Interview iCARE Researcher Showcases Applied Health Data Science at ICML 2025
Research Assistant Joy Li joined thousands of researchers in Vancouver to explore the future of AI.
Joy Li, Research Assistant at iCARE, attended the International Conference on Machine Learning (ICML 2025) in Vancouver, Canada this July. ICML is one of the largest global conferences in artificial intelligence and machine learning, with around 10,000 participants. The event brings together researchers from academia and industry to share cutting-edge work, with a strong focus on theoretical advances.
Joy presented a poster based on research completed with former colleagues, which applied neural operators to solve physical optimisation problems. The poster session provided the opportunity to explain the approach to peers, answer questions, and receive feedback from the wider AI community.
Among the highlights were the poster sessions, each featuring hundreds of projects. Joy noted strong engagement with work on large language models (LLMs) in healthcare. Examples included a comparative study testing six LLM families on medical reasoning tasks, and BoxLM, a model that integrates medical ontology with electronic health records to improve diagnosis prediction.
“The poster sessions were my favourite part,” Joy said. “Being able to walk up to researchers, hear them explain their work, and ask detailed questions made the whole experience much richer than simply reading papers.”
A key theme across the conference was the rapid expansion of LLMs and their applications. Alongside the excitement, challenges remain around the high computational cost of training and deploying these models. Much of the research presented focused on efficiency techniques such as quantisation, knowledge distillation, and parameter-efficient training.
Joy also observed the momentum behind AI agents—systems that use LLMs to perform tasks step by step—which are moving quickly from research to applications in platforms such as Azure AI Foundry and IBM Watsonx Orchestrate.
Reflecting on the sessions, Joy highlighted the importance of grounding research in reality. “Many projects were trained on synthetic or very clean public datasets, but things get much harder with messy, real-world data. That is where iCARE has a significant advantage. We work with valuable clinical data, and there is real potential to adapt advanced models—like LLMs, causal models, and time-series approaches—to make a difference in the NHS.”
ICARE hosts the healthcare data of 2.7 million people in North West London, spanning Imperial College London and Imperial College Healthcare Trust. Its interdisciplinary team uses deidentified data to drive research for improved patient outcomes as part of the Imperial Biomedical Research Centre, funded by the National Institute for Health and Care Research. Projects include digital diabetes wards, cervical cancer prevention, and Covid-19 vaccination studies.
Joy added, “The biggest takeaway for me is that real-world data is where innovation meets impact. At iCARE, we have the opportunity to bridge cutting-edge methods with the practical challenges of healthcare systems.”
iCARE’s participation at international conferences highlights our role in connecting advanced AI research with the challenges and opportunities of real-world healthcare.
