Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/o-laurent/bayes_posterior_symmetry_exploration
Code of the paper A Symmetry-Aware Exploration of Bayesian Neural Network Posteriors published at ICLR 2024.
https://github.com/o-laurent/bayes_posterior_symmetry_exploration
Last synced: 6 days ago
JSON representation
Code of the paper A Symmetry-Aware Exploration of Bayesian Neural Network Posteriors published at ICLR 2024.
- Host: GitHub
- URL: https://github.com/o-laurent/bayes_posterior_symmetry_exploration
- Owner: o-laurent
- License: mit
- Created: 2024-03-15T15:03:18.000Z (10 months ago)
- Default Branch: main
- Last Pushed: 2024-06-21T09:41:21.000Z (7 months ago)
- Last Synced: 2024-11-08T19:49:01.299Z (2 months ago)
- Language: Python
- Homepage: https://ensta-u2is-ai.github.io/Symmetry-Aware-BNN/
- Size: 3.66 MB
- Stars: 1
- Watchers: 2
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# A symmetry-aware exploration of Bayesian posteriors
This is the official repository for the code of the paper [A Symmetry-Aware Exploration of Bayesian Neural Network Posteriors](https://arxiv.org/abs/2310.08287) published at ICLR 2024.
## The Checkpoints dataset
The Checkpoint dataset can be found on [Hugging Face](https://huggingface.co/datasets/torch-uncertainty/Checkpoints). I have put there the most interesting networks and plan to add the rest shortly. Tell me if this is of interest to you in an issue.
The a large part of the models were trained using the [TorchUncertainty](https://github.com/ENSTA-U2IS-AI/torch-uncertainty).
## The code
The code currently includes the scripts to "remove" permutation and scaling symmetries, including the convex optimization using the `cvxpy` package. The whole code of the experiments is still a bit messy but I plan to add it here.
I'll rework a bit the files on the symmetries, but the `scale_resnet.py` should be a good starting point. Tell me if you need examples to understand the how to use the code.
The translation of [mmdagg](https://github.com/antoninschrab/mmdagg-paper) to PyTorch is in the MMD folder. We had trouble concerning the memory consumption of the JAX implementation.
## The poster
I also provide the (new) vertical poster of the paper.
## Something missing?
:construction: I'm currently working on cleaning up the rest of the code, so raise an issue if you would like it to be prioritized.