Hyperspherical Variational Auto-Encoders. Davidson, T., R., Falorsi, L., De Cao, N., Kipf, T., & Tomczak, J., M. arXiv:1804.00891 [cs, stat], 9, 2018.
Hyperspherical Variational Auto-Encoders [link]Website  abstract   bibtex   
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 \$\textbackslashmathcal\S\\$-VAE, is more suitable for capturing data with a hyperspherical latent structure, while outperforming a normal, \$\textbackslashmathcal\N\\$-VAE, in low dimensions on other data types.
@article{
 title = {Hyperspherical Variational Auto-Encoders},
 type = {article},
 year = {2018},
 keywords = {Computer Science - Machine Learning,Statistics - Machine Learning},
 websites = {http://arxiv.org/abs/1804.00891},
 month = {9},
 id = {80a3859e-084c-318b-8598-fbaedd606332},
 created = {2022-03-28T09:45:03.638Z},
 accessed = {2022-03-26},
 file_attached = {true},
 profile_id = {235249c2-3ed4-314a-b309-b1ea0330f5d9},
 group_id = {1ff583c0-be37-34fa-9c04-73c69437d354},
 last_modified = {2022-03-29T08:05:33.433Z},
 read = {false},
 starred = {false},
 authored = {false},
 confirmed = {true},
 hidden = {false},
 citation_key = {davidsonHypersphericalVariationalAutoEncoders2018},
 source_type = {article},
 notes = {arXiv: 1804.00891},
 private_publication = {false},
 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 \$\textbackslashmathcal\S\\$-VAE, is more suitable for capturing data with a hyperspherical latent structure, while outperforming a normal, \$\textbackslashmathcal\N\\$-VAE, in low dimensions on other data types.},
 bibtype = {article},
 author = {Davidson, Tim R and Falorsi, Luca and De Cao, Nicola and Kipf, Thomas and Tomczak, Jakub M},
 journal = {arXiv:1804.00891 [cs, stat]}
}

Downloads: 0