2.00 - 2.05 Welcome/Chair's introduction
 
2.05 - 2.40 Prof. Grazziela Figueredo (Associate Professor, University of Nottingham) – "Using LLMs responsibly: A case study in OMOP mapping"

2.40 - 3.15 Dr. Eghbal Rahimikia (Assistant Professor, University of Manchester) – "Re(Visiting) Large Language Models in Finance"

3.15 - 3.20 Refreshment break
 
3.20 - 3.55 Ms Xinyi Wang (PhD student, University of Nottingham) – "Deep Learning-Based Radiology Report Generation and the Application of Large-Scale Models" 

3.55 - 4.30 Dr Anh Nguyen (Senior Lecturer, University of Liverpool) – "Physical Intelligence"

4.30 - 4.35 Close

Grazielle Figueredo (Associate Professor, Health Data Science,and School of Computer Science, University of Nottingham)
"Using LLMs responsibly: A case study in OMOP mapping"
In this presentation, we introduce Lettuce, an intelligent open-source tool developed to address the inherent complexities associated with converting medical terms into standard concepts defined by the OMOP Common Data Model. We begin by outlining the technical framework underpinning Lettuce, detailing the methodological steps taken to ensure its modularity, adaptability, and capacity to accurately represent domain-specific knowledge. We then explore the added technical value of the tool, with particular emphasis on aspects such as domain-specific data embedding, model selection, and evaluation. A significant component of our work also concerns the responsible integration of large language models (LLMs) as tools for enhancing human collaboration. To this end, we reflect on ongoing dialogues with relevant stakeholders and the broader community of practice, which have informed our efforts to embed principles of decision provenance, accountability, verification, validation, and explainability to improve Lettuce. Ultimately, the aim is to demonstrate that it is feasible to develop scientifically robust tools, with the human in the loop, that expedite the mapping process and adhere to ethical and safety standards.

Eghbal Rahimikia (Assistant Professor, Alliance Manchester Business School, University of Manchester)
"Re(Visiting) Large Language Models in Finance"
This study evaluates the effectiveness of specialised large language models (LLMs) developed for accounting and finance. Empirical analysis demonstrates that these domain-specific models, despite being nearly 50 times smaller, consistently outperform state-of-the-art general-purpose LLMs in return prediction. By pre-training the models on year-specific financial datasets from 2007 to 2023, the study also mitigates look-ahead bias, a common limitation of general-purpose LLMs. The findings highlight the critical importance of addressing look-ahead bias to ensure reliable results. Extensive robustness checks further validate the superior performance of these models.

Xinyi Wang (PhD Student, Health Data Science,and School of Computer Science, University of Nottingham)
"Deep Learning-Based Radiology Report Generation and the Application of Large-Scale Models" 
Medical reports summarize findings from medical images to aid doctors’ decisions. This presentation draws on the review article "A Survey of Deep-learning-based Radiology Report Generation Using Multimodal Inputs" and recent advances in large AI models. “Multimodal inputs” refer to data from images, patient records, medical knowledge, and related sources. Automatically generating radiology reports from such diverse and complex data is challenging. I will first explain the main challenges in handling multimodal data, then review key studies addressing them. Finally, I will discuss how large AI models enhance report generation and explore their practical benefits and current limitations in clinical use.”

Anh Nguyen (Senior Lecturer, University of Liverpool)
"Physical Intelligence"
As artificial intelligence continues to advance, a critical frontier emerges at the intersection of cognition and physical embodiment - physical intelligence. This talk explores how recent breakthroughs in large language models (LLMs) and generative AI are transforming our understanding and implementation of intelligence in the physical world. By linking symbolic reasoning, learned representations, and motor control, we can begin to endow robots with the capacity to perceive, reason, and act adaptively in complex, dynamic environments. Drawing from interdisciplinary insights across machine learning, control theory, and cognitive science, the talk presents a vision for the next generation of AI-driven robots - systems that are not only computationally intelligent but physically agile, resilient, and context-aware. I will showcase examples where LLMs are used to interpret natural language commands, plan complex tasks, and even guide low-level control in robotic systems, illustrating the emergence of embodied, explainable, and generalizable AI. The talk concludes with key challenges and opportunities in building truly intelligent physical agents that can operate robustly in real-world scenarios.

Contact Gareth Liu-Evans for Merseyside Local Group