Dr Long Tran-Thanh | Electronics and Computer Science | University of Southampton

Research interests

Human-aware AI: My main research focus is on combining machine learning, game theory, optimisation, and incentive engineering to tackle optimisation problems within AI systems caused by strategic and selfish human users.

AI for Social Good: I also apply my core AI research to a number of societal challenges, including:

• Using machine learning (ML) and crowdsourced incentive engineering to develop air pollution monitoring system with low cost mobile sensor devices

• Developing smart devices that can use energy efficient algorithms to learn to detect a number of diseases such as TB, or to predict severe health issues such as asthma attacks.

• Designing intelligent housing management systems for homeless people.

• Applying AI and optimisation techniques for efficient suicide prevention.

• Building intelligent solutions for national/cyber security issues.

I also have 2 projects with my colleagues in Vietnam. One is about building low-cost sensor systems for air pollution monitoring in Saigon (joint work with Hien vo from VGU and Huy-Dzung Han from HUST), and the other one is about building stand-alone intelligent devices for tuberculosis testing (with Cuong Pham from PTIT). Apart from these, I am also interested in applying AI to governance (govtech) and education (edtech).   

Online learning: One of my core research areas is bandit theory. In particular, I investigate multi-armed bandit (MAB) models where pulling an arm (i.e., making a decision) requires a cost and the total spending is limited by a finite budget. To tackle this problem, I have introduced a new model, called the budget-limited MAB, and have also proposed a number of arm pulling algorithms for which I have provided both theoretical and empirical performance analyses. I am also interested in applying this bandit model (or its variances) to other domains of AI, such as: (i) decentralised controlling for UAVs; (ii) information collection in wireless sensor networks; and (iii) budget-limited online keyword bidding.

Game theory: My other core research area is game theory: I mainly focus on large coalition formation games from both game theoretical and decision making perspective. In more detail, I look at systems where the number of participants is very large (typically thousands or more). Within these systems, calculating different solution concepts (e.g., the core, nucleolus, Shapley-value, etc.) are very hard. Given this, my goal is to identify approximation techniques that can efficiently provide high quality results. To do so, with some of my colleagues, we have introduced a novel, vector-based, representation model of the participating agents, with which we can calculate the abovementioned concepts in a significantly more efficient way. We have also analysed the error bounds of approximating the Shapley value in large games.

I also study different games with resource allocation from both aspects of classical and behavioural game theory. In particular, I am interested in calculating different equilibria and price of anarchy. 

From the behavioural game theory perspective, I aim to identify players’ favourite strategies when they repeatedly play such games against different opponents (Repeated Colonel Blotto).

Crowdsourcing: More recently, I investigate the performance of different crowdsourcing systems from a theoretical perspective, aiming to provide rigorous performance guarantees for task allocation algorithms. 

Home energy management: I am heavily involved in the research work on home energy management. In particular, we aim to improve the energy consumption profile of home owners, in order to reduce the CO2 emission of the domestic energy sector. To do so, as the first step, we mainly focussed on the accurate learning and prediction of homeowners’ habit, such as appliance usage and heating preferences. Our results were published at ACM E-Energy 2013 and IJCAI 2013.

I am also interested in how to keep user annoyance at an efficient level while interacting with them. With my collaborators we have developed a number of techniques to achieve this goal, and our findings were published at IJCAI 2016 and AAMAS 2018.

Other research interests:

The cost of interference to closed evolving systems: We investigate what is the cost to interfere into closed systems, if we want the system to achieve some desirable states. As a first step, we look at evolving evolutionary games, where an external decision maker can invest his resources into the system (e.g., via a reward scheme) such that in the long term, the agents will follow a preferred behaviour. A preliminary result has been presented at COIN 2014 and NAG 2014, and our most recent results just got accepted to Nature’s Scientific Reports.

Non-monetary referral incentives: I am also investigating how non-monetary referral incentivisation work in social networks. You can find a preliminary version of our work here. For more details, you can visit the website of our project, or watch a video about it. 

Algebraic topology for machine learning: With my PhD student Tom Davies we are also investigating how to make the application of persistent diagrams and other techniques from algebraic topology more efficient and automated in machine learning systems. Our first result is a fuzzy clustering method for persistent diagrams.