Research

My research lies at the intersection of Bayesian statistics, machine learning, and computational imaging. I develop methods for solving imaging inverse problems from incomplete or noisy measurements, with a focus on uncertainty quantification and reliable decision-making.

Bayesian imaging and inverse problems

I study computational methods for reconstructing images from indirect, noisy, or incomplete measurements, such as those arising in medical and scientific imaging. These problems are typically ill-posed, so uncertainty quantification is essential for reliable interpretation.

Self-supervised uncertainty quantification

A central theme of my work is to quantify uncertainty without relying on large collections of ground-truth images. This is important in applications where ground truth is expensive, unavailable, or biased. I develop conformal prediction methods that self-calibrate directly from observed measurements.

Generative posterior sampling

I am interested in diffusion, flow-based, and variational methods for scalable posterior sampling. The goal is to produce multiple plausible reconstructions and calibrated uncertainty estimates from a single observation.

Current directions

• Self-supervised conformal prediction for imaging inverse problems.
• Posterior sampling with generative models for uncertainty quantification.
• Equivariant and symmetry-aware learning for inverse problems.
• Reliable AI tools for scientific and medical imaging.

Research experience