Seminar Data Analytics III: Synthetic Data and Sim-to-Real Learning for Medical AI

Modern medical AI relies heavily on deep learning for analyzing complex medical data, yet its progress is limited by scarce, noisy, and hard-to-share annotated datasets. Privacy constraints, ethical considerations, and high annotation costs often result in models with limited robustness and poor generalization.
Synthetic data and simulation-based learning offer promising avenues to address these challenges by enabling data augmentation, privacy-preserving model development, and controlled experimentation when real data is scarce or inaccessible. However, discrepancies between synthetic and real-world data distributions introduce domain shifts that can hinder real-world performance. This seminar explores methodological advances that leverage synthetic data while ensuring effective transfer to real clinical settings, with a focus on improving data efficiency, robustness, and generalization in human-centered medical AI applications.

The seminar will cover, among other topics:

-Synthetic Data Generation: Generative models and simulation techniques
-Simulation-to-Reality Learning and Domain Adaptation: Methods for reducing domain gaps between synthetic and real data.
-Learning under Data Scarcity and Privacy Constraints: Data-efficient learning strategies for medical applications with limited or sensitive data.
-Applications in Human-Centered and Clinical AI: Use cases in healthcare monitoring, medical imaging, wearable sensing, and assistive systems.

Students are expected to actively participate by selecting up to three research papers related to the seminar topics, either from a list provided by the instructor or based on their own proposal, subject to approval.
Each student will present the selected papers in a 15-minute oral presentation, followed by a 5-minute question-and-answer session.
In addition, students are required to submit a written seminar report that critically analyzes the selected papers, comparing their methodologies, assumptions, and experimental results within the context of the course.

Format: The first two seminar sessions are kick-off meetings (attendance mandatory). After this, there will be a research phase. During this phase, there will be no regular group meetings with all seminar participants. Instead, students will meet with their supervisor on assigned dates during the regular seminar slot in the seminar room. At the end of the semester, there will be final presentation sessions (attendance mandatory). The number of presentation slots depends on the number of participants.