Research projects in Information Technology

This PhD project will investigate the explainability of reinforcement learning (RL) policies in the context of human-robot interaction (HRI), aiming to bridge the gap between advanced RL decision-making and human trust, understanding, and collaboration. The research will critically evaluate and extend state-of-the-art explainability methods for RL, such as policy summarization, counterfactual reasoning, and interpretable model approximations, to make robot decision processes more transparent and intuitive.

Adaptive sequential decisions to maximise information gain and biodiversity outcomes

Development of a GIS-Based Model for Active Citizenry

Street-Level Environment Recognition On Moving Resource-Constrained Devices

Markov Decision Processes (MDPs) are frameworks used to model decision-making in situations where outcomes are partly random and partly under the control of a decision maker. While small MDPs are inherently interpretable for people, MDPs with thousands of states are difficult to understand by humans. The K-MDP problem is the problem of finding the best MDP with, at most, K states by leveraging state abstraction approaches to aggregate states into sub-groups.

Turning Point is a renowned addiction treatment and research centre specialising in the prevention, treatment, and support services for individuals affected by substance use disorders, gambling addiction, and mental health issues. Turning Point operates a network of 26 helplines across the country, ensuring accessible and immediate support for individuals in need. These helplines serve as a vital resource for individuals seeking assistance, information, and guidance related to addiction and mental health concerns.

Classical propositional logic (CPL) captures our basic understanding of the linguistic connectives “and”, “or” and “not”. It also provides a very good basis for digital circuits. But it does not account for more sophisticated linguistic notions such as “always”, “possibly”, “believed” or “knows”. Philosophers therefore invented many different non-classical logics which extend CPL with further operators for these notions.

Classical propositional logic (CPL) captures our basic understanding of the linguistic connectives “and”, “or” and “not”. It also provides a very good basis for digital circuits. But it does not account for more sophisticated linguistic notions such as “always”, “possibly”, “believed” or “knows”. Philosophers therefore invented many different non-classical logics which extend CPL with further operators for these notions.

In many branches of science (e.g., Artificial Intelligence, Engineering etc.), the modelling of the problem is done through the use of functions (e.g., f(x) = y). On a very high-level, we can think of Machine Learning as the problem of approximating function f from the pair of measurements (x,y), and Optimization as the problem of finding the value of input x that maximizes the output y given function f.

In NeuroDistSys (NDS): Optimized Distributed Training and Inference on Large-Scale Distributed Systems, we aim to design and implement cutting-edge techniques to optimize the training and inference of Machine Learning (ML) models across large-scale distributed systems. Leveraging advanced AI and distributed computing strategies, this project focuses on deploying ML models on real-world distributed infrastructures, improving system performance, scalability, and efficiency by optimizing resource usage (e.g., GPUs, CPUs, energy consumption).