attention coefficient of service nodes in the network is calculated by the selfattention mechanism, and different service nodes in the neighborhood are assigned different weights to classify Web services. Through the graph attention network, the content features of Web service can be well integrated with its structure information. The experimental results on the real dataset of ProgrammableWeb platform show that the precision, recall and macro-F1 of the proposed method are greatly improved compared to those of GCN, Node2vec, DeepWalk and Line.
Authors: Mi Peng (Hunan University of Science and Technology), Buqing Cao (Hunan University of Science and Technology), Junjie Chen (Hunan University of Science and Technology), Jianxun Liu (Hunan University of Science and Technology), Bing Li (Hunan University of Science and Technology),
Hide Authors & Abstract

Show Authors & Abstract
08:50 - 09:05
A Novel Multidimensional Comments and Co-preference Aware Recommendation Algorithm

A recommendation system creates a personalized experience for each customer, which helps companies boost the average order value and the amount of revenue generated from each customer. In a typical recommendation system, comments typify the group wisdom of users, which can reflect their feelings toward the product in multiple dimensions. Co-preference mirrors common preference of a group of users. By mining the multidimensional comments and co-preference relationship comprehensively, it is justifiable to recommend products that both have a good reputation and conform to users’ interests. However, the existing related methods have two problems. Firstly, there is lack of further consideration on how to fully utilize comments of products from multiple dimensions for recommendation. Secondly, how to mine co-preference relationship and combine it with multidimensional comments for recommendation is seldom considered. Therefore, a novel recommendation algorithm is proposed, which mines the comments from multiple dimensions and then converges it with co-preference relationship for recommendation. Experiments conducted on two real-world datasets reveal that our proposed method improves the accuracy in terms of MAE and RMSE, compared with state-of-the-art algorithms.
Authors: Nana Song (Central University of Finance and Economics), Yanmei Zhang (Central University of Finance and Economics), Xiaoyi Tang (Central University of Finance and Economics), Huaihu Cao (Central University of Finance and Economics),
Hide Authors & Abstract

Show Authors & Abstract
09:05 - 09:25
A Deep Recommendation Framework for Completely New Users in Mashup Creation

When service business is in evolution from B2B to B2C model, a cold-start problem raises for service composition due to the completely new clients with no historical records. Therefore, it is of great importance to solve the cold-start problem brought by completely new users. In this paper, we pro-pose a recommendation framework for completely new users in Mashup creation based on deep-learning technology. Firstly, this framework extracts the mapping relationship between Mashup description and APIs offline by the deep neural network. Then, when the completely new users have the Mashup demands online, the matching APIs are recommended for them by using the mapping relationship. The experimental results with real-world da-tasets show that our proposed model outperforms the start-of-the-art ones in term of both accuracy and recall rate. The accuracy of the proposed method is 1.34 times higher than that of the state-of-the-art methods, and the recall rate is 1.55 times higher than that of the state-of-the-art methods. Moreover, considering that the new user history invocational data is very sparse, the performance of the proposed method can be greatly improved on the denser dataset.
Authors: Yanmei Zhang (Information School Central University of Finance and Economics), Jinglin Su (Information School Central University of Finance and Economics), Shiping Chen (Software and Computational Systems CSIRO Data61),
Hide Authors & Abstract

Show Authors & Abstract
09:25 - 09:45
Combining Feature Selection Methods with BERT: An In-depth Experimental Study of Long Text Classification

With the introduction of BERT by Google, a large number of pre-training models have been proposed. Using pre-training models to solve text classification problems has become the mainstream. How- ever, the complexity of BERT grows quadratically with the text length, hence BERT is not suitable for processing long text. Then the researchers proposed a new pre-training model XLNet to solve the long text classifi- cation problem. But XLNet requires more GPUs and longer fine-tuning time than BERT. To the best of our knowledge, no attempt has been done before combining traditional feature selection methods with BERT for long text classification. In this paper, we use the classic feature selec- tion methods to shorten the long text and then use the shortened text as the input of BERT. Finally, we conduct extensive experiments on the public data set and the real-world data set from China Telecom. The ex- perimental results prove that our methods are effective for helping BERT to process long text.
Authors: Kai Wang (Zhejiang University of Technology), Jiahui Huang (Zhejiang University of Technology), Yuqi Liu (Zhejiang University of Technology), Bin Cao (Zhejiang University of Technology), Jing Fan (Zhejiang University of Technology),
Hide Authors & Abstract

Show Authors & Abstract
09:45 - 10:05
A novel approach for seizure classification usingpatient specific triggers

The most promising method of non-EEG detection uses a non-invasive multi sensor modality that applies pattern recognition techniques to analyse a patient’s physiological signals. The purpose of this study is to identify whether patient specific triggers (PST) can be used to train a classification model and add a new untested modality to the field of non-EEG seizure detection. Method:For this study we collected data from 2 participants with epilepsy using a Fitbit, smartphone and quantitative questionnaires. Several classification models were then trained, based on different algorithms and techniques.Results: Results show the multi-layer perceptron is the model best suited for this type of classification task with Accuracy scores of 94.73%, 96.87%, 94.11% for datasets D1, D2 and D3 respectively. Conclusion: Our findings indicate that patients were more likely to have a seizure if they had < 6-8 hours of sleep and a stress or fatigue factor ≥ 3.5, indicating that PST can be used as an additional modality for non-EEG seizure classification.
Authors: Jamie Pordoy, Ying Zhang (University of West London),
Hide Authors & Abstract

Show Authors & Abstract
10:05 - 10:25
A unified Bayesian model of community detection in attribute networks with power-law degree distribution

Detecting community structure is an important research topic in complex network analysis. How to improve community detection results by using various features in the network is a very challenging problem. The scale-free and attributes of nodes are the two relatively independent aspects of the complex networks in real world, the former is an inherent structural feature from the global perspective and the later can be used to significantly enhance community detection and community semantics. However, these two aspects are usually modeled and computed independently in previous methods. Based on that, we propose a novel unified Bayesian generative model which combines network topology and node attributes simultaneously to identify community structures via considering to model the scale-free feature. We propose the degree decay variable to preserve the power-law degree characteristic of the network. Specifically, this model composes of two closely correlated parts by a probabilistic transition matrix, one for network topology and the other for nodes attributes. Moreover, we develop a variational EM algorithm to optimize the objective function of the model. Experiments on synthetic and real networks show that our model has better performance compared with some baselines on community detection in attribute networks.
Authors: Zhang Shichong (College of Intelligence and Computing, Tianjin University), Wang Yinghui (College of Intelligence and Computing, Tianjin Universit), Wang Wenjun (College of Intelligence and Computing, Tianjin University), Jiao Pengfei (Center for Biosafety Research and Strategy, Law School, Tianjin University), Lin Pan (School of Marine Science and Technology, Tianjin University),
Hide Authors & Abstract

Show Authors & Abstract

Coffee Break 10:25 - 11:00

35 minutes

Session 10: Resource Management 11:00 - 12:55

11:00 - 11:20
Mobile Edge Server Placement Based on Bionic Swarm Intelligent Optimization Algorithm

By offloading computing tasks from mobile devices to edge servers with sufficient computing resources, network congestion and data propagation delays can be effectively reduced. The placement of edge servers is the core of task offloading and is a multi-objective optimization problem with multiple resource constraints. An optimization model of edge server placement has been established in this paper by minimizing both access delay and workload difference as the optimization goal. Then, based on Glowworm Swarm algorithm, it proposes a mobile edge server placement approach called GSOESP to achieve a multi-objective optimization goal. In this study, we uses the improved Glowworm Swarm Optimization (GSO) algorithm to find the optimal places as the clustering center which is the edge server placement address, and every base station in edge server's neighbour list is allocated to the edge server. After many iterations, we gradually approaches the optimal target. So, the optimal placement scheme is obtained to achieve the goals of minimizing the distance for users to access the edge server and balancing the workload. The GSOESP algorithm is similar to a fast clustering algorithm with good time performance. Experimental results using Shanghai Telecom's real dataset show that the proposed approach achieves an optimal balance between low latency and workload balancing, while guaranteeing service quality, which outperforms several existing representative approaches.
Authors: Feiyan Guo (School of Computer Science and Engineering, Hunan University of Science and Technology, China), Bing Tang (School of Computer Science and Engineering, Hunan University of Science and Technology, China), Linyao Kang (School of Computer Science and Engineering, Hunan University of Science and Technology, China), Li Zhang (School of Computer Science and Engineering, Hunan University of Science and Technology, China),
Hide Authors & Abstract

Show Authors & Abstract
11:20 - 11:40
A MOEAD-based Approach to Solving the Staff Scheduling Problem

Numerous works have been studied on the staff scheduling problem. They often aim to maximize the optimization objectives, and ignore the fairness of their values. Hence, we propose a MOEAD-based approach to address our issue by encoding the individual and population, setting the constrained control of schedules, and selecting feasible schedules. A series of experiments are performed and prove that the proposed method can effectively find the schedule with fairness.
Authors: Hong Feng (Zhejiang University of Technology), hao chen (Zhejiang University of Technology), Bin Cao (Zhejiang University of Technology), Jing Fan (Zhejiang University of Technology),
Hide Authors & Abstract

Show Authors & Abstract
11:40 - 11:55
A Hybrid Collaborative Virtual Environment with Heterogeneous Representations for Architectural Planning

We developed a collaborative virtual environment for architectural planning that facilitates groups of people to work together using three different interfaces that enable users to interact with the scene with varying levels of immersion and different interaction modalities. We conducted a user study to gauge the general usability of the system and to understand how the different interfaces affect the group work. In this paper we present the architecture of the system along with its different interfaces. We also present the user study results and the insights we gained from the study.
Authors: Krishna Bharadwaj (University of Illinois at Chicago), Andrew Johnson (University of Illinois at Chicago),
Hide Authors & Abstract

Show Authors & Abstract
11:55 - 12:15
A deep reinforcement learning based resource autonomic provisioning approach for cloud services

Resource elastic scheduling is a key feature of cloud services. The elastic makes cloud services have the ability to flexibly increase or decrease resources to satisfy user needs, and dynamically allocate resources for cloud services on demand. The amount of resources to be configured is determined at runtime based on the changes in workload to flexibly respond to the fluctuating demands of cloud ser-vices. Appropriate resources need to be configured in advance. In this article, we propose a dynamic resource provisioning framework based on the MAPE loop, and use a two-tier resource elastic configuration for collaborative work. In order to implement the proposed framework, we propose a resource elastic scheduling algorithm based on a combination of the autonomic computing and deep rein-forcement learning (DRL) to reduce task rejection rate of the virtual machine (VM) and increase utilization to obtain as much profit as possible. In this paper, Experimental results using actual Google cluster tracking results show that the proposed policy reduces the total cost about 17%-58% and increases the profit by up to not less than 9%, reduces the service level agreement (SLA) violations to less than 0.4% to better guarantee the quality of service (QoS).
Authors: Qing Zong (Shandong Normal University), Xiangwei Zheng (Shandong Normal University), Yi Wei (Shandong Normal University), Hongfeng Sun (Shandong Women’s University),
Hide Authors & Abstract

Show Authors & Abstract
12:15 - 12:35
End-to-end QoS Aggregation and Container Allocation for Complex Microservice Flows

Microservice is increasingly seen as a rapidly developing architectural style that uses containerization technology to deploy, update, and scale independently and quickly. A complex microservice flow that is composed of a set of microservices can be characterized by a complex request execution path spanning multiple microservices. It is essential to aggregate quality of service (QoS) of individual microservice to provide overall QoS metrics for a complex microservice flow. Besides, leveraging the cost and performance of a complex microservice flow to find an optimal end-to-end container allocation solution with QoS guarantee is also a challenge. In this paper, we define an end-to-end QoS aggregation model for the complex microservice flow, and formulate the end-to-end container allocation problem of microservice flow as a nonlinear optimization problem, and propose an ONSGA2-DE algorithm to solve this problem. We comprehensively evaluate our modeling method and optimization algorithms on the open-source microservice benchmark Sock Shop. The results of experiments show that our method can effectively assist in the QoS management and container allocation of complex microservice flow.
Authors: Min Zhou (School of Big Data and Software Engineering, Chongqing University), Yingbo Wu (School of Big Data and Software Engineering, Chongqing University), Jie Wu (School of Big Data and Software Engineering, Chongqing University),
Hide Authors & Abstract

Show Authors & Abstract
12:35 - 12:55
A DQN-based Approach for Online Service Placement in Mobile Edge Computing

Due to the development of 5G networks, computation intensive applications on mobile devices have emerged, such as augmented reality and video stream analysis. Mobile edge computing is put forward as a new computing paradigm, to meet the low-latency requirements of applications, by moving services from the cloud to the network edge like base stations. Due to the limited storage space and computing capacity of an edge server, service placement is an important issue, determining which services are deployed at edge to serve corresponding tasks. The problem becomes particularly complicated, with considering the stochastic arrivals of tasks, the additional latency incurred by service migration, and the time spent for waiting in queues for processing at edge. Benefiting from reinforcement learning, we propose a deep Q network based approach, by formulating service placement as a Markov decision process. Real-time service placement strategies are output, to minimize the total latency of arrived tasks in a long term. Extensive simulation results demonstrate that our approach works effectively.
Authors: Xiaogan Jie (Shanghai University), Tong Liu (Shanghai University), honghao gao (shanghai university), Chenhong Cao (Shanghai University), Peng Wang (Shanghai University), Weiqin Tong (Shanghai University),
Hide Authors & Abstract

Show Authors & Abstract

Lunch Break 12:55 - 13:25

30 minutes

Session 12: Software & Security 13:25 - 15:20

13:25 - 13:40
Towards a Trusted Collaborative Medical Decision-Making Platform

Traditionally, shared decision-making has been considered as a one-off dyadic encounter between a patient and physician within the confines of the consultation room. In practice, several stakeholders are involved, and the decision-making process involves multiple rounds of interaction and can be influenced by different biopsychosocial, cultural, spiritual, financial, and legal determinants. Thus, there is an opportunity for developing an innovative digital platform for distributed collaborative medical decision-making. However, given the sensitive nature of data and decisions, there are several challenges associated with safeguarding the consent, privacy, and security of the contributors to the decision- making process. In this paper, we propose a conceptual framework and reference architecture for a trusted collaborative medical decision-making (TCMDM) platform that addresses some of these challenges by using a consensus building mechanism built on top of blockchain technology. We illustrate how the TCMDM platform functions by using a real-life use case scenario of an early stage breast cancer patient collaborating with other stakeholders to reach consensus on the best treatment plan.
Authors: Hamza Sellak (CSIRO's Data61), Mohan Baruwal Chhetri (CSIRO's Data61), Marthie Grobler (CSIRO's Data61),
Hide Authors & Abstract

Show Authors & Abstract
13:40 - 14:00
Code Prediction Based on Graph Embedding Model

Code prediction aims to accelerate the efficiency of program- mer development. However, accuracy of prediction is still a great chan- llage. To facilitate the interpretability of the code prediction model and improve the accuracy of prediction. In this paper, the source code’s AST(Abstract Syntax Tree) is used to extract relevant structural paths between nodes and convert them into training graphs. The embedded model is used to convert the structural features of nodes into vectors that are convenient to quantify. Taking the node’s vector similarity as a measurement criterion, the similarity between the candidate value node and the parent node vector of the predicted path is calculated. Finally complete the prediction task of TYPE and VALUE. Experiments show that by using prediction data to increase the weight of related nodes in the graph, the model can extract more useful structural features, es- pecially in Value prediction tasks. The adjustment of the parameters embedded in different graphs can improve the accuracy of the model, and the response of the TYPE prediction task is obvious.
Authors: Kang Yang (Department of Computer Science and Engineering, East China University of Science and Technology), Huiqun Yu (Dept of Computer Science and Engineering,East China University of Science and Technology), Fan Guisheng (East China University of Science amd Technology), Xingguang Yang (Dept of Computer Science and Engineering,East China University of Science and Technology),
Hide Authors & Abstract

Show Authors & Abstract
14:00 - 14:20
Differentially Private Location Protection with Staircase Mechanism under Temporal Correlations

Location-Based Service (LBS) is one of basic services in collaborative appli-cations. However, LBS applications may discloses user’s location privacy, which receives considerable concern. Many methods have been proposed to protect privacy in LBS. Planar Isotropic Mechanism (PIM) is a typical loca-tion privacy preservation method in the scenario of continuous location data release. However, the method is complicated, since it consists of the K-norm mechanism and two convex hull transformation. To solve the problem, we propose a Staircase Mechanism (SM) based location privacy protection method for the scenario of continuous location data release. The proposed method replaces PIM with SM, which makes it easy to implementation and efficient. Furthermore, SM can achieve the same privacy budget with less noise addition, so it can maintain higher quality of services in LBS. Compre-hensive experiments conducted on real location data set demonstrates that the proposed method is efficient and high data utility compared with PIM.
Authors: Rong Fang (Zhejiang Normal University), Jianmin Han (Zhejiang Normal University), Juan Yu (Zhejiang Normal University), Xin Yao (Zhejiang Normal University), Hao Peng (Zhejiang Normal University), Jianfeng Lu (Zhejiang Normal University),
Hide Authors & Abstract

Show Authors & Abstract
14:20 - 14:40
Technical Implementation Framework of AI Governance Policies for Cross-modal Privacy Protection

With the development of virtual community among online users, virtual community groups have become a small society, which can extract the types of privacy related to users through the "virtual traces" left by users' browsing and user-generated content user posted. The privacy type resources can be classified into data resources, information resources and knowledge resources according to their characteristics, which constitute the data-information-knowledge and wisdom Graph (DIKW Graph) of user. There are four circulation processes for privacy type resources in virtual communities,  sensing, storage, transfer, and processing; and there are three circulation participants in the circulation processes, including resource generator, resource communicator, and resource acquirer ; the privacy rights of the participants in the circulation processes include the right to know, the right to participate, the right to be forgotten, and the right to supervise. By clarifying the scope of privacy rights of the three participants in the four circulation processes and combining the value of privacy resources, an AI governance legal framework for privacy protection of virtual communities is established.
Authors: Yucong Duan (Hainan University), Yuxiao Lei (Hainan University),
Hide Authors & Abstract

Show Authors & Abstract
14:40 - 15:00
Defending Use-after-free via Relationship Between Memory and Pointer

Existing approaches to defending Use-After-Free (UAF) exploits are usually done using static or dynamic analysis. However, both static and dynamic analysis su er from intrinsic de ciencies. The existing static analysis is limited in handling loops, optimization of memory representation. The existing dynamic analysis, which is characterized by lacking the maintenance of pointer information, may lead to flaws that the relationships between pointers and memory cannot be precisely identi ed. In this work, we propose a new method called UAF-GUARD without the above barriers, in the aim to defending against UAF exploits using fi ne-grained memory permission management. In particular, we design a key data structure to support the fine-grained memory permission management, which can maintain more information to capture the relationship between pointers and memory. Moreover, we design code instrumentation to enable UAF-GUARD to precisely locate the position of UAF vulnerabilities to further terminate malicious programs when anomalies are detected. We implement UAF-GUARD on a 64-bit Linux system. We carry out experiments to compare UAF-GUARD with the main existing approaches. The experimental results demonstrate that UAF-GUARD is able to effectively and eciently defend against three types of UAF exploits with acceptable space overhead and time overhead.
Authors: Guangquan Xu (Tianjin University), Miao Li (Tianjin University), Xiaotong Li (Tianjin University), Kai Chen (Institute of Information Engineering, Chinese Academy of Sciences), Ran Wang (security center, JD.com), Wei Wang (Beijing Jiaotong University), Kaitai Liang (Surrey Centre of Cyber Security, University of Surrey, U.K.), Qiang Tang (New Jersey Institute of Technology), Shaoying Liu (Hosei Univesity),
Hide Authors & Abstract

Show Authors & Abstract
15:00 - 15:20
API Misuse Detection based on Stacked LSTM

In modern software engineering, API (Application Programming Interface) is widely used to develop applications rapidly by reusing data structure, frameworks, class libs, and etc. However, due to the considerable number of interfaces, lack of documents and timely maintenance and updates, APIs are often used in a wrong way. Therefore, it has become an important problem to detect API misuse in an automatic way. Many existing automatic API detecting methods do not make full use of APIs’ potential semantic information and independent integrity of each API. In this paper, we propose to learn API use specification and detect the API misuse defect based on stacked LSTM. Specifically, first, we obtain ACSG (API Call Syntax Graph) through the static analysis of source code. And then, based on ACSG, we generate API sequences, and transform these sequences into <precious API sequence, next API> for training. Third, in order to represent the APIs in a semantic way, we apply word2vec as a pre-training model to embed features of each API. Though our stacked LSTM model, we regard embedding precious API sequence as the input to model the API use specifications and discover the potential API misuse defects by judging whether the next API is in the output (API probability list) or not. We design experiments to evaluate the effectiveness our method with Java Cryptography APIs and their used code.
Authors: Shuyin Ouyang (Central South University), Fan Ge (Central South University), Li Kuang (Central South University), Yuyu Yin (Hangzhou Dianzi University),
Hide Authors & Abstract

Show Authors & Abstract

Closing Ceremony 15:20 - 15:30

Room #2

Session 8: Internet of Things 08:30 - 10:30

08:30 - 08:50
Networked Multi-Robot Collaboration in Cooperative-Competitive Scenarios under Communication Interference

In this paper, we consider a scenario where a team of predator robots collaboratively survey an area for preventing the invasion from opponent robots. In this scenario, the predator robots can share the sensing information of the prey robots through wireless communication. In order to constrain the surveillance performance of the predator robots, besides the prey robots, some interfering robots are added to break the communication connectivity between the predator robots. This is a typical “cooperative-competitive” decision problem involving multiple optimization variables from electromagnetic and geographic domains, which makes it challenging to solve. For this problem, we first propose the perception and communication models of the robots. Then, with these models, we formulate the problem and adopt multi-agent reinforcement learning (MARL) to solve it. Furthermore, considering the long training-time cost of traditional MARL, we propose a scenario curriculum learning (SCL) training strategy, which can reduce the computation time and improve the performance by evolving the scenarios from simplicity to complexity. The effectiveness of the proposed method is verified by the analysis and simulation results. The results show that the SCL strategy can reduce the training time by 13%.
Authors: Yaowen Zhang (National Innovation Institute of Defense Technology), Dianxi Shi (National Innovation Institute of Defense Technology), Yunlong Wu (National Innovation Institute of Defense Technology), Yongjun Zhang (National Innovation Institute of Defense Technology), Liujing Wang (Tianjin Artificial Intelligence Innovation Center), Fujiang She (National Innovation Institute of Defense Technology),
Hide Authors & Abstract

Show Authors & Abstract
08:50 - 09:10
Multi-UAV Adaptive Path Planning in Complex Environment Based on Behavior Tree

In this paper, we consider a scenario where multiple tracking unmanned aerial vehicles (UAVs) pursue a target UAV in a complex environment. Consider the fast airspeed of the UAV, the path planning needs to be finished in a limited time. Moreover, the complex environment may involve diverse geographical areas, which raises the challenges for the path planning algorithms. For the first challenge, we will adopt the real-time algorithms to keep the efficiency of path planning. For the challenge of environment diversity, we involve the behavior tree (BT) model and propose a BT-organized path planning (BT-OPP) method aiming at achieving adaptive scheduling of different path planning algorithms in different geographical areas. Furthermore, in order to take the advantages of multiple tracking UAVs, we propose a virtual-target-based tracking (VTB-T) method which can make the tracking UAVs pursue the target UAV collaboratively. The effectiveness of the proposed BT-OPP method and the VTB-T method are verified by analysis and numerical results for different system configurations, showing that a substantial target tracking efficiency improvement may be achieved in comparison with the benchmark.
Authors: Wendi Wu (National University of Defense Technology), Jinghua Li (Tianjin Artificial Intelligence Innovation Center), Yunlong Wu (National Innovation Institute of Defense Technology), Xiaoguang Ren (National Innovation Institute of Defense Technology), Yuhua Tang (National University of Defense Technology),
Hide Authors & Abstract

Show Authors & Abstract
09:10 - 09:30
Adaptive Online Estimation of Thrashing-avoiding Memory Reservations for Long-lived Containers

Data intensive computing systems in cloud datacenters create long-lived containers and allocate resources such as memory for containers to execute long-running applications. To avoid wasting resources, it is a challenge to exactly estimate how much memory should be reserved for the containers so that applications can run smoothly. The current state-of-the-art work has the following two limitations: 1) The prediction accuracy is restricted by the monotonicity of the iterative search, and 2) the application performance fluctuates due to the termination conditions. In this paper, we propose two improved strategies based on MEER, called MEER+ and Deep-MEER, to assist in memory allocation adaptively upon resource manager like YARN. MEER+ has one more step of approximation than MEER, which makes the iterative search bi-directional to better approach the optimal value. Based on reinforcement learning, Deep-MEER can achieve thrashing-avoiding estimation without involving termination conditions. We evaluated MEER+ and Deep-MEER with four typical benchmark workloads on a pseudo cluster consisting of virtual machines. The results show that MEER+ and Deep-MEER are both more accurate than MEER. Moreover, Deep-MEER has good generalization ability, namely, a model trained with data from a few applications can be applied to other applications. It also guarantees stable performance for submitted applications during recurring executions.
Authors: Lin Jiayun (Sun Yat-sen University), Liu Fang (Sun Yat-sen University), Cai Zhenhua (Sun Yat-sen University), Huang Zhijie (The University of Texas at Arlington),
Hide Authors & Abstract

Show Authors & Abstract
09:30 - 09:50
A Novel Probabilistic-Performance-Aware Approach to Multi-workflow Scheduling in the Edge Computing Environment

Edge computing is a decentralized computing infrastructure in which data, compute, storage and applications are located somewhere between the data source and the computing facilities. While the edge servers enjoy the close proximity to the end-users to provide services at reduced latency and lower energy costs, they use from limitations in computational and radio resources, which calls for smart, quality-of-service (QoS) guaranteed and efficient task scheduling methods and strategies. For addressing the edge-environment-oriented multi-workflow scheduling problem, in this paper, we propose a probabilistic-QoS-aware approach to multi-workflow scheduling over edge servers with time-varying QoS. Our proposed method leverages a probability-mass function-based QoS aggregation model and a discrete firefly algorithm for generating the multi-workflow scheduling plans. In order to prove the effectiveness of our proposed method, we conducted an experimental case study based on varying types of workflow templates and a real-world dataset for edge server positions. It can be seen that our method clearly outperforms its peers in terms of completion time, cost, and deadline validation rate.
Authors: Yuyin Ma (School of Computers, Chongqing University), Ruilong Yang (School of Computers, Chongqing University), Yiqiao Peng (Bashu Secondary School), Mei Long (ZBJ NETWORK Co. Ltd.), Xiaoning Sun (School of Computers, Chongqing University), Wanbo Zheng (Kunming University of Science and Technology), Xiaobo Li (Chongqing Animal Husbandry Techniques Extension Center Chongqing), Yong Ma (School of Computer and Information Engineering, Jiangxi Normal UniversityNanchang),
Hide Authors & Abstract

Show Authors & Abstract
09:50 - 10:10
RCFC: A Region-Based POI Recommendation Model with Collaborative Filtering and User Context

In the past few years, mobile application has been innovated by leaps and bounds, which leads the prevalence of location-based social networks (LBSNs). Point of interest (POI) recommendation aims to recommend satisfactory locations to users in mobile environment and plays an important role in LBSNs. However, there are still two challenges to be solved. One is the data sparseness caused by users who just visit a few POIs. The other is that it’s hard to make reasonable explanation of recommendation from the perspective of real world. Hence, firstly we propose a region-based collaborative filtering to alleviate the data sparseness by clustering locations into regions. Secondly, we model the impact of two kinds of user contexts like geographical distance and POI category to make POI recommendation more reasonable. Finally, we present a joint model called RCFC which combines the two parts mentioned above. Results of experiments on two real-world datasets demonstrate the model we propose outperforms the popular recommendation algorithms and is more in line with the situation in real world.
Authors: Jun Zeng (Chongqing University), Haoran Tang (Chongqing University), Xin He (Chongqing University),
Hide Authors & Abstract

Show Authors & Abstract
10:10 - 10:30
Delay Constraint Energy Efficient Cooperative offloading in MEC for IoT

Diverse Internet of Things (IoT) applications are usually latency-critical and computation-intensive, on the other hand IoT mobile devices (IMDs) are resource-limited. Mobile edge computing (MEC) is a promising approach to settle the conflict by offloading computing tasks to MEC servers. In this paper, we propose a neighbor-aided cooperative offloading scheme with delay constraint to improve the energy efficiency in MEC for the IoT edge network. The network consists of an IMD with some IMD neighbors, and an access point (AP) integrated with an MEC server. The latency-constrained and computation-intensive tasks in the IMD can be partially offloaded to the selected neighbor or to the MEC server for execution. Different from other offloading cooperation schemes, we propose a scheme by selecting the most energy-efficient one among all the neighbors of the IMD as the offloading helper since some tasks are indivisible or can only be partitioned into a limited number of segments. By minimizing the total energy consumption while satisfying the computation delay constraint, we get the most energy-efficient neighbor with the optimized division of tasks through solving the formulated problem. We also design an easy neighbor selection scheme with lower time complexity according to the weighted value of the transmit data rate. Numerical results show that the proposed scheme outperforms benchmark schemes significantly with the low energy consumption and the improved computation capacity.
Authors: Haifeng Sun (Southwest University of Science and Technology), jun wang (Shenzhen University), Haixia Peng (University of Waterloo), Lili Song (Southwest University of Science and Technology), Mingwei Qin (Southwest University of Science and Technology),
Hide Authors & Abstract

Show Authors & Abstract

Coffee Break 10:30 - 11:00

30 minutes

Session 11: Smart Transportation 11:00 - 13:05

11:00 - 11:20
T2I-CycleGAN: Maritime Road Network Extraction from Crowdsourcing Spatio-Temporal AIS Trajectory Data

Maritime road network is composed of detailed maritime routes and is vital in many applications such as threats detection, traffic control. However, the vessel trajectory data, or Automatic Identification System (AIS) data, are usually large in scale and collected with different sampling rates. And, what’s more, it is difficult to obtain enough accurate road networks as paired training datasets. It is a huge challenge to extract a complete maritime road network from such data that matches the actual route of the ship. In order to solve these problems, this paper proposes an unsupervised learning-based maritime road network extraction model T2I-CycleGAN based on CycleGAN. The method translates trajectory data into unpaired input samples for model training, and adds dense layer to the CycleGAN model to handle trajectories with different sampling rates. We evaluate the approach on real-world AIS datasets in various areas and compare the extracted results with the real ship coordinate data in terms of connectivity and details, achieving effectiveness beyond the most related work.
Authors: Xuankai Yang (North China University of Technology), Guiling Wang (North China University of Technology), Jiahao Yan (North China University of Technology), Jing Gao (North China University of Technology),
Hide Authors & Abstract

Show Authors & Abstract
11:20 - 11:40
Where is the Next Path? A Deep Learning Approach to Path Prediction without Prior Road Networks

Trajectory prediction plays an important role in many urban and marine transportation applications, such as path planning, logistics and traffic management. The existing prediction methods of moving objects mainly focus on trajectory mining in Euclidean space. However, moving objects generally move under road network constraints in the real world. It provides an opportunity to take use of road network constraints or de-facto regular paths for trajectory prediction. As yet, there is little research work on trajectory prediction under road network constraints. And these existing work assumes prior road network information is given in advance. However in some application scenarios, it is very difficult to get road network information, for example the maritime traffic scenario on the wide open ocean. To this end, we propose an approach to trajectory prediction that can make good use of road network constrains without depending on prior road network information. More specifically, our approach extracts road segment polygons from large scale crowdsourcing trajectory data (e.g. AIS positions of ships, GPS positions of vehicles etc.) and translates trajectories into road segment sequences. Useful features such as movement direction and vehicle type are extracted. After that, a LSTM neural network is used to infer the next road segment of a moving object. Experiments on real-world AIS datasets confirm that our approach outperforms the state-of-the-art methods.
Authors: Guiling Wang (North China University of Technology), Mengmeng Zhang (North China University of Technology), Jing Gao (North China University of Technology), Yanbo Han (North China University of Technology),
Hide Authors & Abstract

Show Authors & Abstract
11:40 - 12:00
Distributed Color-based Particle Filter for Target Tracking in Camera Network

Colour-based particle filters have appeared in some literatures, however, there are still some important drawback in tracking targets, such as illumination changes, occlusion and low tracking accuracy. To solve these problems, in this paper, we propose a distributed color-based particle filter (DCPF) for target tracking in camera network, which can track targets accurately in a large-scale camera network with less data transmission and less computation. This is achieved by an local color-based particle filter and a separate fusion filter which is designed to consistently the local filtering distributions into the global posterior by average consensus method. Compared with the previous algorithms, the algorithm in this paper has two obvious advantages. First, the DCPF framework merges color features in the state model to obtain better robustness. Second, it considers the situation where the target is disappear in some camera because of limited field of view (FoV). Convincing results are confirmed that the performance analysis of the proposed algorithm in this paper is very close to the centralized particle filter approaching.
Authors: Yueqing Jing (Anhui University), Yanming Chen (Anhui University),
Hide Authors & Abstract

Show Authors & Abstract
12:00 - 12:20
HMM-based Traffic State Prediction And Adaptive Routing Method In VANETs

As the number of vehicles increases, the traffic environment becomes more complicated. It is important to find a routing method for different scenarios in the vehicular ad hoc networks(VANETs). Although there are many routing methods, these methods rarely consider multiple road traffic states. In this paper, we propose a traffic state prediction method based on Hidden Markov Model(HMM), and then choose different routing methods according to different traffic states. We consider that GPS may cause measurement errors, so Kalman Filter is used to estimate the observation, which makes observation more accurately. For different road states, we can make appropriate methods to improve routing performance. For example, when the road is in rush hour, we will use Extended Kalman Filter to predict vehicle information in a short time to reduce the number of broadcasts, which can reduce channel load. The result show that our method is useful for reducing the number of packets and improving the delivery rate.
Authors: Kaihan Gao (Hefei University of Technology), Xu Ding (Hefei University of Technology), Juan Xu (Hefei University of Technology), Fan Yang (Hefei University of Technology), Chong Zhao (Hefei University of Technology),
Hide Authors & Abstract

Show Authors & Abstract
12:20 - 12:45
A Hybrid Deep Learning Approach for Traffic Flow Prediction in Highway Domain

With the development of cities, intercity highway plays a vital role in people's daily travel. The traffic flow on the highway network is also increasingly concerned by road managers and road participants. However, due to the influence of highway network topology, as well as the feature such as weather, traffic flow prediction becomes more complicated. So, it is much difficult to construct a multi-feature matrix and predict the traffic flow of the whole network at a time. A novel prediction method of multi-feature whole network traffic flow that based on a hybrid deep learning model, which can learn multi-feature and predict a whole network traffic flow is proposed. The experiment shows that the prediction accuracy of this method is significantly better than the existing methods, and it has a good performance in the whole network prediction.
Authors: Zhe Wang (North China University of Technology), Weilong Ding (North China University of Technology), Hui Wang (Beijing China-Power Information Technology Company Limited),
Hide Authors & Abstract

Show Authors & Abstract
12:45 - 13:05
HomoNet: Unified License Plate Detection and Recognition in Complex Scenes

Although there are many commercial systems for license plate detection and recognition (LPDR), existing approaches based on object detection and Optical Character Recognition (OCR) are difficult to achieve good performance in both efficiency and accuracy in complex scenes (e.g., varying viewpoint, light, weather condition, etc). To tackle this problem, this work proposed a unified end-to-end trainable fast perspective LPDR network named HomoNet for simultaneous detection and recognition of twisted license plates. Specifically, we state the homography pooling (HomoPooling) operation based on perspective transformation to rectify tilted license plates. License plate detection was replaced with keypoints location to obtain richer information and improve the speed and accuracy. Experiments show that our network outperforms the state-of-the-art methods on public datasets, such as 95.58%@22.5ms on RP and 97.5%@19ms on CCPD.
Authors: yang yuxin (Xi'an Jiaotong University), xi wei (Xi'an Jiaotong University), zhu chengkai (SenseTime Group Ltd.), zhao yihan (Xi'an Jiaotong University),
Hide Authors & Abstract

Show Authors & Abstract

Lunch Break 13:05 - 13:30

25 minutes

Session 13: Resource Management; Artificial Intelligence 13:30 - 15:05

13:30 - 13:45
Reactive Workflow Scheduling in Fluctuant Infrastructure-as-a-Service Clouds using Deep Reinforcement Learning

As a promising and evolving computing paradigm, cloud computing benefits scientific computing-related computational-intensive applications, which usually orchestrated in terms of workflows, by providing unlimited, elastic, and heterogeneous resources in a pay-as-you-go way. Given a workflow template, identifying a set of appropriate cloud services that fulfill users’ functional requirements under pre-given constraints is widely recognized to be a challenge. However, due to the situation that the supporting cloud infrastructures can be highly prone to performance variations and fluctuations, various challenges such as guaranteeing user-perceived performance and reducing the cost of the cloud-supported scientific workflow need to be properly tackled. Traditional approaches tend to ignore such fluctuations when scheduling workflow tasks and thus can lead to frequent violations to Service-Level-Agreement (SLA). On the contrary, we take such fluctuations into consideration and formulate the workflow scheduling problem as a continuous decision-making process and propose a reactive, deep-reinforcement-learning-based method, named DeepWS, to solve it. Extensive case studies based on real-world workflow templates show that our approach outperforms significantly than traditional ones in terms of SLA-violation rate and total cost.
Authors: Qinglan Peng (Chongqing University), Wanbo Zheng (Data Science Research Center, Kunming University of Science and Technology, Kunming 650031, China), Yunni Xia (Chongqing University), Chunrong Wu (Chongqing University), Yin Li (Institute of Software Application Technology, Guangzhou & Chinese Academy of Sciences), Mei Long (ZBJ Network Co. Ltd.), Xiaobo Li (Chongqing Animal Husbandry Techniques Extension Center),
Hide Authors & Abstract

Show Authors & Abstract
13:45 - 14:10
BPA:The Optimal Placement of Interdependent VNFs in Many-core System

Network function virtualization (NFV) brings the potential to provide the flexible implementation of network functions and reduce overall hardware cost by running service function chains on commercial off-the-shelf servers with many-core processors. Towards this direction, both academia and industry have spent vast amounts of effort to address the optimal placement challenges of NFV middleboxes. Most of the servers usually are equipped with Intel X86 processors, which adopt Non-Uniform Memory Access (NUMA) architecture. However, existing solutions for placing SFCs in one server either ignore the impact of hardware architecture or overlook the dependency between middleboxes. Our empirical analysis shows that the placement of virtual network functions (VNFs) with interdependency in a server needs more particular consideration. In this paper, we first manage the optimal placement of VNFs by jointly considering the discrepancy of cores in different NUMA nodes and interdependency between network functions (NFs), and formulate the optimization problem as a Non-Linear Integer Programming model. Then we find a reasonable metric to describe the dependency relation formally. Finally, we propose a heuristic-based backtracking placement algorithm (BPA) to find the near-optimal placement solution. The evaluation shows that, compared with two state-of-art placement strategies, our algorithm can improve the aggregate performance by an average of 20% or 45% within an acceptable time range.
Authors: Youbing Zhong (University of Chinese Academy of Sciences), Zhou Zhou (Institute of Information Engineering), Xuan Liu (Southeast University), Da Li (Dept. of Electrical and Computer Engineering, University of Missouri-Columbia), Meijun Guo (School of Mathematical Sciences, University of Chinese Academy of Sciences), shuai Zhang (University of Chinese Academy of Sciences), Qingyun Liu (Institute of Information Engineering, Chinese Academy of Sciences), Li Guo (Institute of Information Engineering),
Hide Authors & Abstract

Show Authors & Abstract
14:10 - 14:25
Cooperative Source Seeking in Scalar Field: A Virtual Structure-based Spatial-Temporal Method

Source seeking problem has been faced in many fields, especially in search and rescue applications such as first-response rescue, gas leak search, etc. We proposed a virtual structure based spatial-temporal method to realize cooperative source seeking using multi-agents. Spatially, a circular formation is considered to gather collaborative information and estimate the gradient direction of the formation center. In terms of temporal information, we make use of the formation positions in time sequence to construct a virtual structure sequence. Then, we fuse the sequential gradient as a whole. A control strategy with minimum movement cost is proposed. This strategy rotates the target formation by a certain angle to make the robot team achieve the minimum moving distance value when the circular team moves to the next position. Experimental results show that, compared with state-of-the-art, the proposed method can quickly find the source in as few distances as possible, so that the formation can minimize the movement distance during the moving process, and increase the efficiency of source seeking. Numerical simulations confirm the efficiency of the scheme put forth. Compared with state-of-the-art source seeking methods, the iterative steps of our proposed method is reduced by 20\%, indicating that the method can find the signal source with higher efficiency and lower energy consumption, as well as better robustness.
Authors: Cheng Xu (University of Science & Technology Beijing), Yulin Chen (University of Science & Technology Beijing), Shihong Duan (University of Science & Technology Beijing), Hang Wu (University of Science & Technology Beijing),
Hide Authors & Abstract

Show Authors & Abstract
14:25 - 14:45
BiC-DDPG: Bidirectionally-Coordinated Nets for Deep Multi-Agent Reinforcement Learning

Multi-agent reinforcement learning (MARL) often faces the problem of policy learning under large action space. There are two reasons for the complex action space: first, the decision space of a single agent in a multi-agent system is huge. Second, the complexity of the joint action space caused by the combination of the action spaces of different agents increases exponentially from the increase in the number of agents. How to learn a robust policy in multi-agent cooperative scenarios is a challenge. To address this challenge we propose an algorithm called bidirectionally-coordinated Deep Deterministic Policy Gradient (BiC-DDPG). In BiC-DDPG three mechanisms were designed based on our insights against the challenge: we used a centralized training and decentralized execution architecture to ensure Markov property and thus ensure the convergence of the algorithm, then we used bi-directional rnn structures to achieve information communication when agents cooperate, finally we used a mapping method to map the continuous joint action space output to the discrete joint action space to solve the problem of agents' decision-making on large joint action space. A series of fine grained experiments in which include scenarios with cooperative and adersarial relationships between homogeneous agents were designed to evaluate our algorithm. The experiment results show that our algorithm out performing the baseline.
Authors: Gongju Wang (Artificial Intelligence Research Center (AIRC), National Innovation Institute of Defense Technology (NIIDT), Beijing 100166, China), Dianxi Shi (Artificial Intelligence Research Center (AIRC), National Innovation Institute of Defense Technology (NIIDT), Beijing 100166, China), Chao Xue (Artificial Intelligence Research Center (AIRC), National Innovation Institute of Defense Technology (NIIDT), Beijing 100166, China), Hao Jiang (College of Computer, National University of Defense Technology, Changsha 410073, China), Yajie Wang (College of Computer, National University of Defense Technology, Changsha 410073, China),
Hide Authors & Abstract

Show Authors & Abstract
14:45 - 15:05
FocAnnot: Patch-wise Active Learning for Intensive Cell Image Segmentation

In the era of deep learning, data annotation becomes an essential but costly work, especially for the biomedical image segmentation task. To tackle this problem, active learning (AL) aims to select and annotate a part of available images for modeling while retaining accurate segmentation. Existing AL methods usually treat an image as a whole during the selection. However, for an intensive cell image that includes similar cell objects, annotating all similar objects would bring duplication of efforts and have little benefit to the segmentation model. In this study, we present a patch-wise active learning method, namely FocAnnot (focal annotation), to avoid such worthless annotation. The main idea is to group different regions of images to discriminate duplicate content, then evaluate novel image patches by a proposed cluster-instance double ranking algorithm. Instead of the whole image, experts only need to annotate specific regions within an image. This reduces the annotation workload. Experiments on the real-world dataset demonstrate that FocAnnot can save about 15% annotation cost to obtain an accurate segmentation model or provide a 2% performance improvement at the same cost.
Authors: Bo Lin (Zhejiang University), Shuiguang Deng (Zhejiang University), Jianwei Yin (Zhejiang University), Jindi Zhang (Zhejiang University), Ying Li (Zhejiang University), Honghao Gao (Shanghai University),
Hide Authors & Abstract

Show Authors & Abstract
Room #3

Session 9: Collaborative Robotics and Autonomous Systems 08:30 - 10:25

08:30 - 08:50
Self-organised Flocking with Simulated Homogeneous Robotic Swarm

Flocking is a common behaviour observed in social animals such as birds and insects, which has received considerable attention in swarm robotics research studies. In this paper, a homogeneous self-organised flocking mechanism was implemented using simulated robots to verify a collective model. We identified and proposed solutions to the current gap between the theoretical model and the implementation with real-world robots. Quantitative experiments were designed with different factors which are swarm population size, desired distance between robots and the common goal force. To evaluate the group performance of the swarm, the average distance within the flock was chosen to show the coherency of the swarm, followed by statistical analysis to investigate the correlation between these factors. The results of the statistical analysis showed that compared with other factors, population size had a significant impact on the swarm flocking performance. This provides guidance on the application with real robots in terms of factors and strategic design.
Authors: Zhe Ban (Dept Electrical and Electronic Engineering, The University of Manchester), Craig West (Bristol Robotic Lab, UK), Barry Lennox (Dept Electrical and Electronic Engineering, The University of Manchester), Farshad Arvin (Dept Electrical and Electronic Engineering, The University of Manchester),
Hide Authors & Abstract

Show Authors & Abstract
08:50 - 09:10
Investigation of Cue-based Aggregation Behaviour in Complex Environments

Swarm robotics is mainly inspired by the collective behaviour of social animals in nature. Among different behaviours such as foraging and flocking performed by social animals; aggregation behaviour is often considered as the most basic and fundamental one. Aggregation behaviour has been studied in different domains for over a decade. In most of these studies, the settings are over-simplified that are quite far from reality. In this paper, we investigate cue-based aggregation behaviour using BEECLUST in a complex environment having two cues --one being the local optimum and the other being the global optimum-- with an obstacle between the two cues. We measured the aggregation size on both cues with and without the obstacle varying the number of robots. The simulations were performed on a custom open-source simulation platform, Bee-Ground, using custom MONA robots. The results showed that the aggregation behaviour was able to overcome a certain degree of environmental complexities revealing the robustness of the method. We also verified these results using our stock-flow model.
Authors: Shiyi Wang (Dept Electrical and Electronic Engineering, The University of Manchester), Ali Turgut (Middle East Technical University), Thomas Schmickl (University of Graz, Austria), Barry Lennox (Dept Electrical and Electronic Engineering, The University of Manchester), Farshad Arvin (Dept Electrical and Electronic Engineering, The University of Manchester),
Hide Authors & Abstract

Show Authors & Abstract
09:10 - 09:25
A Tamper-resistant and Decentralized Service for Cloud Storage based on Layered Blockchain

With the ever-increasing scale of data, IP traffic of data centers will gradually suffer from a serious shortage. Centralized clouds may no longer deliver satisfactory storage services. To alleviate network pressure, transmitting source data to the edge of network instead of the remote cloud is becoming more and more important. Based on the blockchain technology, we propose a decentralized cloud storage service with tamper-resistant function. Our service provides users with metadata storage and management capabilities. In order to overcome the inherent defects of blockchain, in our design, the blockchain network has a layered and collaborative structure. After the storage capacity problem is solved, our service allows users to upload digests of source data to make the query function more user-friendly. For privacy protection, we subjoin the access management function as well. Finally, the experimental results show that the performance and availability of the blockchain network are able to support our service to provide efficient functions to users.
Authors: Fuxiao Zhou (Shanghai Jiao Tong University), Haopeng Chen (Shanghai Jiao Tong University), Zhijian Jiang (Shanghai Jiao Tong University),
Hide Authors & Abstract

Show Authors & Abstract
09:25 - 09:45
A Blockchain Based Cloud integrated IoT Application using a Hybrid Design

The Internet of Things (IoT) plays a vital role in society due to its positive features like portability and accessibility. We can enhance the efficiency of an IoT based system by adding other advanced technologies like cloud infrastructure and blockchain technology. IoT applications can be accessed at any time and from any place. However, a database is required for storing and computing application information. Cloud infrastructure provides services on data like storage, computation, and analysis. Hence utilization of cloud-based IoT applications has increased in a global fashion. The major drawback of these applications are their inability to provide privacy and security preservation to data that has been maintained on the cloud due to its centralized architecture. Blockchain technology can help in overcoming these drawbacks with features like immutability, transparency, and distributed structure. In this paper, a blockchain-based cloud-integrated IoT application is proposed that can help identify intruders through virtual monitoring. The main advantage of this application is that it can operate in surroundings where manual monitoring is difficult and data is stored on a blockchain-based tamper-free environment.
Authors: Rupa Ch (VR Siddhartha Engineering College), Gautam Srivastava (Brandon University), Thippa Reddy G (Vellore Institute of Technology), Praveen Reddy (Vellore Insitute of Technology), Sweta Bhattacharya (VIT Vellore),
Hide Authors & Abstract

Show Authors & Abstract
09:45 - 10:05
Towards Efficient and Privacy-Preserving Service QoS Prediction with Federated Learning

With the widespread adoption of cloud computing, large-scale online applications composed of services have been deployed in many critical areas. In order to ensure the performance of cloud applications, Quality of Service (QoS) is a key indicator commonly used for service selection and adaptation. To facilitate QoS-based selection and adaptation, previous studies have proposed collaborative QoS prediction approaches to estimate personalized QoS values as a supplement to sparse user-perceived QoS data. However, collaborative QoS prediction encounters privacy problems in practice, which makes users reluctant to collaborate through sharing data. As a result, privacy threat has become a key challenge to make QoS prediction approaches practical. In this paper, we proposed a privacy-preserving QoS prediction approach employing federated learning techniques to tackle this grand challenge. We further propose several efficiency improvement techniques to significantly reduce system overhead and make the federated privacy-preserving QoS prediction approach feasible. The proposed approach is evaluated on a large-scale real-world QoS dataset, and the experimental results confirm its effectiveness and efficiency.
Authors: Yilei Zhang (Anhui Normal University), Xiao Zhang (Anhui Normal University), Xinyuan Li (Anhui Normal University),
Hide Authors & Abstract

Show Authors & Abstract
10:05 - 10:25
A Reinforcement Learning based Approach to Identify Resource Bottlenecks for Multiple Services Interactions in Cloud Computing Environments

Cloud service providers are provisioning resources including a variety of virtual machine instances to support customers that migrate their services to the cloud. From the customers’ perspective, selecting the appropriate amount of resources is tightly coupled with performance and cost. By identifying the potential resource bottlenecks in the early stage of the service deployment process, resource planning can be significantly optimized. However, due to the unpredictable workloads and heterogeneous resources, it is difficult to identify resource bottlenecks that can degrade system performance. To support system non-functional requirements (NFR) in a better manner, we propose a reinforcement learning based approach to support the NFR management of system concerning the multiple services interactions scenario by identifying the potential resource bottleneck and optimizing the demanded resources. The proposed approach can predict the resource bottleneck for multiple services interactions, e.g. bottleneck in CPU or overloads in specific service, and provide guidance for resource planning. We modeled and simulated the proposed approach using an extended version of the CloudSim toolkit. Comprehensive evaluations with realistic data from Siemens MindSphere and Alibaba Cloud service show that our proposed approach can achieve high accuracy in terms of performance metrics, such as response time, queries per second (QPS), and resource usage.
Authors: Lingxiao Xu (School of Software and Information Engineering, University of Electronic Science and Technology of China, China), Minxian Xu (Shenzhen Institutes of Advanced Technology, Chinese Academy of Sciences, China), Richard Semmes (Siemens Industry Software (Chengdu) Co., Ltd , China), Hui Li (Siemens Industry Software (Chengdu) Co., Ltd , China), Hong Mu (Siemens Industry Software (Chengdu) Co., Ltd , China), Shuangquan Gui (Siemens Industry Software (Chengdu) Co., Ltd , China), Wenhong Tian (School of Software and Information Engineering, University of Electronic Science and Technology of China, China), Kui Wu (Department of Computer Science of University of Victoria, Canada.), Rajkumar Buyya (School of Computing and Information Systems, University of Melbourne, Australia),
Hide Authors & Abstract

Show Authors & Abstract