TH1: AI for emerging inverse problems in computational imaging
Tuesday Feb. 20, 8:30am-12:30pm
AAAI 2024 - Vancouver, Canada
Overview
This tutorial focuses on emerging computational imaging applications, particularly addressing lesser-known inverse problems. Our objective is to shed light on crucial yet lesser-studied areas like snapshot compressive imaging and single-photon imaging, offering insights into their mathematical modeling, advancements, and the limitations of current solutions. In doing so, we also emphasize how AI/ML approaches are instrumental in addressing these challenges, enhancing traditional optimization-based methods to develop state-of-the-art algorithms in computational imaging.
Schedule
Overview of Inverse Problems in Imaging
Classic Approach to Solving Inverse Problems:
Utilizing simple convex regularizers to model the source structure
Formulating the inverse problem as a convex optimization problem
Challenges and Limitations of Traditional Methods:
Convex regularization methods and their limitations in capturing complex source structures
Predominant focus on linear or generalized linear models
Deep Learning in Inverse Problems:
Objectives:
Advancing beyond simplistic source models
Developing versatile algorithms for various inverse problems
Potential Deep Learning-Based Approaches and Their Pros and Cons:
End-to-end solutions
Plug-and-play iterative methods
Unrolled network architectures
Defining the Problem: Recovering a 3D data cube from a single 2D projection
Applications of SCI
Mathematical Modeling of the SCI Inverse Problem
Deep Learning-Based Methods for SCI
Introduction to Ultrafast Imaging Technologies
Exploring Single-Photon Imaging and Its Mathematical Model
Non-Line-of-Sight Imaging:
Problem Definition
Classical Approaches to Non-Line-of-Sight Imaging
Deep Learning-Based Approaches to Non-Line-of-Sight Imaging
Coherent Imaging: Classic Challenges in an Unsolved Context
Problem Definition: Imaging from Compressive Measurements in the Presence of Speckle Noise
Review of Mathematical Modeling and Recent Theoretical Developments
Deep Learning-Based Methods for Compressive Coherence Imaging
Audience
Target audience: Machine learning and applied math researchers and practitioners
Prerequisite information:
Foundational understanding of basic linear algebra and convex optimization.
Prior knowledge in deep learning and underdetermined linear inverse problems is advantageous, but not required.
Registration information can be found here:
https://aaai.org/aaai-conference/registration/
Organizers
Shirin Jalali
Rutgers University
David Lindell
University of Toronto
Xin Yuan
Westlake University