https://github.com/VincentStimper/resampled-base-flows

Normalizing Flows with a resampled base distribution
https://github.com/VincentStimper/resampled-base-flows

normalizing-flows pytorch resampled-base-distribution

Last synced: 4 months ago
JSON representation

Normalizing Flows with a resampled base distribution

• Host: GitHub
• URL: https://github.com/VincentStimper/resampled-base-flows
• Owner: VincentStimper
• Created: 2020-08-20T14:00:52.000Z (almost 4 years ago)
• Default Branch: master
• Last Pushed: 2022-09-07T15:49:44.000Z (almost 2 years ago)
• Last Synced: 2024-01-16T10:45:21.570Z (5 months ago)
• Topics: normalizing-flows, pytorch, resampled-base-distribution
• Language: Python
• Homepage: https://proceedings.mlr.press/v151/stimper22a
• Size: 1.35 MB
• Stars: 34
• Watchers: 3
• Forks: 2
• Open Issues: 0
• Metadata Files:
  • Readme: README.md

Lists

• awesome-stars - VincentStimper/resampled-base-flows - Normalizing Flows with a resampled base distribution (Python)

README

        
# Resampling Base Distributions of Normalizing Flows

## Overview

Normalizing flows are a popular class of models for approximating probability 

distributions. However, their invertible nature limits their ability to model 

target distributions with a complex topological structure, such as Boltzmann 

distributions. Several procedures have been proposed to solve this problem 

but many of them sacrifice invertibility and, thereby, tractability of the 

log-likelihood as well as other desirable properties. To address these limitations, 

we introduce a base distribution for normalizing flows based on learned rejection 

sampling in our article

[Resampling Base Distributions of Normalizing Flows](https://proceedings.mlr.press/v151/stimper22a),

allowing the resulting normalizing flow to model complex topologies without giving

up bijectivity. In this repository, we implemented this class of base distributions

and provide the script for various experiments comparing them to other commonly used

base distributions. Some results of applying our method to 2D distributions are shown

below.

![Normalizing flows with different base distributions](https://github.com/VincentStimper/resampled-base-flows/blob/master/images/2d_distributions.png "Normalizing flows with different base distributions")

This packages builds upon the normalizing flow library 

[`normflows`](https://github.com/VincentStimper/normalizing-flows). The Boltzmann 

generator experiments are realized via the 

[`boltzgen`](https://github.com/VincentStimper/boltzmann-generators) library.

## Methods of Installation

The latest version of the package can be installed via pip

```

pip install --upgrade git+https://github.com/VincentStimper/resampled-base-flows.git

```

If you want to use a GPU, make sure that PyTorch is set up correctly by

by following the instructions at the

[PyTorch website](https://pytorch.org/get-started/locally/).

To run the Boltzmann generator experiments it is necessary to install OpenMM.

Instructions on how to do this can be found 

[here](http://docs.openmm.org/7.0.0/userguide/application.html#installing-openmm).

## Citation

If you use our code in your own research, please cite our paper:

> Vincent Stimper, Bernhard Schölkopf, José Miguel Hernández-Lobato. Resampling Base 

> Distributions of Normalizing Flows. In Proceedings of The 25th International Conference 

> on Artificial Intelligence and Statistics, volume 151, pp. 4915–4936, 2022.

**Bibtex**

```

@inproceedings{Stimper2022,

  title = 	 {Resampling {B}ase {D}istributions of {N}ormalizing {F}lows},

  author =       {Vincent Stimper and Bernhard Sch{\"o}lkopf and Jos{\'e} Miguel Hern{\'a}ndez-Lobato},

  booktitle = 	 {Proceedings of The 25th International Conference on Artificial Intelligence and Statistics},

  pages = 	 {4915--4936},

  year = 	 {2022},

  volume = 	 {151}

}

```