Research projects in Information Technology

This project focuses on simulating the organic growth of criminal communication networks by leveraging techniques such as Reinforcement Learning and Graph Theory. The goal is to curate a synthetic dataset that models the evolving structure and dynamics of illegal networks, taking into account factors like social connections, communication patterns, and resource allocation. By using graph-based models, the project aims to create realistic representations of how criminal groups form, expand, and operate under various conditions.

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.

Background and Motivation

Modern deep learning models have achieved remarkable success in computer vision and natural language processing. However, they typically produce overconfident predictions and lack reliable mechanisms to quantify uncertainty. This limitation becomes particularly problematic in high-stakes applications, such as healthcare diagnosis, autonomous systems, and scientific discovery.

Project Background & Motivation

Project Background and Motivation

The "inverse design" of molecules from analytical spectra (such as MS2, NMR, or IR) is a fundamental bottleneck in analytical chemistry, metabolomics, and drug discovery. While deep generative models have shown promise in proposing novel molecular structures, they typically require massive, cleanly labelled datasets to train effectively.

We are seeking a motivated PhD candidate to work on unsupervised music emotion tagging within the broader field of affective computing. The project aims to develop reproducible machine learning approaches for automatic emotion recognition in music, with stronger theoretical grounding, transparent model implementation, and rigorous validation.

Using the Project-1 hiPSC platform, this project builds AI pipelines to learn disease-relevant representations from cellular images, fused with multi-omics. Models will classify diagnosis and predict treatment response with strict donor-level splits, cross-regional external validation, and fairness audits (sex/ethnicity stratification). Interpretable AI (e.g., attribution maps, SHAP) will nominate mechanism-anchored biomarkers and candidate interventions. Tooling will be containerised and open to support reproducibility and clinical translation.

This research focuses on developing and evaluating methodologies for the optimal design of control charts within the framework of Statistical Process Control (SPC). The study aims to determine the best configuration of chart parameters, such as sample size, sampling interval, and control limits, to minimize detection time for process shifts while controlling false alarm rates. It explores both traditional and advanced optimization techniques, including analytical models, simulation-based approaches, and data-driven algorithms.

The project involves building a high-fidelity digital twin of complex urban or environmental systems by integrating GIS data, IoT sensor measurements, and domain-specific information. AI models will be developed to predict dynamic behaviors and key risk zones, and these models will be embedded into the digital twin to enable real-time simulation and visualization. The system’s performance will be evaluated in terms of prediction accuracy, computational efficiency, and usability to ensure it provides actionable insights for decision-making and planning.

The project involves building and curating a comprehensive food image dataset suitable for mobile AI applications. High-accuracy deep learning models will be trained on this dataset and then compressed into lightweight student models using knowledge distillation, enabling efficient real-time inference on mobile devices. The distilled models will be deployed and optimized on mobile platforms, with their performance evaluated in terms of classification accuracy, computational speed, and overall usability.