About this Event
Deep representation learning for clinical predictive models
Abstract: The translation of vast datasets into actionable clinical insights is a pervasive pursuit cutting across informatics, machine learning, healthcare, and medicine. The crux of the matter lies in extracting knowledge from multi-modal, multi-source, high-dimensional clinical data—a challenge that necessitates the development of prediction models for clinical decision-making. This thesis research tackles these fundamental challenges through a novel representation learning-based framework, which capitalizes on recent advancements in generative modeling and latent variable models to encapsulate patients' characteristics in a lower-dimensional representation. This representation is semi-supervised by the clinical outcomes of interest to healthcare professionals. Extending the scope of our approach, we delve into more intricate clinical settings where the models estimate individualized treatment effects to identify patients likely to derive maximum benefit from specific treatment decisions. Moreover, our research explores the expansion of representation learning models to diverse data modalities, including the incorporation of large language models to extract information from clinical texts. This dissertation contributes to AI for health, offering innovative solutions to the challenges of knowledge extraction from complex clinical data and facilitating the translation of predictive outcomes into practical clinical insights.
Event Details
See Who Is Interested
0 people are interested in this event