Energy-Efficient Service Function Orchestration in 5G Mobile Networks
Lead Research Organisation:
University of Exeter
Department Name: Computer Science
Abstract
Recent years have witnessed an increasing diversity of network services and novel applications, which are subject to a wide spectrum of service requirements with respect to bandwidth, throughput, and latency. Energy-efficiency of networks and systems becomes a dramatically growing concern, due to both increasing energy cost and CO2 emission. 5G mobile networks are being recognised as the next-generation Internet to meet the varying needs of diversified services and to reduce the carbon emissions of telecom infrastructure and data centres. In contrast to the network functions built with dedicated equipment in the current network, network elements in 5G mobile networks are provided as Virtual Network Functions (VNF, software implementations of network functions), which are elastically orchestrated in the form of Service Function Chains (SFC) to meet dynamic service demands and energy-efficiency goals.
The SFC orchestration can be performed in two steps: (i) VNF chaining, to define the order in which network functions are connected to form an SFC; (ii) SFC placement, to define how the SFC is embedded into the physical network. The essential features in 5G, such as cross-domain differentiated service and network dynamicity and uncertainty, have made traditional chaining methods and placement models inapplicable for 5G mobile services. Therefore, significant efforts have been devoted to tackling the research challenges of SFC orchestration.
Existing studies have primarily leveraged obvious mutual VNF dependencies to perform VNF chaining. However, such a method cannot determine the global ordering for all involved VNFs of an end-to-end 5G service. Furthermore, it narrows the channel to create SFCs with the aim of optimising the differentiated services and energy-efficiency when being instantiated in a physical network. In addition, SFC placement has been considered as a deterministic scheduling problem with the complete service and network information known as a priori. However, when applied to 5G mobile networks, additional challenges are presented in that network resource and service requirements vary at runtime due to the factors of e.g. time-varying user mobility and dynamic network management. Until now, an energy-efficient SFC orchestration in 5G mobile networks, considering both optimal VNF chaining and dynamic service demands and substrate resource capacities, has not been reported in the existing literature.
This project will investigate VNF chaining and SFC placement in order to achieve energy-efficient and proactive SFC orchestration in the context of 5G mobile networks. To tackle this challenging problem progressively, our work will be focused on three major objectives: 1) propose a uniform VNF chaining framework, that can integrate VNF dependencies, differentiated service policies, and energy-efficiency objectives, to efficiently construct SFCs; 2) develop a proactive model, considering future network and service variations at the initialisation of SFC placement, to perform practical service deployment in 5G mobile networks; 3) propose a real-time deterministic model to perform SFC placement for unforeseen variations with the aim of minimising the negative effects on network performance due to service demand violations. The insights into the energy-efficient VNF chaining framework obtained in Objective 1 will be fed into the SFC placement models in Objective 2 and Objective 3. The research proposed in this project is the first of its kind on the energy-efficient SFC orchestration in 5G mobile networks. In the long-term aim, the implications of this research will contribute to 5G service orchestration in both theoretical and practical sides and pave the way for future green 5G mobile networks.
The SFC orchestration can be performed in two steps: (i) VNF chaining, to define the order in which network functions are connected to form an SFC; (ii) SFC placement, to define how the SFC is embedded into the physical network. The essential features in 5G, such as cross-domain differentiated service and network dynamicity and uncertainty, have made traditional chaining methods and placement models inapplicable for 5G mobile services. Therefore, significant efforts have been devoted to tackling the research challenges of SFC orchestration.
Existing studies have primarily leveraged obvious mutual VNF dependencies to perform VNF chaining. However, such a method cannot determine the global ordering for all involved VNFs of an end-to-end 5G service. Furthermore, it narrows the channel to create SFCs with the aim of optimising the differentiated services and energy-efficiency when being instantiated in a physical network. In addition, SFC placement has been considered as a deterministic scheduling problem with the complete service and network information known as a priori. However, when applied to 5G mobile networks, additional challenges are presented in that network resource and service requirements vary at runtime due to the factors of e.g. time-varying user mobility and dynamic network management. Until now, an energy-efficient SFC orchestration in 5G mobile networks, considering both optimal VNF chaining and dynamic service demands and substrate resource capacities, has not been reported in the existing literature.
This project will investigate VNF chaining and SFC placement in order to achieve energy-efficient and proactive SFC orchestration in the context of 5G mobile networks. To tackle this challenging problem progressively, our work will be focused on three major objectives: 1) propose a uniform VNF chaining framework, that can integrate VNF dependencies, differentiated service policies, and energy-efficiency objectives, to efficiently construct SFCs; 2) develop a proactive model, considering future network and service variations at the initialisation of SFC placement, to perform practical service deployment in 5G mobile networks; 3) propose a real-time deterministic model to perform SFC placement for unforeseen variations with the aim of minimising the negative effects on network performance due to service demand violations. The insights into the energy-efficient VNF chaining framework obtained in Objective 1 will be fed into the SFC placement models in Objective 2 and Objective 3. The research proposed in this project is the first of its kind on the energy-efficient SFC orchestration in 5G mobile networks. In the long-term aim, the implications of this research will contribute to 5G service orchestration in both theoretical and practical sides and pave the way for future green 5G mobile networks.
Planned Impact
Economic Impact
This project makes a significant contribution to the proactive and green service deployment in 5G mobile networks, by addressing two main research challenges: service function chaining and placement. Research outcomes will contribute to the UK leadership in 5G, which plays a key role in defining industry standards to support an estimated 5 to 6% of UK GDP per annum indicated by a recent UK Government report. In addition, research outcomes will have wider impact by contributing to the Internet service providers, mobile network/service operators, content providers, and Internet services and applications, through providing necessary mechanisms to satisfy dynamic user demands and time-varying substrate network resources in an energy-efficient manner. This will open up new business models for network operators and widen the market for service providers.
Mobile Internet Users
Mobile users are increasingly expected to receive content with high Quality-of-Experience (QoE). 5G mobile network is expected to provide excellent QoE for mobile Internet users, by realising swift, flexible, scalable, and differentiated service deployment. The proposed research is vital and investigates necessary components on achieving these goals. Research outcomes can help bring 5G-enabled service deployment closer to practice, which will ultimately offer mobile users best QoE in an energy-efficient way by minimising the service response time and optimising the network resource utilisation.
Mobile Network/Service Operators
Scenarios for example online service deployment with uncertain service demands, addressed by this project are not considered by today's network due to enormously high cost of conventional solutions. One of the main objectives of this project is to propose solutions for such scenarios that are commercially viable and academically challenging. The market for mobile network/service operators will thus naturally expand. Energy-efficiency is a growing concern due to both increasing energy costs and CO2 emissions. Due to explosive growth of emerging and diverse mobile services, it would be desirable to adopt dynamic and proactive orchestration approaches to reduce energy consumption, which is the general research theme of this project. Hence, the project outcome facilitates more sustainable ICT in UK.
Internet Industries
The exponential growth of mobile traffic poses a significant challenge because the Internet was designed to accommodate static service deployment, which impacts every part of the Internet ecosystem. The whole Internet industry needs novel and economical ways to solve these problems. A possible solution to provide dynamic service deployment is to develop flexible 5G mobile networks. In moving to 5G, Internet industries will find themselves at an advantage with the outcomes of this research, in terms of low energy consumption, high QoE, and proactive service deployment, which will undoubtedly enhance their competitiveness and profitability.
Public Impact
To facilitate the understanding of this scientific technologies by public, this project will deliver the public impact through Exeter's public engagement and outreach programmes, such as School Visits, and through science fairs, such as The Big Bang Fair. In addition, a version of the project website will be maintained using plain English and animations to facilitate public understanding and engage public participation.
Education and Training
University of Exeter aims to constantly enhance its teaching quality by introducing new research findings and applied technologies into the curriculum. The outcomes of this research are highly relevant to the Computer Science programme. The project endeavours to enhance the knowledge and skill base of the UK through training and developing researchers. The PDRA employed on the project will receive valuable training in this highly marketable sector, which is useful for future career advancement.
This project makes a significant contribution to the proactive and green service deployment in 5G mobile networks, by addressing two main research challenges: service function chaining and placement. Research outcomes will contribute to the UK leadership in 5G, which plays a key role in defining industry standards to support an estimated 5 to 6% of UK GDP per annum indicated by a recent UK Government report. In addition, research outcomes will have wider impact by contributing to the Internet service providers, mobile network/service operators, content providers, and Internet services and applications, through providing necessary mechanisms to satisfy dynamic user demands and time-varying substrate network resources in an energy-efficient manner. This will open up new business models for network operators and widen the market for service providers.
Mobile Internet Users
Mobile users are increasingly expected to receive content with high Quality-of-Experience (QoE). 5G mobile network is expected to provide excellent QoE for mobile Internet users, by realising swift, flexible, scalable, and differentiated service deployment. The proposed research is vital and investigates necessary components on achieving these goals. Research outcomes can help bring 5G-enabled service deployment closer to practice, which will ultimately offer mobile users best QoE in an energy-efficient way by minimising the service response time and optimising the network resource utilisation.
Mobile Network/Service Operators
Scenarios for example online service deployment with uncertain service demands, addressed by this project are not considered by today's network due to enormously high cost of conventional solutions. One of the main objectives of this project is to propose solutions for such scenarios that are commercially viable and academically challenging. The market for mobile network/service operators will thus naturally expand. Energy-efficiency is a growing concern due to both increasing energy costs and CO2 emissions. Due to explosive growth of emerging and diverse mobile services, it would be desirable to adopt dynamic and proactive orchestration approaches to reduce energy consumption, which is the general research theme of this project. Hence, the project outcome facilitates more sustainable ICT in UK.
Internet Industries
The exponential growth of mobile traffic poses a significant challenge because the Internet was designed to accommodate static service deployment, which impacts every part of the Internet ecosystem. The whole Internet industry needs novel and economical ways to solve these problems. A possible solution to provide dynamic service deployment is to develop flexible 5G mobile networks. In moving to 5G, Internet industries will find themselves at an advantage with the outcomes of this research, in terms of low energy consumption, high QoE, and proactive service deployment, which will undoubtedly enhance their competitiveness and profitability.
Public Impact
To facilitate the understanding of this scientific technologies by public, this project will deliver the public impact through Exeter's public engagement and outreach programmes, such as School Visits, and through science fairs, such as The Big Bang Fair. In addition, a version of the project website will be maintained using plain English and animations to facilitate public understanding and engage public participation.
Education and Training
University of Exeter aims to constantly enhance its teaching quality by introducing new research findings and applied technologies into the curriculum. The outcomes of this research are highly relevant to the Computer Science programme. The project endeavours to enhance the knowledge and skill base of the UK through training and developing researchers. The PDRA employed on the project will receive valuable training in this highly marketable sector, which is useful for future career advancement.
Organisations
- University of Exeter (Lead Research Organisation)
- University of Sydney (Collaboration)
- Macau University of Science and Technology (Collaboration)
- Norwegian University of Science and Technology (NTNU) (Collaboration)
- Tsinghua University China (Collaboration)
- UNIVERSITY OF BRITISH COLUMBIA (Collaboration)
- UNIVERSITY OF EXETER (Collaboration)
People |
ORCID iD |
| Yulei Wu (Principal Investigator) |
Publications
Cheng X
(2018)
Network Function Virtualization in Dynamic Networks: A Stochastic Perspective
in IEEE Journal on Selected Areas in Communications
Cheng X
(2020)
Safeguard Network Slicing in 5G: A Learning Augmented Optimization Approach
in IEEE Journal on Selected Areas in Communications
Huang Z
(2022)
An Energy-efficient And Trustworthy Unsupervised Anomaly Detection Framework (EATU) for IIoT
in ACM Transactions on Sensor Networks
Miao W
(2019)
Stochastic Performance Analysis of Network Function Virtualization in Future Internet
in IEEE Journal on Selected Areas in Communications
Wang H
(2022)
A Graph Neural Network-Based Digital Twin for Network Slicing Management
in IEEE Transactions on Industrial Informatics
Wu Y
(2021)
Cloud-Edge Orchestration for the Internet of Things: Architecture and AI-Powered Data Processing
in IEEE Internet of Things Journal
Wu Y
(2021)
Deep learning for privacy preservation in autonomous moving platforms enhanced 5G heterogeneous networks
in Computer Networks
Wu Y
(2021)
Convergence of Blockchain and Edge Computing for Secure and Scalable IIoT Critical Infrastructures in Industry 4.0
in IEEE Internet of Things Journal
Wu Y
(2021)
Deep reinforcement learning for blockchain in industrial IoT: A survey
in Computer Networks
Wu Y
(2022)
Graph Neural Networks for Anomaly Detection in Industrial Internet of Things
in IEEE Internet of Things Journal
| Description | Network slicing is proposed as a promising paradigm to meet the strict requirements of a wide variety of mission-critical applications and services in the fifth generation communication systems (5G). A paramount issue to be solved for network slices is to ensure that the service function orchestrator meet the quality-of-service (QoS) requirement for all services running on slices across different network domains. The inherent network dynamics and uncertainties from 5G infrastructure, resources and applications are slowing down the further adoption of service function orchestrator in many emerging networking applications. Motivated by this, we investigate the issues of network utility degradation when implementing service function orchestrator in dynamic networks, and design a proactive orchestration solution from a fully stochastic perspective. Unlike existing deterministic solutions, which assume given network capacities and/or static service quality demands, the proposed solution explicitly integrates the knowledge of influential network variations into a two-stage stochastic resource utilisation model. By exploiting the hierarchical decision structures in this problem, a distributed computing framework with two-level decomposition is designed to facilitate a distributed implementation of the proposed model in large-scale networks. Experimental results demonstrate that the proposed solution not only improves 3~5 folds of network performance, but also effectively reduces the risk of service quality violation. In the event that prior environmental knowledge is absent for the decision of service function orchestrator in dynamic networks like 5G, we propose an online two-stage slicing optimisation model with time-averaged metrics to safeguard the network slicing, where prior environmental knowledge can be partially observed at runtime. Directly solving an off-line solution to this problem is intractable since the future system realisations are unknown before decisions. Therefore, we propose a learning augmented optimisation approach with deep learning and Lyapunov stability theories. This enables the system to learn a safe slicing solution from both historical records and run-time observations. We prove that the proposed solution is always feasible and nearly optimal, up to a constant additive factor. Finally, we demonstrate up to 2.6× improvement in the simulation when compared with three state-of-the-art algorithms. With the development of the Internet of Things (IoT) and the pervasive 5G network, the interaction and convergence between the physical world and the cyber world of many industries are getting faster. Digital twin (DT) technology paves a way to cyber-physical integration, through creating virtual replicas of physical objects in the digital way to simulate their behaviours. As the complexity of 5G communication networks is exacerbated by the demands of various applications, managing a network of this unprecedented complexity would benefit greatly from a network digital twin. A network DT refers to a computer simulation model of the communication network, along with its operating environment and the carried application traffic. The DT can be used to study the behaviour of its physical counterpart under a diverse set of operating conditions, including cyber-attacks, and adjust its actual network configuration as needed. It could also be used for the effective management of network infrastructure and keeping track of configuration changes, particularly those in remote locations which can be difficult to maintain. To this end, we designed a data-driven model and established a network DT based on the model to monitor the performance and predict the end-to-end (E2E) latency of network slices in the network, mapping between the observed network status and performance metrics. A DT of network slicing is important to achieve cost-efficient and performance-optimal slicing management and keep monitoring the performance under a diverse set of operating conditions without impacting the physical network. Three main innovations have been achieved through this work. Firstly, we model the intricate interactions between multiple slices deployed on a shared physical infrastructure into a graph structure and design a novel Graph Neural Network (GNN)-based virtual representation model. Unlike other machine learning tasks, network slices and their substrate network are naturally in the form of graphs. Nevertheless, most of the existing learning frameworks are not designed to learn from data in a non-Euclidean structure of real networks and are limited in providing accurate results on graph data and hard to achieve generalisation on dynamic topologies and configurations. Therefore, we developed a graph-based representation for slice enabled networks to solve the challenges caused by the irregular topologies of graphs and the interdependency between nodes by modelling the whole substrate network into a directed graph. Secondly, we propose a DT of network slicing to investigate the composite traffic generated by slices and produce accurate predictions of E2E slice latency. The DT adopts the inductive graph learning framework to achieve generalisation and perform prediction under various environments. At last, we conduct various experiments to demonstrate the high accuracy and the generalisation capability of the DT for estimating the E2E latency of slices under arbitrary topologies, diverse slice deployment and traffic distributions that have not been seen during the training process. Hypothetical what-if scenarios are performed using the DT to evaluate the E2E performance with QoS constraints. With appropriate design and development, E2E network slicing is feasible to accommodate smart transportation, smart energy and smart factory. |
| Exploitation Route | The proposed proactive orchestration solution and online two-stage slicing optimisation model have considered the dynamic nature of 5G, and thus they can be applied in real-world networking environment to efficiently orchestrate network functions for meeting the requirements of diverse applications. The proposed solutions can be applied by Internet service providers like BT, mobile service providers like O2 and Vodafone, and content service providers like BBC and Netflix. We are currently negotiating with BT to find a suitable way of collaboration and implementation of the proposed solutions. For the proposed DT, it can be integrated with various deep reinforcement learning (DRL)-based methods. DRL is one of the state-of-the-art machine-learning branches and has attracted many efforts from both the academic and non-academic sectors. Training a DRL algorithm relies on the interaction between an agent and an environment. However, it is hard to have an accurate environment for experiments, and sometimes the cost of training is high. Our DT model can be treated as an environment and generate various training data samples acting as the inputs for the DRL model. This can be achieved through manipulating the traffic and allocated resources to the slices/users thousands of times. The results of the project show a mutual benefit between Artificial Intelligence (AI) and DT. On the one hand, leveraging AI, DT models can be created based on observed historical data rather than just the design information. It can help discover previously unseen patterns. On the other hand, the DT can produce simulated data. A virtual environment can go through an infinite number of repetitions and scenarios. The simulated data produced can then be used to train the AI model under potential real-world conditions, that might otherwise be very rare or still in the testing phase. The proposed solutions can be used to engage with industry partners in health, transportation and energy. |
| Sectors | Digital/Communication/Information Technologies (including Software) |
| Description | Based on the outcome of this research, we have created five Year-12 projects for Exeter Mathematics School. It gives students an early insight into the world of research on 5G and 6G, and provides a chance for students to learn how to utilise machine learning to solve the resource optimisation problem in 5G/6G network slicing and network digital twins. We aim to deepen and broaden their understanding of computer science and its related fields. We also intent to support and encourage their development as problem-solving scientists, especially for the girls. As the complexity of today's communication networks is exacerbated by the demands of 5G/6G and IoT, managing a network of this unprecedented complexity would benefit greatly from a digital twin. The proposed network DT model achieves a vital step towards realising the ambitious vision of autonomous management of network slicing by providing the ability to map the network status to E2E QoS performance for complex slicing-enabled networks. Such ability ensures that the QoS requirements are constantly met after any changes performed on slices. For the network operators like BT, the network DT can be treated as a virtual model that mirrors the physical slicing-enabled networks in the real-world. The network slicing DT model designed in the project can generate accurate predictions of E2E slicing latency under various network configurations. The main benefit is that network operators can use the DT to achieve a timely response to dynamic traffic utilisation in networks. Analysing and allocating resources in real-time are crucially important but very hard to address, because collecting large volumes of data is expensive and poses high overhead. The DT model can generate and process its own data to enable near real-time analysis and prediction, and then streams the insights into the network management application. Such ability can be further used to optimise network performance and make long-term network planning. Network operators can simulate hypothetical what-if scenarios and resource allocation to evaluate the resulting performance without affecting the applications running on the physical network slices. We have worked with Exeter Science Centre to produce two infographics, 4 detailed paper summaries (3 for network slicing, 1 for network digital twin) and two YouTube videos in order to engage the general public, which are all presented via a newly designed and populated website. As of 9th January 2022, the YouTube videos received more than 500 views. |
| First Year Of Impact | 2018 |
| Sector | Digital/Communication/Information Technologies (including Software) |
| Impact Types | Societal |
| Description | Exeter-Tsinghua Academic Fellowship |
| Amount | £2,970 (GBP) |
| Organisation | University of Exeter |
| Sector | Academic/University |
| Country | United Kingdom |
| Start | 01/2019 |
| End | 07/2019 |
| Title | A scalable and generalised Graph Neural Networks (GNNs) framework |
| Description | We address the scalability issues through building the GNN-based model to focus on only leveraging information of node features for effective network node embedding. Instead of training the embedding for each node, the model learns a function that generates embeddings by sampling and aggregating features from nearby neighbours. In this way, we don't need to process the information of the complete graph in each training step, but only sample a subset of the neighbours for the training aggregation approach. Therefore, our model can work essentially on huge graphs. To evaluate this, we convert the time series data into hundreds of graph screenshots, i.e., we model the slicing traffic matrices and allocate resources at time slot t as a single graph. We then combine 500 single graphs with different configurations into a huge graph that contains 12000 nodes and 37000 links, and evaluate our developed digital twin model. The generalisation is achieved through an inductive graph framework. A generalised graph model is designed with symmetric and orderless characteristics, and thus are invariant to permutations of inputs, and hence can be suited to various graph topologies. A sampled set of neighbours is also used here to make the model topology-free. To improve the generalisation of the DT, we remove 30 percent of the nodes including the links connecting them from the graph. The DT is trained on the reduced graph with the remaining 70 percent of the nodes from the original graph. To evaluate the generalisation capability of the DT, we apply it to the GEANT2 topology but with different slice deployment strategies that were not seen during the training. In each experiment, we deploy different numbers of slices ranging from 10 to 100 with an interval of 10, and each slice has its own resource utilisation. The slices are deployed in the network using various strategies that are derived from 10 different routing algorithms. |
| Type Of Material | Computer model/algorithm |
| Year Produced | 2021 |
| Provided To Others? | Yes |
| Impact | The model resolves the scalability and generalisation issues of network digital twins and network slicing management. |
| Description | Research collaboration - Albert Zomaya |
| Organisation | University of Sydney |
| Country | Australia |
| Sector | Academic/University |
| PI Contribution | We jointly worked on the research of network slicing. My team's contributions include the design and experiments of the solution. |
| Collaborator Contribution | We jointly worked on the research of network slicing. My partner's contributions are about the input of comments and suggestions. |
| Impact | Xiangle Cheng, Yulei Wu, Geyong Min, Albert Y. Zomaya, Xuming Fang: Safeguard Network Slicing in 5G: A Learning Augmented Optimization Approach. IEEE J. Sel. Areas Commun. 38(7): 1600-1613 (2020) Xiangle Cheng, Yulei Wu, Geyong Min, Albert Y. Zomaya: Network Function Virtualization in Dynamic Networks: A Stochastic Perspective. IEEE J. Sel. Areas Commun. 36(10): 2218-2232 (2018) |
| Start Year | 2019 |
| Description | Research collaboration - HongNing Dai - Hao Wang |
| Organisation | Macau University of Science and Technology |
| Country | Macao |
| Sector | Academic/University |
| PI Contribution | I contribute the research work on edge computing and network slicing. |
| Collaborator Contribution | My partners contribute the research work on network security. |
| Impact | We produced several publications Y. Wu, Y. Ma, H.-N. Dai, and H. Wang, "Deep Learning for Privacy Preservation in Autonomous Moving Platforms Enhanced 5G Heterogeneous Networks," Computer Networks, doi: 10.1016/j.comnet.2020.107743. Y. Wu, H.-N. Dai, and H. Wang, "Convergence of Blockchain and Edge Computing for Secure and Scalable IIoT Critical Infrastructures in Industry 4.0," IEEE Internet of Things Journal, doi: 10.1109/JIOT.2020.3025916. H. Wang, S. Ma, C. Guo, Y. Wu, H.-N. Dai and D. Wu, "Blockchain-based Power Energy Trading Management," ACM Transactions on Internet Technology, accepted. |
| Start Year | 2020 |
| Description | Research collaboration - HongNing Dai - Hao Wang |
| Organisation | Norwegian University of Science and Technology (NTNU) |
| Country | Norway |
| Sector | Academic/University |
| PI Contribution | I contribute the research work on edge computing and network slicing. |
| Collaborator Contribution | My partners contribute the research work on network security. |
| Impact | We produced several publications Y. Wu, Y. Ma, H.-N. Dai, and H. Wang, "Deep Learning for Privacy Preservation in Autonomous Moving Platforms Enhanced 5G Heterogeneous Networks," Computer Networks, doi: 10.1016/j.comnet.2020.107743. Y. Wu, H.-N. Dai, and H. Wang, "Convergence of Blockchain and Edge Computing for Secure and Scalable IIoT Critical Infrastructures in Industry 4.0," IEEE Internet of Things Journal, doi: 10.1109/JIOT.2020.3025916. H. Wang, S. Ma, C. Guo, Y. Wu, H.-N. Dai and D. Wu, "Blockchain-based Power Energy Trading Management," ACM Transactions on Internet Technology, accepted. |
| Start Year | 2020 |
| Description | Research collaboration - Niccolo Tempini - Hao Yin |
| Organisation | Tsinghua University China |
| Country | China |
| Sector | Academic/University |
| PI Contribution | My team's contributions include the development of technical solutions to social problems, i.e., considering the impact and factors of social problems into the design of technical solutions e.g. machine learning models. |
| Collaborator Contribution | Tempini's contributions are about the input on social problems. Yin's contributions are on big data analysis. |
| Impact | We jointly supervised PhD students and published the paper below Z. Huang, Y. Wu, N. Tempini, H. Lin, H. Yin, "An Energy-efficient and Trustworthy Unsupervised Anomaly Detection Framework (EATU) for IIoT," ACM Transactions on Sensor Networks, doi: 10.1145/3543855. |
| Start Year | 2020 |
| Description | Research collaboration - Niccolo Tempini - Hao Yin |
| Organisation | University of Exeter |
| Department | Department of Sociology, Philosophy and Anthropology |
| Country | United Kingdom |
| Sector | Academic/University |
| PI Contribution | My team's contributions include the development of technical solutions to social problems, i.e., considering the impact and factors of social problems into the design of technical solutions e.g. machine learning models. |
| Collaborator Contribution | Tempini's contributions are about the input on social problems. Yin's contributions are on big data analysis. |
| Impact | We jointly supervised PhD students and published the paper below Z. Huang, Y. Wu, N. Tempini, H. Lin, H. Yin, "An Energy-efficient and Trustworthy Unsupervised Anomaly Detection Framework (EATU) for IIoT," ACM Transactions on Sensor Networks, doi: 10.1145/3543855. |
| Start Year | 2020 |
| Description | Research collaboration - Victor Leung |
| Organisation | University of British Columbia |
| Country | Canada |
| Sector | Academic/University |
| PI Contribution | We jointly work on the research of deep reinforcement learning for blockchain in industrial IoT. My team's contribution is about the networking and communications. |
| Collaborator Contribution | We jointly work on the research of deep reinforcement learning for blockchain in industrial IoT. My partner's contribution is about the consensus mechanism. |
| Impact | Yulei Wu, Zehua Wang, Yuxiang Ma, Victor C. M. Leung: Deep reinforcement learning for blockchain in industrial IoT: A survey. Comput. Networks 191: 108004 (2021) |
| Start Year | 2019 |
| Description | An invited talk at the ISPA'20 workshop - HW |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | The invited talk titled Applying Machine Learning towards Dynamic Resource Allocation and Autonomous Management in Network Slicing is based on the research results of this project. The talk has raised the international awareness of the work and engaged with wide audiences. |
| Year(s) Of Engagement Activity | 2020 |
| Description | Exeter Maths School - Network Digital Twins |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Schools |
| Results and Impact | The concept of Digital Twin (DT) and Graph Neural Networks (GNNs) have been organised and form a challenging project for the students at Exeter Maths School. It gives the students an early insight into the world of research and a chance to learn utilising machine learning algorithms to solve the realistic problems. We bring these students together, to make them to work as a team, to provide stimuli that will deepen and broaden their understanding of computer science and its related fields and provide them with a unique and invaluable experience. We also support and encourage their development as problem-solving scientists, especially for the girls. By the end of this project, the students explore and understand 5 challenging tasks. 1) Understand the fundamental concepts of DT and the process of interaction and information exchange between the physical system and virtual models. 2) Discover different types of deep neural networks of machine learning, including Convolutional Neural Networks (CNNs), Recurrent Neural Networks (RNNs) and GNNs. 3) Design and implement a DT of computer networks using one deep learning model. The designed DT will need to be able to predict various key performance metrics of the network. 4) Perform network optimisation by leveraging the DT, to improve the performance or maximise the resource utilisation according to the corresponding QoS demands. 5) Write a professional-quality article and make the corresponding presentation to introduce your findings to the audiences. |
| Year(s) Of Engagement Activity | 2020 |
| Description | Exeter Maths School - Network Slicing |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | Regional |
| Primary Audience | Schools |
| Results and Impact | A part of the research has been organised and form two Year-12 projects for the students at Exeter Mathematics School. We encourage the students to use stochastic optimisation and machine learning (especially deep learning), to solve the resource optimisation problems in 5G network slicing. In particular, Reinforcement Learning (RL), a branch of machine learning has gained tremendous attractions recently. Students are encouraged to 1) learn the concept of Deep Reinforcement Learning (DRL), which enables RL to scale to the decision-making problems that were previously intractable, 2) investigate the challenges in 5G network slicing, and 3) design a mechanism that learns to manage and allocate resources directly from interaction with the network based on DRL. Through the project, students are expected to grasp the related concepts of stochastic optimisation, machine learning (deep learning and reinforcement learning), and network slicing (virtualisation technologies, SDN and NFV). They will implement their own DRL-based optimisation algorithms and obtain the experimental results to evaluate the performance of the developed solution. A formal research report and presentation will be generated after the project. The project will deepen and broaden the insights of the students in science, bringing them together to work as a team and providing them a unique and invaluable experience. |
| Year(s) Of Engagement Activity | 2018,2019 |
| Description | International Conference HPCC |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Industry/Business |
| Results and Impact | Present the resource outcome on using deep learning for service function orchestration. |
| Year(s) Of Engagement Activity | 2019 |
| Description | Invited Talk at the ISPA'20 workshop |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | Give a talk on Digital Twinning for Intelligent Network Management |
| Year(s) Of Engagement Activity | 2020 |
| Description | Invited Talk at the TrustCom'20 workshop |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | Give a talk on Robust Anomaly Detection in Networked Systems |
| Year(s) Of Engagement Activity | 2020 |
| Description | Invited talk at DT and IoT Forum |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Postgraduate students |
| Results and Impact | About 100 PhD students and academics attended the talk, which trigger many follow-up discussions and potential collaborations. |
| Year(s) Of Engagement Activity | 2022 |
| Description | Invited talk at UbiSec 2022 conference |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Postgraduate students |
| Results and Impact | This is an invited talk at an international conference, where triggered follow-up questions and discussions. |
| Year(s) Of Engagement Activity | 2022 |
| Description | Invited talk at Xidian 111 programme |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Postgraduate students |
| Results and Impact | About 100 postgraduate students and academics attended the talk, which trigger follow-up discussions and potential research collaborations. |
| Year(s) Of Engagement Activity | 2021 |
| Description | Invited talk at the International Workshop on Emerging Trends Issues, and Challenges in Computer Science and Applications |
| Form Of Engagement Activity | A talk or presentation |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Professional Practitioners |
| Results and Impact | Give a talk on Intelligent Networking: Current Status, Trend, and the Future |
| Year(s) Of Engagement Activity | 2021 |
| Description | Organising MECN-2019 Workshop |
| Form Of Engagement Activity | Participation in an activity, workshop or similar |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Postgraduate students |
| Results and Impact | The workshop on Multi-access Edge Computing and Networking (MECN-2019) is organised to promote the research outcomes and impact of the project. The workshop is held in conjunction with the 18th IEEE International Conference on Ubiquitous Computing and Communications (IUCC-2019) between 21 to 23 October 2019, in Shenyang, China. The workshop provides an international forum for scientists, engineers and researchers to discuss and exchange novel views, results, experiences and work-in-process regarding all aspects of edge computing and 5G technologies, as well as to identify the emerging research topics and open issues. |
| Year(s) Of Engagement Activity | 2019 |
| Description | Public engagement - website |
| Form Of Engagement Activity | Engagement focused website, blog or social media channel |
| Part Of Official Scheme? | No |
| Geographic Reach | International |
| Primary Audience | Media (as a channel to the public) |
| Results and Impact | The outputs of the activity are two infographics, 4 detailed paper summaries (3 for network slicing, 1 for network digital twin) and two YouTube videos, which are all presented via a newly designed and populated website. |
| Year(s) Of Engagement Activity | 2021 |
| URL | https://wu-lab.exeter.ac.uk/outreach/ |