There is a long history of algorithmic development for solving inverse problems arising in sensing and imaging systems and beyond. Examples include medical and computational imaging, compressive sensing, as well as community detection in networks. Until recently, most algorithms for solving inverse problems in the imaging and network sciences were based on static signal models derived from physics or intuition, such as wavelets or sparse representations.
Today, the best performing approaches for the aforementioned image reconstruction and sensing problems are based on deep learning, which learn various elements of the method including i) signal representations, ii) stepsizes and parameters of iterative algorithms, iii) regularizers, and iv) entire inverse functions. For example, it has recently been shown that solving a variety of inverse problems by transforming an iterative, physics-based algorithm into a deep network whose parameters can be learned from training data, offers faster convergence and/or a better quality solution. Moreover, even with very little or no learning, deep neural networks enable superior performance for classical linear inverse problems such as denoising and compressive sensing. Motivated by those success stories, researchers are redesigning traditional imaging and sensing systems.
However, the field is mostly wide open with a range of theoretical and practical questions unanswered. In particular, deep-neural network based approaches often lack the guarantees of the traditional physics based methods, and while typically superior can make drastic reconstruction errors, such as fantasizing a tumor in an MRI reconstruction.
This workshop aims at bringing together theoreticians and practitioners in order to chart out recent advances and discuss new directions in deep neural network based approaches for solving inverse problems in the imaging and network sciences.
|8:30 - 8:40||Opening Remarks|
|8:40 - 9:10||Lenka Zdeborova: The spiked matrix model with generative priors|
|9:10 - 9:40||Shuang Qiu, Xiaohan Wei, Zhuoran Yang: Robust One-Bit Recovery via ReLU Generative Networks: Improved Statistical Rate and Global Landscape Analysis|
|9:40 - 10:30||Coffee Break|
|10:30 - 11:00||Laura Waller: Computational microscopy in scattering media|
|11:00 - 11:30||Guillermo Sapiro: Basis Decomposition of Deep Learning|
|11:30 - 12:00||Stephan Hoyer, Jascha Sohl-Dickstein, Sam Greydanus: Neural Reparameterization Improves Structural Optimization|
|12:00 - 2:00||Lunch Break|
|2:00 - 2:30||Raquel Urtasun|
|2:30 - 3:00||Josh Batson: Blind Denoising, Self-Supervision, and Implicit Inverse Problems|
|3:00 - 3:30||Venkat Chandrasekaran: Learning Regularizers from Data|
|3:30 - 4:15||Break and Posters|
|4:15 - 6:00||Poster Session|
Neural reparameterization improves structural optimization
Stephan Hoyer, Jascha Sohl-Dickstein, Sam Greydanus
Robust One-Bit Recovery via ReLU Generative Networks: Improved Statistical Rate and Global Landscape Analysis
Shuang Qiu, Xiaohan Wei, Zhuoran Yang
Extreme Few-view CT Reconstruction using Deep Inference
Hyojin Kim, Rushil Anirudh, K. Aditya Mohan, Kyle Champley
Improving Limited Angle CT Reconstruction with a Robust GAN Prior
Rushil Anirudh, Hyojin Kim, Jayaraman J. Thiagarajan, K. Aditya Mohan, Kyle Champley
A Hybrid Architecture for On-Device Compressive Machine Learning
Yang Li, Thomas Strohmer
Generative Models for Low-Dimensional Video Representation and Compressive Sensing
Rakib Hyder, M. Salman Asif
PatchDIP Exploiting Patch Redundancy in Deep Image Prior for Denoising
Muhammad Asim, Fahad Shamshad, Ali Ahmed
Auto-encoders for compressed sensing
Pei Peng, Shirin Jalali, Xin Yuan
Compressed Sensing and Overparametrized Networks: Overfitting Peaks in a Model of Misparametrized Sparse Regression in the Interpolation Limit
Partha P Mitra
Lower Bounds for Compressed Sensing with Generative Models
Akshay Kamath, Sushrut Karmalkar, Eric Price
Y-net: A Physics-constrained and Semi-supervised Learning Approach to the Phase Problem in Computational Electron Imaging
Nouamane Laanait, Junqi Yin, Albina Borisevich
Unsupervised Deep Basis Pursuit: Learning inverse problems without ground-truth data
Jonathan I. Tamir, Stella X. Yu, Michael Lustig
AlgoNet: $C^\infty$ Smooth Algorithmic Neural Networks for Solving Inverse Problems
Felix Petersen, Christian Borgelt, Oliver Deussen
Retrieving Signals with Deep Complex Extractors
Chiheb Trabelsi, Olexa Bilaniuk, Ousmane Dia, Ying Zhang, Mirco Ravanelli, Jonathan Binas, Negar Rostamzadeh, Christopher J Pal
Unrolled, model-based networks for lensless imaging
Kristina Monakhova, Joshua Yurtsever, Grace Kuo, Nick Antipa, Kyrollos Yanny, Laura Waller
GAN priors for Bayesian inference
Dhruv V. Patel, Assad A. Oberai
Learning Network Parameters in the ReLU Model
Arya Mazumdar, Ankit Singh Rawat
Learned imaging with constraints and uncertainty quantification
Felix J. Herrmann, Ali Siahkoohi, Gabrio Rizzuti
Generative Inpainting Network Applications on Seismic Image Compression and Non-Uniform Sampling
Xiaoyang Rebecca Li, Nikolaos Mitsakos, Ping Lu, Yuan Xiao, Cheng Zhan, Xing Zhao
Exploring Properties of the Deep Image Prior
Andreas Kattamis, Adrian Weller
Learning-Based Low-Rank Approximations
Piotr Indyk, Ali Vakilian, Yang Yuan
Sample Complexity Lower Bounds for Compressive Sensing with Generative Models
Zhaoqiang Liu, Jonathan Scarlett
Energy Dissipation with Plug-and-Play Priors
Hendrik Sommerhoff, Andreas Kolb, Michael Moeller
Precise asymptotics for phase retrieval and compressed sensing with random generative priors
Benjamin Aubin, Bruno Loureiro, Antoine Baker, Florent Krzakala, Lenka Zdeborova
Learning to Recover Sparse Signals
Sichen Zhong, Yue Zhao, Jianshu Chen
Subsampled Fourier Ptychography via Pretrained Invertible and Untrained Network Priors
Fahad Shamshad, Asif Hanif, Ali Ahmed
Learning to Solve Linear Inverse Problems in Imaging with Neumann Networks
Davis Gilton, Greg Ongie, Rebecca Willett
Robust and interpretable blind image denoising via bias-free convolutional neural networks
Zahra Kadkhodaie, Sreyas Mohan, Eero P. Simoncelli, Carlos Fernandez-Granda
Phase Retrieval using Untrained Neural Network Priors
Gauri Jagatap, Chinmay Hegde
A GAN based solver of black-box inverse problems
Michael Gillhofer, Hubert Ramsauer, Johannes Brandstetter, Sepp Hochreiter
Co-Generation with GANs using AIS based HMC
Tiantian Fang, Alexander G. Schwing
Memory-efficient Learning for Large-scale Computational Imaging
Michael Kellman, Jon Tamir, Emrah Bostan, Michael Lustig, Laura Waller
Gradient-Based Neural DAG Learning
Sébastien Lachapelle, Philippe Brouillard, Tristan Deleu, Simon Lacoste-Julien
Low Shot Learning with Untrained Neural Networks for Imaging Inverse Problems
Oscar Leong, Wesam Sakla
Call for Papers and Submission Instructions
Submission is closed!
We invite researchers to submit anonymous extended abstracts of up to 4 pages (excluding references) which will be considered for contributed talks and posters. No specific formatting is required. Authors may use the NeurIPS style file, or any other style as long as it has standard font size (11pt) and margins (1in).
Submission at openreview open now until the submission deadline on
September 9 September 13.
We invite works on inverse problems in the imaging sciences and new developments in non-Euclidean domains such as graphs, including contributions on the development of new architectures for natural signal priors (for examples GANs, non adversarially trained generators, unlearned neural networks, and combinations thereof), theoretical foundations (including rigorous recovery guarantees, provable convergence, and bounds on representation errors), and applications in imaging and beyond. We especially encourage submissions in the following areas:
Learned generative models for solving inverse problems: Recent years have seen great advances in generative modeling for a variety of signals, in particular images. The corresponding priors, when enforced in reconstruction algorithms, enable lower sample complexity and higher robustness to noise than conventional approaches. In this workshop, we discuss progress and research directions in generative models for finding solutions to inverse problems efficiently and accurately.
Un-trained models for solving inverse problems: Even without any learning, deep neural networks have been shown to be effective models through the so-called deep image priors, suggesting that deep neural networks are inherently good at representing natural images or more generally, signals. In this workshop, we will discuss progress and research directions in the understanding of the inductive bias brought by deep architectures and by gradient-descent optimization.
Learning to solve inverse problems end-to-end: Neural networks applied to inverse problems yield impressive results. Trained on large amounts of training data they are often able to run faster and yield more accurate results than existing methods. However, those advances come at the cost of a lack of recovery guarantees, require large amounts of training data, and can sometimes lead to undesirable behaviour, such as hypothesizing parts of an image from training data. We encourage contributions on methods and algorithms for learning to solve inverse problems and contributions that explore application scenarios, for example in computational imaging.
Inverse problems beyond Euclidean data: Deep networks are emerging as a powerful tool to solve inverse problems on data with an underlying graph or manifold structure such as social networks or surfaces in computer graphics. We highly welcome contributions on solving inverse problems in geometric deep learning.
- Submission Deadline:
September 9thExtended until September 13th, 2019.
- Notification: October 1st, 2019
- Workshop: Friday, December 13, 2019.
- Reinhard Heckel (TUM)
- Paul Hand (Northeastern)
- Richard Baraniuk (Rice University)
- Joan Bruna (NYU)
- Alex Dimakis (UT Austin)
- Deanna Needell (UCLA)
Please email firstname.lastname@example.org with any questions.