Roummel Marcia
University of California-Merced, CA·Mathematics & Statistics
Many imaging and sensing problems mean recovering a clear signal or image from noisy, indirect measurements. Deep neural networks can do this well but usually need large training sets.
- Fields
Mathematics & Statistics
Computer Science
Education Research
- Topics
data science
machine learning
deep learning
- Grants
- 1 known grant, through Jun 2027
What they work on
- Extend deep unrolling networks for signal recovery and image reconstruction by training them with quasi-Newton optimization methods that exploit previous iterates and gradients.
- Investigate loss functions that better reflect the noise distribution of measurements when training physics-informed neural networks.
- Develop optimization techniques for deep unrolling with nonlinear physical sensing models.
- Train graduate students in machine learning and optimization research and create undergraduate research projects and a linear algebra course for a data science major.
More about the research
In many physics-based problems, scientists already know the math describing how measurements are made. This is called the observation operator. Building that knowledge into the network can make learning more efficient.
Why it matters
Better methods for reconstructing images and signals from noisy measurements could benefit physics-based imaging and sensing applications. By incorporating known physics, these methods could need less training data. The program also prepares undergraduate and graduate students from underserved communities for data science careers in academia, industry and government.
Known grants
From NIH and NSF research grants. Labs often have other funding we can't see.
Closest labs
Andrei Draganescu
University of Maryland-Baltimore County, MD·Mathematics & Statistics
Develop low-rank gradient flow (LRGF), a first-order optimization method that builds quadratic approximations with low-rank Hessians and computes their gradient flow exactly, ending each step with a Wolfe-condition line search.
Jonathan Tamir
University of Texas at Austin, TX·Engineering·Early-career
Develop deep generative physical models that pair known imaging physics with image patterns learned from data, so computational imaging systems can adapt to dynamically changing operating conditions.
Xin Li
Texas A&M University-College Station, TX·Engineering
Develop physics-informed neural networks, AI models constrained by wave mechanics, that estimate nearshore bathymetry, waves and currents simultaneously from optical and lidar images of the surf zone.
The school
University of California-Merced: 3 labs on PI Pages, $77M a year on research, 112 research doctorates a year.
Reach out
A start for your email. Fill in the brackets.