Harry D. Sun

Research Interests

Vision Language Model: Medicine frequently presents challenges that are inherently multi-modal, requiring both textual and visual understanding. My research in Vision Language Modeling seeks to enhance the AI's capability to interpret and act upon complex medical scenarios presented through images and textual prompts. Beyond this, I'm delving into building clinician trust in AI through transparent and reliable model behaviors.

AI-based Diagnosis and Discovery: With the increasing complexities in medical diagnostics, I'm investigating the potential of AI to match or even surpass clinician accuracy in certain diagnostic tasks. My recent endeavors have led to the development of advanced tools tailored for leukemia detection and diagnosis, offering promising avenues for rapid and accurate disease identification.

Synthetic Data & Generative AI: Facing both scarcity and privacy concerns in medical data, I view synthetic data as a transformative solution. Yet, its true potential is unlocked only when meeting clinical standards. I've pioneered human-in-the-loop techniques to achieve this, and I'm enthusiastic about pushing boundaries in this domain further.

Self Supervised Learning in Medical Images: The intricacies and sparse annotations of medical imaging demand an approach beyond traditional supervised techniques. Venturing into self-supervised learning, I'm delving into the potential of discerning complex patterns in the absence of annotations. My interest lies in developing robust representation learning techniques, which can serve as a foundational bedrock when bridging image modality with other modalities such as clinical notes, electronic health records, genomic sequences, and wearable device data.