Skip to main content

Education Center

Webinar: Provable Probabilistic Imaging using Generative Diffusion Priors (video)

SHARE:
Category
Proficiency
Language
Media Type
Pricing

SPS Members $0.00
IEEE Members $11.00
Non-members $15.00

Authors
Date

"Provable Probabilistic Imaging using Generative Diffusion Priors" presented by Dr. Yu Sun.  Based on the IEEE Xplore® article under the same title. Published: IEEE Transactions on Computational Imaging, August 2024. Visit article article here.

Inverse problems are often ill-posed, meaning that multiple images may be consistent with the same set of measurements. As a result, solving an inverse problem requires not only recovering a high-quality image but also characterizing the ambiguity inherent in the reconstruction process. Diffusion models (DMs) have recently emerged as powerful image priors, delivering remarkable performance across a wide range of imaging applications. However, many existing DM-based approaches rely on approximations to the underlying generative process, which can lead to solutions that deviate from the true Bayesian posterior.

In this webinar, the presenter will present two diffusion-based sampling algorithms that leverage the expressive power of diffusion models without requiring such approximations. The proposed methods can be viewed as complementary sampling extensions of classical plug-and-play priors (PnP), with one based on gradient descent and the other on variable splitting. He will discuss the theoretical foundations of these algorithms, including distributional convergence guarantees, and demonstrate their performance on challenging inverse problems. In particular, results on black hole imaging highlight their ability to characterize multiple plausible solutions and perform uncertainty quantification in highly ill-posed settings.

Duration
00:54:07
Subtitles
✖

IEEE SPS Education Center FAQs

The IEEE SPS Education Center is your hub for educational resources in signal processing. It offers a variety of materials tailored for students and professionals alike. You can explore content based on your specific interests and skill levels.

Select the program and click on the external link to the IEEE SPS Resource Center.

Educational credits in the form of professional development hours (PDHs) or continuing education units (CEUs) are available on select educational programs.