Learning Disentangled Joint Continuous and Discrete Representations

Emilien Dupont
3/31/2018
stat.MLcs.LG

Official code: Schlumberger/joint-vae (469 stars, MIT, PyTorch, last commit 2019-04-02).

Abstract

We present a framework for learning disentangled and interpretable jointly continuous and discrete representations in an unsupervised manner. By augmenting the continuous latent distribution of variational autoencoders with a relaxed discrete distribution and controlling the amount of information encoded in each latent unit, we show how continuous and categorical factors of variation can be discovered automatically from data. Experiments show that the framework disentangles continuous and discrete generative factors on various datasets and outperforms current disentangling methods when a discrete generative factor is prominent.

View on arXivView PDF

Code Implementations(1)

Schlumberger/joint-vaeOfficial100%
PyTorchMIT

Pytorch implementation of JointVAE, a framework for disentangling continuous and discrete factors of variation :star2:

46960Python, Jupyter NotebookMay 8, 20187 years agoMIT
disentangled-representationsdisentanglementgenerative-modelsgumbel-softmaxpytorch+1 more

Similar Papers

Disentanglement with Factor Quantized Variational Autoencoders

88% match1 repo52024

Learning Disentangled Discrete Representations

86% match1 repo2023

Variational Learning of Disentangled Representations

85% match2025

Disentanglement with Holographic Reduced Representations

83% match2026

Inferencing Based on Unsupervised Learning of Disentangled Representations

83% match1 repo72018

α-TCVAE: On the relationship between Disentanglement and Diversity

83% match1 repo22024

Cite this paper

@article{dupont2018learning,
  title  = {Learning Disentangled Joint Continuous and Discrete Representations},
  author = {Emilien Dupont},
  year   = {2018},
  eprint = {1804.00104},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1804.00104}
}

Add a badge to your README

Shows whether official code is linked for this paper.

Code of Paper badge
[![Code of Paper](https://codeofpaper.com/badge/1804.00104.svg)](https://codeofpaper.com/paper/1804.00104)

Discussion