CS Talk
Vincent Monardo
Host: Holly Rushmeier
Title: Plug-and-Play Image Reconstruction Meets Stochastic Variance-Reduced Gradient Methods
Abstract:
Plug-and-play (PnP) methods have recently emerged as a powerful framework for image reconstruction that can flexibly combine different physics-based observation models with data-driven image priors in the form of denoisers, and achieve state-of-the-art image reconstruction quality in many applications. In this work, we aim to further improve the computational efficacy of PnP methods by designing a new algorithm that makes use of stochastic variance-reduced gradients (SVRG), a nascent idea to accelerate runtime in stochastic optimization. Compared with existing PnP methods using batch gradients or stochastic gradients, the new algorithm, called PnP-SVRG, achieves comparable or better accuracy of image reconstruction at a much faster computational speed. Extensive numerical experiments are provided to demonstrate the benefits of the proposed algorithm through the application of phase retrieval in conjunction with a wide variety of popular image denoisers.
Bio:
Vincent Monardo is a PhD Candidate in the Electrical and Computer Engineering Department at Carnegie Mellon University advised by Dr. Yuejie Chi. Vincent received his BSE with Honors in Electrical Engineering from Arizona State University with a Minor in Applied Mathematics. Vincent’s research interests lie in the domain of designing and analyzing efficient algorithms for solving inverse problems–especially within the realm of image reconstruction–which improve upon the computational and statistical performance of state-of-the-art methods. In the future, Vincent is motivated to teach future generations the beauty of machine learning–in 2019, he received the CMU ECE Outstanding Teaching Award. In his spare time, Vincent enjoys trivia, crosswords, music, and rock climbing.