The research team has been running projects for Exeter Mathematics School A-level students since 2018. These projects are designed to challenge the students, and give them an insight into the research methods used in our Lab.
In each project, the students learn the basics of the digital 5G infrastructure, and then solve a particular problem relating to network slicing and digital twin technology. At the end of the project, they present their work in the form of a report and a public presentation. Read more about each project below:
Project 1: Mathematical modelling and optimization for network routing in 5G
In this project, the students are tasked to create a computer routing model, to understand how information is sent around a telecommunications network to different destinations. To do this, they learn about and use a mathematical tool called “combinatorial optimization”, which helps to find the optimal (shortest, fastest) path through the network.
The students then test and improve their model to work efficiently even when uncertainties are present. A real 5G network has lots of uncertainties, such as: how many users will need the system at once, what are their service requirements,… So, this project is a good introduction to the technical challenges of making a 5G network.
Project 2: Resource optimization of network slicing in 5G
This is a mini-project related to our research in 5G network slicing, most suitable for students who are already familiar with programming languages. The students start by learning about the principles of 5G, network slicing, machine learning and AI techniques.
The students then have the challenge of putting all this together, to first create a network slicing model, and then use DRL to use the network resources most effectively. This is an opportunity for students to try out a multidisciplinary research challenge; using techniques from artificial intelligence and applying it to telecommunications networks.
Project 3: Artificial Intelligence-based Digital Twin Models for Network Management
This mini-project is related to our work in designing digital twins, most suitable for students with programming experience and an interest in neural networks (used in machine learning). The students start with building their understanding of how a digital twin models the physical network, and also learning about the different types of neural networks.
The students then choose their preferred neural network to help create a digital twin that learns from historical data, makes predictions and tries to optimise the allocation of resources. This is the most challenging project we run, particularly interesting for students who would like a deep understanding of machine learning techniques.
Project 4: Explainable Anomaly Detection in Industrial Internet of Things
This project aims to investigate the anomaly detection methods in IIoT and dive into explainable machine learning models to explain anomaly detection. To do this, students start with discovering supervised and unsupervised anomaly detection methods with relevant IIoT datasets, and understanding of explainable machine learning models.
Then students choose their preferred anomaly detection model and explainer to design an explainable anomaly detection framework, which requires high anomaly detection accuracy and correspondingly detection explanations. This project makes a good introduction to the concept of ‘explainability’ in machine learning, which lead students to have a critical thinking on the trustworthiness of AI in real life implementation.
Project 5: Explainable Reinforcement Learning for Atari Games
This mini-project is related to our work in ethical autonomous systems in terms of explainable reinforcement learning in OpenAI Gym environment. Students start with investigating the reinforcement learning methods (such as A3C and Deep Q-Learning) for OpenAI Gym Atari environment and understanding the reasons and methods of using explainable reinforcement learning models.
The students then have the challenge of designing and implementing an explainable reinforcement learning model using their preferred Atari game and suitable explainer. The designed model needs to be able to finish the game successfully and generate explanations. This is a challenging project, but provokes students’ interest on understanding of how AI makes decisions.