PEOPLE
LEONARDO STELLA - DLC Research Group
OPENINGS
I am currently looking for PhD students that want to work with me in one of the following topics:
Game Theory: theory and applications of game theory, evolutionary game theory and mean-field game theory. Examples include bio-inspired models, consensus, cooperation and competition, rational and irrational players.
Reinforcement Learning and Multi-Agent Systems: assessing generalisability, robustness, efficiency and scalability of multi-agent reinforcement learning algorithms in control and games.
Data-driven control: use of data-driven approaches to study, design and control dynamical systems.
Applicants for a PhD position should hold a BSc and/or a Master degree in relevant areas, have a strong motivation for research in one of the above topics and proved academic performance or relevant project experience.
We offer an excellent research environment within the School of Computer Science and a strong supervisory team, with great opportunities to grow.
If you are interested, please send me an email with your CV and transcripts. Also, a short project proposal would be an indication of your motivation to conduct research in one of these areas.
POSTDOCTORAL RESEARCHERS
Dr Aditya Acharya
Project: HuMaT - Reinforcement Learning for Human-Agent Teams.
The focus of this project is look at multi-agent reinforcement learning (MARL) to improve and automate human-agent teaming. This project is in collaboration with the Alan Turing Institute and ARL.
Dr Claire Palmer
Project: Automating Deployment of Virtual Reality Training Solutions.
The project focussed on automating the creation of information models to support virtual reality training solutions. The project was sponsored by Innovate UK and was in partnership with Bloc Digital, a leading company in XR solutions located in Derby.
Dr Claire Palmer is now a research associate at Loughborough University.
CURRENT PHD STUDENTS
Mr Aju Ani Justus
Project: HuMaT - Reinforcement Learning for Human-Agent Teams.
First supervisor: Prof Christopher Baber.
Reinforcement Learning (RL) has achieved exceptional success in recent years. Recently, a prominent area of research involves the extension to multi-agent reinforcement learning (MARL). The focus of this project is on building reliable and robust human-agent teams. In particular, the aims are: i) explainability – of the agents to the human, and of the human to the agents; ii) team behaviour in multi-human and multi-agent teams.
Mr Ziyue Chu
Project: Provably efficient model-free episodic MARL for collective decision-making, with application in swarms of drones.
Second Supervisor: Prof Christopher Baber.
For multi-agent reinforcement learning systems to learn complicated behaviour, several millions of interactions are frequently needed. Furthermore, the ability to generalise the acquired behaviour is typically limited. The combined difficulties with generalisation and sample efficiency significantly restrict the potential uses of multi-agent reinforcement learning. The goal of this research is to enable agents to learn robust, reusable skills that transfer to different contexts and to learn effective behaviours with less data by utilising distributed information and the multi-agent aspect of such systems. On the application side, the project aims to provide control of a swarm of drones for collective decision-making.
Mr Tuo Zhang
Project: Developing Scalable Multi-Agent Reinforcement Learning Algorithms through Game Theory.
Second Supervisor: Prof Per Kristian Lehre.
MARL is highly promising in solving real-world problems as it learns by sampling from the environment, bypassing the need for explicit problem modelling, which is often infeasible. However, its unpredictability in complex settings hinders its broader application, largely due to the absence of a solid theoretical foundation. Game theory, particularly evolutionary game theory, aligns closely with MARL in form, offering robust tools for theoretical analysis and foundation building. This report includes a literature review, research question statement, completed work, and future plans.
VISITING RESEARCHERS
Mr Chrystian Pool Edmundo Yuca Huanca
Project: A combined multi-agent reinforcement learning and MDP framework for coverage control in swarm of drones.
Supervisors: Dr Gian Paolo Incremona, Prof Patrizio Colaneri.
Abstract.
Ms Ilaria Lagalante
Project: A physics-based reinforcement learning approach for the characterisation of dross in L-PBF additive manufacturing.
Supervisors: Prof Moataz Attallah, Dr Diego Giovanni Manfredi, Prof Mariangela Lombardi.
Abstract.
GRADUATE STUDENTS
MSc Projects - Summer 2024
Mr Alan Binu
Title: Developing a simulation framework for coverage control using MARL in swarm robotics.
Mr Kieran Brevern
Title: Agent-based modelling for collective decision-making in honeybees.
Ms Nicoletta Lambrou
Title: Physics-based NNs for accurate measurements in additive manufacturing.
Ms Pei-Tzu Lee
Title: A web application for process parameter optimisation in additive manufacturing.
Mr Sam Robbins
Title: A reinforcement learning approach for the characterisation of dross in additive manufacturing.
Mr Thomas Ross
Title: A game theoretic framework to solve the changing lane problem in autonomous driving.
Mr Mastan Sharief Shaik
Title: Cascading failures in financial markets via mean-field games.
Mr Amanjot Singh
Title: A multi-agent reinforcement learning approach to minimise cascading failures in financial markets.
Mr Filip Szczepanski
Title: A reinforcement learning framework with enthalpy-based reward functions for process parameter optimisation in additive manufacturing.
MSc Projects - Summer 2023
Mr Thomas Astley
Title: Using Machine Learning to optimise Laser Powder Bed Fusion process parameters for Dimensional Tolerance.
In this dissertation, machine learning combined with image analysis is used to obtain predictions for optimum values for two parameters of the laser powder bed fusion process, beam compensation (BC) and contour distance (CD), when A20X powder is used. The main impact of these parameters is on the dimensional tolerance and accuracy of the smaller sections of the resulting product.
Mr Oliver Halliday
Title: Using Machine Learning to Optimise Process Parameters in Additive Manufacturing.
This project aims to reduce the need for lengthy experimentation in additive manufacturing by producing a machine learning (ML) algorithm that utilises a limited amount of training data and is consequently able to predict the relative density of a print, as a measure of its quality.
Mr Ahmed Muadh Faizan Mohamed
Title: Process Parameter Optimisation in Additive Manufacturing: A Reinforcement Learning Approach.
The aim of this project was to develop a reinforcement learning approach to optimise process parameters for metal additive manufacturing, in collaboration with Francesco Careri and Prof Moataz Attallah.
Miss Madhurima Sarkar
Title: Optimal Financial Investments to Mitigate Cascading Failures.
This study enlightens the broader effects of using artificial intelligence in the financial context. We use a model-free Q-learning approach to mitigate against cascading failures. Our objective is to develop investment strategies in the form of portfolio holdings that would be optimised through a thorough analysis of past performance and potential rewards or penalties.
Mr Aiden Swartzberg
Title: Multi-Agent Reinforcement Learning for Mitigating Cascading Failures in Financial Networks through Adjusting Equity Goals.
This project aims to provide a solution to mitigate systemic financial collapse by offering an automated approach for defining the equity goals within a network of interconnected institutions. We introduce a novel cooperative multi-agent reinforcement learning model designed to imitate the role of a central governing body, which sets the equity requirements of each company at different time periods.
Mr Haoming Zhang
Title: Network Model Analysis in Financial Contagion.
Financial contagion is a crucial issue and its consequences can lead from the bankruptcy of a single institution to the onset of a financial crisis. This project studies the properties of the Watts-Strogatz network model and the Barabási-Albert network model. This project primarily focuses on the discussion and analysis of the effects of varying share-holding weights among multiple organisations in different network models.
MSc Projects - Summer 2022
Mr Jishag Azhikodan Chenarath
Title: Winter – The Medical Bot. A virtual buddy to keep you in company.
The aim of this project was to develop an intelligent system or a bot that could act as a virtual assistant and a front-line defence against growing mental health issues, focusing on user-based information.
Mr Daniel Fitzpatrick
Title: Applying Machine Learning to Analyse Poker Gameplay.
In this project, a set of dynamical Logistic Regression and Linear Regression models are compared, to see if we can predict other players' approximate hand qualities and aggression throughout a game in real-time. The models are trained utilising historical data from each player on the table.
Mr Kishan Odedra
Title: Investigating pedagogical approaches for spatial and logical reasoning tasks in virtual reality environments.
The report aims to construct and compare a virtual reality application which teaches a learning task to conventional pedagogical mediums. Unity and Oculus Interaction SDK are the tools used for the development of the prototype. The learning task is primarily facilitated through a visualisation of an algorithm and subsequent testing of knowledge with immediate feedback.
Mr Conor Tansey
Title: EscapeNow A new social network built to connect aspiring travellers.
The goal of this project was to build a full stack social network web application that facilitated users in finding other people to travel and explore the world with as well as sharing their own adventures and being inspired by others travel experiences.
Mr Henry Taylor
Title: A Comparative Analysis of Q-Learning, Minmax and Monte Carlo Tree Search for Connect-4 Artificial Intelligence.
This dissertation aims to answer the question 'What is the strongest algorithm enabling AI to play a game Connect-4?'. Three popular game algorithms (Q-Learning, Minmax, and Monte Carlo Tree Search) are compared to more advanced versions.
Mr Rajesh Kanna Vaidyanathan
Title: Intelligent Connected Autonomous Vehicles Traffic Flow Regulation study based on Unity ML-Agents.
This study describes the process of designing, developing, simulating, and evaluating a Multi-Agent Reinforcement Learning method for an intelligent connected autonomous vehicles traffic flow regulation scenario using the Unity ML-Agents platform.
Miss Shangqing Wei
Title: A Scalable Approach For Cooperative Multi-Agent Reinforcement Learning in Minecraft.
Scalability issue is one of the bottlenecks of multi-agent reinforcement learning (MARL). As the number of agents increases, especially to a large-scale, the state-action space will grow exponentially, which leads to the curse of dimensionality and high computation complexity. Aiming to address the scalability issue of MARL, this project propose a scalable approach for a cooperative task in Minecraft.