Computer Science PGR Seminar Thu Nguyen and Yunzhe Yuan

All are welcome to listen to our speakers Thu Nguyen and Yunzhe Yuan.

Thu Nguyen will present Longitudinal physical-mental multimorbidity trajectories by sociodemographic factors in Scotland.

Did you know that around one in four adults in the UK live with two or more chronic conditions? This is known as multimorbidity and is becoming increasingly common. It is one of the most pressing challenges facing healthcare systems worldwide now and in the coming years. This talk will be about multimorbidity, its trajectories, the role of sociodemographic factors, and why it demands a different approach to care.

Thu is a PhD student, co-supervised by the School of Computer Science, Medicine, Mathematics and Statistics, and Geography and Sustainable Development. Her main research area is health data science, with emphasis on multimorbidity, integrated physical-mental care, social determinants of health, and health inequality.

Yunzhe Yuan will present Efficient Dependency Parsing via Lightweight Fine-tuning and Constrained Decoding.

Syntactic structure is a fundamental component in many NLP applications, ranging from information extraction to machine translation. While LLMs have recently set new performance benchmarks for structured prediction tasks like dependency parsing, training and deploying such models can be costly and impractical in resource-constrained environments. This talk presents a lightweight alternative that combines pretrained sequence-to-sequence models with explicit structural constraints, ensuring that outputs form valid dependency trees. Experiments on Universal Dependencies demonstrate that compact models, when guided by appropriate inductive biases, can achieve competitive performance comparable to much larger architectures.

Yunzhe Yuan is a PhD student in Artificial Intelligence at the School of Computer Science. His research focuses on syntactic parsing and grammar-based analysis, specifically tailored for low-resource and resource-constrained environments. His current work explores how explicit structural constraints, such as constrained decoding, can optimise the accuracy-efficiency trade-off for dependency parsing.