Speaker: Dr. Jyothi Rikhab Chand, Postdoctoral Research Associate ,University of Virginia, Charlottesville Virginia, USA.
Date/Time: Monday, 24th February 2025, 3:00 PM – 4:00 PM
Venue: SV Narsaiah Auditorium, IAP
Abstract: Deep learning-based reconstruction algorithms have shown significant improvement in recovering MRI images from undersampled measurements. However, they are memory intensive or require contraction constraints for fixed-point convergence. The former issue restricts the usage of the recovery algorithm in high-dimensional inverse problems, while the latter limits the expressivity of the network leading to decreased recovery performance. In this talk, I will present novel memory-efficient energy models parameterized by Convolutional Neural Networks to represent the negative log-prior of the MRI images. Once trained, these models can be used to solve the inverse problem using any off-the-shelf optimization algorithms with convergence guarantees without any constraints on the network that translates to improved performance. These models can also perform sampling, like diffusion models, to estimate the uncertainty of the recovered image.
Biography: Dr. Jyothi Rikhab Chand is a postdoctoral research associate working with Prof. Mathews Jacob at the University of Virginia, USA. Her current research interest lies in developing algorithms for solving problems in medical imaging by combining traditional model-based optimization algorithms with modern deep learning approaches. Previously, she finished her PhD at IIT-Delhi, where she developed optimization algorithms for various problems in signal processing and machine learning. She can be contacted at jyothi-rikhabchand@virginia.edu
Please mark your calendar for the next seminar series on 24th February 2025 at 3:00PM