Model Based Deep Learning - Applications to Imaging and Communications - Yonina Eldar

Explore model-based deep learning applications in imaging and communications through this comprehensive conference talk. Delve into the integration of parametric models with optimization tools, leading to efficient and interpretable networks. Discover approaches to overcome challenges in deep neural networks, such as their black-box nature and large training set requirements. Examine case studies in image deblurring, separation, and super-resolution for ultrasound and microscopy. Learn about efficient communication systems and the application of model-based methods for COVID-19 diagnosis using X-ray and ultrasound. Gain insights into topics including deep symbol detection, data-driven hybrid algorithms, deep unrolling for blind deblurring, and sparsity-based ultrasound super-resolution hemodynamic imaging. Understand the balance between classical statistical modeling techniques and deep learning approaches in signal processing and communications.

Intro
The Deep Learning Revolution
Challenges
Model Based Signal Processing
Motivation: Symbol Detection
Deep Symbol Detection
Model-Based vs. Deep Learning
Data Driven Hybrid Algorithms
DUBLID: Deep Unrolling for Blind Deblurring
Deep Unfolding with Normalizing Flow Priors
DL for Clutter Suppression
Deep Adaptive Beamforming
Super Resolution Microscopy
SPARCOM Recovery
SPARCOM: Super Resolution Correlation Microscopy
SUSHI: Sparsity-Based Ultrasound Super- resolution Hemodynamic Imaging
Rationale
Viterbi Algorithm
Factor Graph Methods