It Has Potential: Gradient-Driven Denoisers for Convergent Solutions to Inverse Problems
Dec 6, 2021
Speakers
Regev Cohen
Řečník · 0 sledujících
Yochai Blau
Řečník · 1 sledující
Daniel Freedman
Řečník · 0 sledujících
Ehud Rivlin
Řečník · 0 sledujících
About
In recent years there has been increasing interest in leveraging denoisers for solving general inverse problems. Two leading frameworks are regularization-by-denoising (RED) and plug-and-play priors (PnP) which incorporate explicit likelihood functions with priors induced by image denoising algorithms. RED and PnP have shown state-of-the-art performance in diverse imaging tasks when powerful denoisers are used, such as convolutional neural networks (CNNs). However, such denoisers typically do not exhibit symmetric Jacobians, hence, they cannot be interpreted as maximum a posteriori (MAP) nor minimum mean square error (MMSE) estimators. Furthermore, the Lipschitz constant of these CNN denoisers often has to be controlled (during training) to lead to stable inverse algorithms whose convergence can only be shown empirically. In this work, we introduce image denoisers derived as the gradients of smooth scalar-valued deep neural networks, acting as potentials. This ensures two things: (1) the proposed denoisers display symmetric Jacobians, allowing for a MAP estimator interpretation; (2) the denoisers may be integrated into optimization schemes with backtracking step size, removing the need for enforcing their Lipschitz constant. We develop a simple inversion method that utilizes the proposed denoisers, and we theoretically establish its convergence to stationary points of an underlying objective function consisting of the learned potentials. We numerically validate our method through various imaging experiments, showing improved results compared to RED and PnP, and with additional provable stability.In recent years there has been increasing interest in leveraging denoisers for solving general inverse problems. Two leading frameworks are regularization-by-denoising (RED) and plug-and-play priors (PnP) which incorporate explicit likelihood functions with priors induced by image denoising algorithms. RED and PnP have shown state-of-the-art performance in diverse imaging tasks when powerful denoisers are used, such as convolutional neural networks (CNNs). However, such denoisers typically do no…
Organizer
NeurIPS 2021
Účet · 1,9k sledujících
About NeurIPS 2021
Neural Information Processing Systems (NeurIPS) is a multi-track machine learning and computational neuroscience conference that includes invited talks, demonstrations, symposia and oral and poster presentations of refereed papers. Following the conference, there are workshops which provide a less formal setting.
Like the format? Trust SlidesLive to capture your next event!
Professional recording and live streaming, delivered globally.
Sharing
Recommended Videos
Presentations on similar topic, category or speaker
Successor Feature Landmarks for Long-Horizon Goal-Conditioned Reinforcement Learning
Zhlédnout později
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
Submission Strategies
Zhlédnout později
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
Real-Time Optimization for Model Predictive Control
Zhlédnout později
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
Online Multi-Armed Bandits with Adaptive Inference
Zhlédnout později
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
Conservative Offline Distributional Reinforcement Learning
Zhlédnout později
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %
Object-aware Contrastive Learning for Debiased Scene Representation
Zhlédnout později
Sangwoo Mo, …
Pro uložení prezentace do věčného trezoru hlasovalo 0 diváků, což je 0.0 %