Riemannian Convex Potential Maps

Last update: Nov 28, 2022

Related tags

Overview

Riemannian Convex Potential Maps

This repository is by Brandon Amos, Samuel Cohen and Yaron Lipman and contains the JAX source code to reproduce the experiments in our ICML 2021 paper on Riemannian Convex Potential Maps.

Modeling distributions on Riemannian manifolds is a crucial component in understanding non-Euclidean data that arises, e.g., in physics and geology. The budding approaches in this space are limited by representational and computational tradeoffs. We propose and study a class of flows that uses convex potentials from Riemannian optimal transport. These are universal and can model distributions on any compact Riemannian manifold without requiring domain knowledge of the manifold to be integrated into the architecture. We demonstrate that these flows can model standard distributions on spheres, and tori, on synthetic and geological data.

Reproducing our experiments

config.yaml contains the basic config for setting up our experiments. We currently use hydra 1.0.3. By default it contains the options to reproduce the multimodal sphere flow:

This can be run with:

$ ./main.py
workspace: /private/home/bda/repos/rcpm/exp_local/2021.06.21/053411
Iter 1000 | Loss -10.906 | KL 0.017 | ESS 96.74% | 9.54e-02s/it
Iter 2000 | Loss -10.908 | KL 0.013 | ESS 97.43% | 1.90e-02s/it
Iter 3000 | Loss -10.911 | KL 0.012 | ESS 97.71% | 1.75e-02s/it
Iter 4000 | Loss -10.912 | KL 0.010 | ESS 98.02% | 1.63e-02s/it
Iter 5000 | Loss -10.912 | KL 0.009 | ESS 98.19% | 1.46e-02s/it
...
Iter 30000 | Loss -10.915 | KL 0.006 | ESS 98.75% | 1.78e-02s/it

This will create a work directory in exp_local with the models and debugging information. You can use plot-components.py to further analyze the components of the learned flow, and plot-demo.py to produce the grid visualization from Figure 2 of our paper.

Other JAX sphere flow library

katalinic/sdflows provides a great JAX re-implementation of Normalizing Flows on Tori and Spheres.

Citations

If you find this repository helpful for your publications, please consider citing our paper:

@inproceedings{cohen2021riemannian,
  title={{Riemannian Convex Potential Maps}},
  author={Cohen*, Samuel and Amos*, Brandon and Lipman, Yaron},
  booktitle={ICML},
  year={2021},
}

Licensing

This repository is licensed under the CC BY-NC 4.0 License.

You might also like...

Image morphing without reference points by applying warp maps and optimizing over them.

Differentiable Morphing Image morphing without reference points by applying warp maps and optimizing over them. Differentiable Morphing is machine lea

380 Dec 19, 2022

PyTorch implementations of the paper: "Learning Independent Instance Maps for Crowd Localization"

IIM - Crowd Localization This repo is the official implementation of paper: Learning Independent Instance Maps for Crowd Localization. The code is dev

91 Nov 10, 2022

Spatial Intention Maps for Multi-Agent Mobile Manipulation (ICRA 2021)

spatial-intention-maps This code release accompanies the following paper: Spatial Intention Maps for Multi-Agent Mobile Manipulation Jimmy Wu, Xingyua

70 Jan 2, 2023

Spatial Action Maps for Mobile Manipulation (RSS 2020)

spatial-action-maps Update: Please see our new spatial-intention-maps repository, which extends this work to multi-agent settings. It contains many ne

27 Nov 30, 2022

Reimplementation of the paper `Human Attention Maps for Text Classification: Do Humans and Neural Networks Focus on the Same Words? (ACL2020)`

Human Attention for Text Classification Re-implementation of the paper Human Attention Maps for Text Classification: Do Humans and Neural Networks Foc

15 Dec 13, 2021

Range Image-based LiDAR Localization for Autonomous Vehicles Using Mesh Maps

Comments

Lie group application

Thanks for your great work on applying optimal transport to Manifold Normalizing Flow. I wonder, can we apply this general framework to any manifold of the Lie group? especially for SO(2) or SO(3) manifolds. Both have well-defined rotation geodesics and closed-form exponential maps.

opened by huyphan168 0

Riemannian Convex Potential Maps

Related tags

Overview

Riemannian Convex Potential Maps

Reproducing our experiments

Other JAX sphere flow library

Citations

Licensing

You might also like...

Image morphing without reference points by applying warp maps and optimizing over them.

PyTorch implementations of the paper: "Learning Independent Instance Maps for Crowd Localization"

Spatial Intention Maps for Multi-Agent Mobile Manipulation (ICRA 2021)

Spatial Action Maps for Mobile Manipulation (RSS 2020)

Reimplementation of the paper `Human Attention Maps for Text Classification: Do Humans and Neural Networks Focus on the Same Words? (ACL2020)`

Range Image-based LiDAR Localization for Autonomous Vehicles Using Mesh Maps

Implementation of "Fast and Flexible Temporal Point Processes with Triangular Maps" (Oral @ NeurIPS 2020)

Deep Compression for Dense Point Cloud Maps.

Neural Surface Maps

Comments

Lie group application

Owner

Facebook Research

Code in PyTorch for the convex combination linear IAF and the Householder Flow, J.M. Tomczak & M. Welling

Universal Probability Distributions with Optimal Transport and Convex Optimization

Convex optimization for fun and profit.

ESGD-M - A stochastic non-convex second order optimizer, suitable for training deep learning models, for PyTorch

Differentiable molecular simulation of proteins with a coarse-grained potential

CPF: Learning a Contact Potential Field to Model the Hand-object Interaction

Using Hotel Data to predict High Value And Potential VIP Guests

Implementation of Graph Transformer in Pytorch, for potential use in replicating Alphafold2

[ICCV 2021] Excavating the Potential Capacity of Self-Supervised Monocular Depth Estimation

Neural Scene Flow Fields using pytorch-lightning, with potential improvements