Uncertainty Quantification via Neural Posterior Principal Components
Authors: Elias Nehme, Omer Yair, Tomer Michaeli
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We showcase our method on multiple inverse problems in imaging, including denoising, inpainting, super-resolution, colorization, and biological image-to-image translation. Our method reliably conveys instanceadaptive uncertainty directions, achieving uncertainty quantification comparable with posterior samplers while being orders of magnitude faster. |
| Researcher Affiliation | Academia | Elias Nehme Technion Israel Institute of Technology seliasne@campus.technion.ac.il Omer Yair Technion Israel Institute of Technology omeryair@campus.technion.ac.il Tomer Michaeli Technion Israel Institute of Technology tomer.m@ee.technion.ac.il |
| Pseudocode | No | The paper does not contain any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | Code and examples are available on our webpage. |
| Open Datasets | Yes | Handwritten digits Figure 4 demonstrates NPPC on denoising and inpainting of handwritten digits from the MNIST dataset. Faces To test NPPC on faces, we trained on the Celeb A-HQ dataset using the original split inherited from celeb A [24]...Here, we applied NPPC to a dataset of migrating cells imaged live for 14h (1 picture every 10min) using a spinning-disk microscope [52]. The dataset consisted of 1753 image pairs of resolution 1024 1024, out of which 1748 were used for training, and 5 were used for testing following the original split by the authors. |
| Dataset Splits | Yes | resulting in 24183 images for training, 2993 images for validation, and 2824 images for testing. |
| Hardware Specification | No | The paper does not provide specific details on the hardware (e.g., GPU/CPU models, memory) used to run the experiments. |
| Software Dependencies | No | The paper mentions using 'variants of the U-Net architecture [11, 40]' but does not provide specific version numbers for any software dependencies (e.g., programming languages, libraries, frameworks). |
| Experiment Setup | Yes | Full details regarding the architectures, the scheduler, and the per-task setting of λ1, λ2 are in App. A. |