Morteza Mardani
Morteza Mardani, Ph.D.
Email: mormardani [@] gmail [DOT] com
Google Scholar
Twitter
About me
I am currently a senior AI researcher at NVIDIA research and a visiting researcher at Stanford. I am aiming to bridge the gap between theory and practice of machine learning with a focus on generative learning. My background is in statistical learning and optimization.
Prior to NVIDIA, I was a research staff and postdoc at Stanford, working with John Pauly, David Donoho, and Shreyas Vasanawala. I received my PhD from the University of Minnesota, Twin Cities, under the supervision of Georgios Giannakis. As part of the program, I was a visiting scholar at UC Berkeley, RISE Lab, working with Michael Mahoney.
Research
I am interested in fundamental research on generative learning with product impact. I am currently focused on diffusion models. In the past, I worked on architecture design for deep learning, understanding inductive bias of neural networks, and solving inverse problems using deep learning. For my PhD, I contributed to large-scale data science and received the “Young Author Best Paper Award” from IEEE Signal Processing Society in 2017.
Opportunities
I am looking for motivated PhD interns and research scientists interested in generative learning to join our team. Please drop me an email at mmardani@nvidia.com.
Latest news
Our work Corr-diff to learn generaitve mapping between low-resolution and high-resolution states using residual diffusion models appeard on arXiv.
Our papers on RED-diff for MRI and blind MRI accepted for NeurIPS-2023 workshop on learning for inverse problems.
Our RED-diff paper for solving inverse problems using diffusion models appeared on arXiv. Code is released.
Our AutoSamp work for auto-encoding MRI sampling appears on arXiv. Code is released.
I have been elected as a technical committee member for IEEE computational imaging.
Our work on loss-guided diffusion models accepted for ICML 2023.
Our work on pseudo-inverse guided diffusion models (PGDM) accepted for ICLR 2023.
Our work on multiscale attention via Wavelet neural operators appeared on arxiv.