multiviewae 1.1.2

A library for running multiview autoencoder models

Multi-view-AE: Multi-modal representation learning using autoencoders

multi-view-AE is a collection of multi-modal autoencoder models for learning joint representations from multiple modalities of data. The package is structured such that all models have fit, predict_latents and predict_reconstruction methods. All models are built in Pytorch and Pytorch-Lightning.

For more information on implemented models and how to use the package, please see the documentation at https://multi-view-ae.readthedocs.io/en/latest/

Models Implemented

Below is a table with the models contained within this repository and links to the original papers.

Model class	Model name	Number of views	Original work
mcVAE	Multi-Channel Variational Autoencoder (mcVAE)	>=1	link
AE	Multi-view Autoencoder	>=1
AAE	Multi-view Adversarial Autoencoder with separate latent representations	>=1
DVCCA	Deep Variational CCA	2	link
jointAAE	Multi-view Adversarial Autoencoder with joint latent representation	>=1
wAAE	Multi-view Adversarial Autoencoder with joint latent representation and wasserstein loss	>=1
mmVAE	Variational mixture-of-experts autoencoder (MMVAE)	>=1	link
mVAE	Multimodal Variational Autoencoder (MVAE)	>=1	link
me_mVAE	Multimodal Variational Autoencoder (MVAE) with separate ELBO terms for each view	>=1	link
JMVAE	Joint Multimodal Variational Autoencoder(JMVAE-kl)	2	link
MVTCAE	Multi-View Total Correlation Auto-Encoder (MVTCAE)	>=1	link
MoPoEVAE	Mixture-of-Products-of-Experts VAE	>=1	link
mmJSD	Multimodal Jensen-Shannon divergence model (mmJSD)	>=1	link
weighted_mVAE	Generalised Product-of-Experts Variational Autoencoder (gPoE-MVAE)	>=1	link
VAE_barlow	Multi-view Variational Autoencoder with barlow twins loss between latents.	2	link,link
AE_barlow	Multi-view Autoencoder with barlow twins loss between latents.	2	link,link
DMVAE	Disentangled multi-modal variational autoencoder	>=1	link
weighted_DMVAE	Disentangled multi-modal variational autoencoder with gPoE joint posterior	>=1

Installation

Clone this repository and move to folder:

git clone https://github.com/alawryaguila/multi-view-AE
cd multi-view-AE

Create the customised python environment:

conda create --name mvae python=3.9

Activate python environment:

conda activate mvae

Install the multi-view-AE package:

pip install ./

Citation

If you have used multi-view-AE in your research, please consider citing our JOSS paper:

Aguila et al., (2023). Multi-view-AE: A Python package for multi-view autoencoder models. Journal of Open Source Software, 8(85), 5093, https://doi.org/10.21105/joss.05093

Bibtex entry:

@article{Aguila2023, 
doi = {10.21105/joss.05093}, 
url = {https://doi.org/10.21105/joss.05093}, 
year = {2023}, 
publisher = {The Open Journal}, 
volume = {8}, 
number = {85}, 
pages = {5093}, 
author = {Ana Lawry Aguila and Alejandra Jayme and Nina Montaña-Brown and Vincent Heuveline and Andre Altmann}, 
title = {Multi-view-AE: A Python package for multi-view autoencoder models}, journal = {Journal of Open Source Software} 
}

Contribution guidelines

Contribution guidelines are available at https://multi-view-ae.readthedocs.io/en/latest/