Hyperspherical Variational Auto-Encoders

Tim R. Davidson, Luca Falorsi, Nicola De Cao, Thomas Kipf, Jakub M. Tomczak
4/3/2018
stat.MLcs.LG

Official code: nicola-decao/s-vae-pytorch (392 stars, MIT, PyTorch, last commit 2020-03-21). 1 more implementation below.

Abstract

The Variational Auto-Encoder (VAE) is one of the most used unsupervised machine learning models. But although the default choice of a Gaussian distribution for both the prior and posterior represents a mathematically convenient distribution often leading to competitive results, we show that this parameterization fails to model data with a latent hyperspherical structure. To address this issue we propose using a von Mises-Fisher (vMF) distribution instead, leading to a hyperspherical latent space. Through a series of experiments we show how such a hyperspherical VAE, or $\mathcal{S}$-VAE, is more suitable for capturing data with a hyperspherical latent structure, while outperforming a normal, $\mathcal{N}$-VAE, in low dimensions on other data types. Code at http://github.com/nicola-decao/s-vae-tf and https://github.com/nicola-decao/s-vae-pytorch

View on arXivView PDF

Code Implementations(2)

nicola-decao/s-vae-pytorchOfficial100%
PyTorchMIT

Pytorch implementation of Hyperspherical Variational Auto-Encoders

39260PythonJun 29, 20186 years agoMIT
deep-learninghyperspherical-vaemachine-learningmanifold-learningpytorch+3 more
nicola-decao/s-vae-tfOfficial100%
TensorFlowMIT

Tensorflow implementation of Hyperspherical Variational Auto-Encoders

24139PythonMar 25, 20187 years agoMIT
deep-learninghyperspherical-vaemachine-learningmanifold-learningtensorflow+3 more

Similar Papers

The Hyperspherical Geometry of CLIP Latent Space: A Semantic Mixture Model

84% match2026

HYPO: Hyperspherical Out-of-Distribution Generalization

79% match1 repo192024

Hyperspherical Variational Autoencoders Using Efficient Spherical Cauchy Distribution

78% match2025

SPHERE-JEPA: Spherical Prediction with Homogeneous Embeddings

78% match2026

Prior Specification for Bayesian Matrix Factorization via Prior Predictive Matching

77% match2 repos32019

Learning over von Mises-Fisher Distributions via a Wasserstein-like Geometry

77% match2025

Cite this paper

@article{davidson2018hyperspherical,
  title  = {Hyperspherical Variational Auto-Encoders},
  author = {Tim R. Davidson and Luca Falorsi and Nicola De Cao and Thomas Kipf and Jakub M. Tomczak},
  year   = {2018},
  eprint = {1804.00891},
  archivePrefix = {arXiv},
  url    = {https://arxiv.org/abs/1804.00891}
}

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/1804.00891.svg)](https://codeofpaper.com/paper/1804.00891)

Discussion