|
Mapping Linguistic Diversity Across NLP Research AreasPrinciple Investigator: Tharindu RanasingheResearch Profile Project DescriptionThis project will investigate how well different human languages are represented across research in Natural Language Processing (NLP), the field concerned with enabling computers to understand and generate human language. Rather than treating a language as simply “high-resource” or “low-resource”, we will examine how its level of research attention differs across areas such as machine translation, speech, dialogue, language modelling and legal NLP. We will analyse more than 60,000 research papers published over the last decade using large AI models to identify the languages and research areas studied in each paper.
|
|
GPU-Accelerated Surrogate Modelling for Complex SimulationsPrinciple Investigator: Congye WangResearch Profile Project DescriptionMany scientific and engineering problems rely on computer simulations that can take a long time to run, making large-scale analysis difficult or impractical. This project develops surrogate models: fast, data-driven approximations that learn how these expensive simulations behave and can then provide accurate predictions at a fraction of the computational cost. The research will explore modern machine-learning methods for building reliable surrogate models and will assess how well they perform across a range of complex problems. High-performance computing will be used to generate simulation data, train the models, and test them at scale. The long-term aim is to make computationally intensive scientific analysis faster and more accessible.
|
|
Molecular simulations of biomolecular interactions and assembliesPrinciple Investigator: Jamshed AnwarResearcher Profile Project DescriptionThe project aims to develop a fundamental understanding of biomolecular assemblies using molecular simulation. The interest is in self-assembly, phase transformations, interactions between assembled structures, and how assembled structures can be perturbed. Self-organised molecular aggregates play a key role in biology, pathology (disease), and nanotechnology applications. Indeed, all biological structures from biomolecular complexes, membranes, organelles, and upwards rely on molecular self-assembly. Soft-matter aggregates are also of technological interest and include micellar solutions and microemulsions, free-standing and supported lipid layers, lipid vesicles (currently being used for the delivery of mRNA-based vaccines), and block co-polymer complexes that are being exploited in molecular sensing devices. Publications
|
|
GPU-isation of the CASINO quantum Monte Carlo codePrinciple Investigator: Neil DrummondResearcher Profile Project DescriptionThe structural, electronic, and optical properties of materials are largely determined by the behaviour of the electrons that bind their atoms together. The fundamental equations describing electrons in materials have been known since the 1930s, but they are extremely challenging to solve. Quantum Monte Carlo methods, such as those implemented in our code CASINO, provide brute-force, approximate numerical solutions to those equations. However, computational expense is one of the major limiting factors on what can be achieved in quantum Monte Carlo calculations. This project aims to ensure that CASINO can efficiently exploit modern computer hardware such as graphics processing units.Publications
|
|
Re-thinking the use of VAEs in the Bayesian Optimisation of Structured SpacesPrinciple Investigator: Henry MossResearch Profile Project DescriptionScientists in expensive fields like drug discovery, materials science, and engineering can now run many experiments quickly and use powerful computers to tackle bigger challenges than ever. Generative AI—software that can create new images, molecules, or even engineered structures—could transform how experiments are planned, performed, and refined. This project is developing the key methodology needed to unlock Generative AI’s full potential in designing experiments, helping speed up innovation for both science and industry.
|
|
Efficient Point Cloud Video ProcessingPrinciple Investigator: Hossein RahmaniResearch Profile Project DescriptionWe are exploring efficient point cloud video processing methods. As 3D sensors such as RGB-D cameras and LiDARs have become more affordable and widespread, 3D point cloud video processing has become an important and hot research topic. Compared to 2D images, 3D data contains rich geometric information, which could give the model a better understanding of the scene. The Point Cloud is a commonly used 3D data format since it keeps the original geometric information of the 3D data. Therefore, it is used in many critical scene-understanding tasks such as robotic and automatic driving systems. In these applications, the model always needs to give the result or make the decision in a limited time with limited computational resources. Therefore, we need to explore efficient point cloud video processing methods for these real-world applications' requirements.
|
|
Exploring cross-modal training for low-resource language modelsPrinciple Investigator: Hossein RahmaniResearch Profile Project DescriptionLow-resource languages present a challenge for NLP because most training requires large quantities of well-annotated data. Our training explores an approach to train language models for many languages on minimal data using images instead.
|
|
Monitoring of the evaluation metrics during the task of French Financial Narrative SummarizationPrinciple Investigator: Nadhem ZmandarResearch Profile Project DescriptionThis projects Monitors the behaviour of different summarisation evaluation metrics during the task of French Financial Narrative Summarization
|