VALIANT Deeper Dive is a virtual seminar series that highlights recently published research aligned with the center鈥檚 mission in artificial intelligence and translational computational science. Each session features an invited author presenting an in-depth exploration of a recent publication. Talks emphasize the problem context, technical approach, key findings, and broader implications of the work, followed by moderated discussion and audience engagement.
Upcoming Seminars:
The VALIANT Deeper Dive seminar series has concluded for the semester. Thank you to everyone who joined us. Stay tuned for new sessions when programming resumes in the upcoming academic year.
Previous Seminars:
Apr. 28th: "Targeting Model Inversion from Generative Models"
Presented by Mingxing (Ethan) Rao
In this seminar, Ethan presented his latest work on Targeting Model Inversion from Generative Models, examining how sensitive training data may be reconstructed from large-scale generative models such as diffusion and flow-based architectures. This research shed light on emerging risks related to data privacy, copyright, and model security, while offering new perspectives on how to identify and mitigate vulnerabilities in modern AI systems.
Mingxing (Ethan) Rao is a fourth-year Ph.D. student in Computer Science at 自慰视频 University, advised by Prof. Daniel Moyer. His research interests primarily lie in Generative Models, Model Inversion, Membership Inference, and Monocular Depth Estimation. He is broadly interested in adversarial attacks and defenses. His current research focuses on inverting training datasets from pre-trained generative models, specifically diffusion-based and flow-matching architectures. Additionally, he contributes to surgical robotics through research on Monocular Depth Estimation. (Personal website:聽)
Mar. 18th: "Integrating Multimodal Imaging and Non-Imaging Data via Graph Learning"
Presented by Weifeng Yu
In this seminar, Weifeng presented Integrating Multimodal Imaging and Non-Imaging Data via Graph Learning, showcasing how advanced machine learning techniques can combine brain imaging and behavioral data to uncover new insights into neurodevelopment and mental health. This work highlighted the growing role of AI in advancing psychiatric research and improving our understanding of the human brain.
Weifeng Yu is a Research Assistant at the University of Virginia School of Data Science. His research explores how computational neuroimaging and AI-driven methods can deepen our understanding of brain connectivity, behavior, and psychiatric disorders.
Feb. 24th: "Harmonizing Brain MRI Across Sites Without Paired Data"
Presented by Mengqi Wu
In this session Menqi presents his latest work on Harmonizing Brain MRI Across Sites Without Paired Data, introducing UMH-a groundbreaking approach that overcomes traditional barriers in multi-site MRI harmonization by using an innovative image style-guided latent diffusion model. This method not only improves cross-site image alignment but also preserves critical biological features, paving the way for more robust AI diagnostic tools.
Mengqi Wu is a Ph.D. Candidate in Biomedical Engineering at UNC Chapel Hill. He conducts cutting-edge research in AI-driven neuroimaging harmonization, developing novel deep learning frameworks that enable large-scale, multi-site analyses vital for advancing diagnostics of neurodegenerative disorders.
Feb. 3rd: "Distribution-Guided Multi-Tracer Brain PET Synthesis from Structural MRI with Class-Conditioned Weighted Diffusion"
Presented by Minhui Yu
In this session, Minhui presents her work on Distribution-Guided Multi-Tracer Brain PET Synthesis from Structural MRI with Class-Conditioned Weighted Diffusion. She will detail a novel normalized diffusion framework (NDF) that generates high-quality PET images across multiple tracers, addressing critical issues of cost, radiation exposure, and tracer availability. This cutting-edge approach leverages class-conditioned diffusion models and distributional constraints to ensure accurate and consistent image synthesis, with promising results validated on a large multi-subject dataset.
Minhui Yu is a Ph.D. candidate in Biomedical Engineering at UNC Chapel Hill. Her innovative research tackles key challenges in neurodegenerative disease diagnosis by synthesizing multi-tracer brain PET images from structural MRI data using advanced generative deep learning models.
Got Questions?
Contact Lianrui Zuo (lianrui.zuo@vanderbilt.edu) or Yihao Liu (yihao.liu@vanderbilt.edu).