Welcome!
We’re the VOILA Lab at Georgia Tech. Our lab’s research applies techniques from machine learning, signal processing, and optimization to solve inverse problems in computational imaging, including tasks in computer vision, medical imaging, and scientific imaging.
Please see our research areas page for our lab’s recent research, and our papers page for an exhaustive listing of our lab’s work.
Research areas:
Thinking about joining the group?
In general, I plan to hire 1-2 PhD students per year, but that may fluctuate year to year.
Please read the Prospective page for detailed instructions before contacting me.
Our lab has a funded PhD fellowship opening (starting Fall 2027) for research in diffusion models for computational imaging, with application to event-based cameras. Due to funder requirements, this fellowship is restricted to US citizens. If interested, please apply to the DSP research area in the ECE PhD program and list me as a potential advisor.
What strong applicants usually have: a solid foundation in linear algebra, optimization, signal processing, probability & statistics, and algorithms & data structures. Most projects use Python (including GPU and autodiff libraries like PyTorch, JAX, and CuPy), but familiarity with lower-level languages like CUDA or C++ is a bonus
News
October 27th, 2026
Sara will be giving two talks at Asilomar 2026 on accurate, provable, and fast nonlinear tomographic reconstruction and on rate-distortion for compressive signal parameterizations.
October 22nd, 2026
New preprint on going beyond best-of-k evaluation for nonlinear inverse problems, joint with Andreas Geiger, is available on arXiv.
October 22nd, 2026
New preprint on a local out-of-distribution detection framework based on statistical typicality and superpixels, joint with Rohan Sanda, is available on arXiv.
October 5th, 2026
Sara will be giving a talk at the 2026 Georgia Statistics Day on a finite-sample lens on when, why, and how diffusion posterior samplers fail.
September 24th, 2026
Our paper Bounding Global and Local Compression Error of Signal Parameterizations has been accepted to NeurIPS. Congrats Quang!
July 15th, 2026
Sara will be giving a two-part talk in the Mathematics of 3D Reconstruction workshop at ICERM on both rate-distortion for compressive signal parameterizations and a finite-sample lens on when, why, and how diffusion posterior samplers fail.
June 17th, 2026
Our paper 3D Field of Junctions: A Noise-Robust, Training-Free Structural Prior for Volumetric Inverse Problems, joint with Prof. Justin Romberg, has been accepted to ECCV. Congrats Namhoon and Narges!
June 9th, 2026
New website is up!
June 3rd, 2026
Our group will be presenting two papers this week at CVPR: KLIP (main track) and 3D Field of Junctions (workshop). Sara will also be giving talks in the Computational Cameras and Displays and 4D World Models workshops on a finite-sample lens on when, why, and how diffusion posterior samplers fail and rate-distortion for compressive signal parameterizations respectively.
May 28th, 2026
New preprint on predicting the reconstruction error of compressive signal parameterizations without access to ground truth is available on arXiv.