Plug-and-Play Posterior Sampling under Mismatched Measurement and Prior Models
Authors: Marien Renaud, Jiaming Liu, Valentin De Bortoli, Andres Almansa, Ulugbek Kamilov
ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We numerically validate our theory on several inverse problems such as sampling from Gaussian mixture models and image deblurring. Our results suggest that the sensitivity of the sampling distribution of Pn P-ULA to a mismatch in the measurement model and the denoiser can be precisely characterized. |
| Researcher Affiliation | Academia | Marien Renaud Washington University in St. Louis St. Louis, MO 63130, USA Jiaming Liu Washington University in St. Louis St. Louis, MO 63130, USA Valentin de Bortoli ENS, CNRS, PSL University Paris, 75005, FRANCE Andr es Almansa CNRS, Universit e Paris Cit e Paris, 75006, FRANCE Ulugbek Kamilov Washington University in St. Louis St. Louis, MO 63130, USA |
| Pseudocode | No | The paper does not contain any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | Source code used in our experiments can be found in Pn P ULA posterior law sensivity code. It contains a README.md file that explains step by step how to run the algorithm and replicate the results of the paper. |
| Open Datasets | Yes | Images (from CBSD68 dataset (Martin et al., 2001)) of size 256 256 in grayscale. A total of 17 distinct Dn CNN denoising models (Ryu et al., 2019) were trained... The denoiser is implemented using the DRUNet architecture (Zhang et al., 2021) and has been trained on the Celeb A dataset (Liu et al., 2015). |
| Dataset Splits | Yes | In our experimental setup, we address the deblurring inverse problem using the Celeb A validation set, which consists exclusively of images of women s faces resized to RGB dimensions of 256 256 pixels. |
| Hardware Specification | Yes | The computational time required for training one denoiser an NVIDIA Ge Force RTX 2080 GPU was approximately 2 hours. So a total of 34 hours of computation was needed to train the gray-scale image denoisers. The computational time required for one Pn P-ULA sampling on an NVIDIA Ge Force RTX 2080 GPU was approximately 40 seconds per image. |
| Software Dependencies | No | The paper mentions software components like 'Dn CNN denoising models' and 'DRUNet architecture' but does not specify their version numbers or other crucial software dependencies with versions. |
| Experiment Setup | Yes | With Pn P-ULA parameters : ϵ = 0.05, δ = 0.05 and N = 100000. The initialization of the Markov chain is taken at x = 0. Parameters are chosen following the recommendation of (Laumont et al., 2022). The initialization of Pn P-ULA was performed using the observation vector y, and a total of N = 1000 images were saved as samples, with one image being saved every 10 steps in the sampling process. |