Variational auto-encoders with Student’s t-prior

Najmeh Abiri; Mattias Ohlsson

Variational auto-encoders with Student’s t-prior

Najmeh Abiri, Mattias Ohlsson

eSSENCE: The e-Science Collaboration

Halmstad University

Forskningsoutput: Kapitel i bok/rapport/Conference proceeding › Konferenspaper i proceeding › Peer review

Sammanfattning

We propose a new structure for the variational auto-encoders (VAEs) prior, with the weakly informative multivariate Student’s t-distribution. In the proposed model all distribution parameters are trained, thereby allowing for a more robust approximation of the underlying data distribution. We used Fashion-MNIST data in two experiments to compare the proposed VAEs with the standard Gaussian priors. Both experiments showed a better reconstruction of the images with VAEs using Student’s t-prior distribution.

Originalspråk	engelska
Titel på värdpublikation	ESANN 2019 - Proceedings
Undertitel på värdpublikation	The 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning
ISBN (elektroniskt)	978-287-587-065-0
Status	Published - 2019
Evenemang	27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning - Bruges, Belgien Varaktighet: 2019 apr. 24 → 2019 apr. 26

Konferens

Konferens	27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning
Förkortad titel	ESANN 2019
Land/Territorium	Belgien
Ort	Bruges
Period	2019/04/24 → 2019/04/26

Ämnesklassifikation (UKÄ)

Datavetenskap (datalogi)
Annan data- och informationsvetenskap

Tillgång till dokument

https://www.elen.ucl.ac.be/Proceedings/esann/esannpdf/es2019-42.pdf

AI Lund: AI-forskning vid Lunds universitet
Åström, K. (Projektkoordinator), Malec, J. (Projektkoordinator), Larsson, S. (Projektkoordinator), Ohlsson, M. (Projektkoordinator), Balkenius, C. (Projektkoordinator), Dutceac Segesten, A. (Projektkoordinator), Haider, J. (Projektkoordinator), Willim, R. (Projektkoordinator), Ledendal, J. (Projektkoordinator), Aits, S. (Projektkoordinator), Hedlund, M. (Forskare), Wisbrant, J. (Projektkommunikatör), Heiberg, E. (PI), Topp, E. A. (Forskare), Janneck, J. (Forskare), Klang, M. (Forskare), Brinck, I. (Forskare), Sundin, O. (Forskare), Westin, E. (Administratör), Blom Lussi, E. (Projektkoordinator) & Axhamn, J. (Projektkoordinator)
2018/01/01 → …
Projekt: Nätverk

Citera det här

@inproceedings{24babff9628842aba0994272559768c3,

title = "Variational auto-encoders with Student{\textquoteright}s t-prior",

abstract = "We propose a new structure for the variational auto-encoders (VAEs) prior, with the weakly informative multivariate Student{\textquoteright}s t-distribution. In the proposed model all distribution parameters are trained, thereby allowing for a more robust approximation of the underlying data distribution. We used Fashion-MNIST data in two experiments to compare the proposed VAEs with the standard Gaussian priors. Both experiments showed a better reconstruction of the images with VAEs using Student{\textquoteright}s t-prior distribution.",

author = "Najmeh Abiri and Mattias Ohlsson",

year = "2019",

language = "English",

booktitle = "ESANN 2019 - Proceedings",

note = "27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, ESANN 2019 ; Conference date: 24-04-2019 Through 26-04-2019",

}

Abiri, N & Ohlsson, M 2019, Variational auto-encoders with Student’s t-prior. i ESANN 2019 - Proceedings: The 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning, Bruges, Belgien, 2019/04/24. <https://www.elen.ucl.ac.be/Proceedings/esann/esannpdf/es2019-42.pdf>

TY - GEN

T1 - Variational auto-encoders with Student’s t-prior

AU - Abiri, Najmeh

AU - Ohlsson, Mattias

PY - 2019

Y1 - 2019

N2 - We propose a new structure for the variational auto-encoders (VAEs) prior, with the weakly informative multivariate Student’s t-distribution. In the proposed model all distribution parameters are trained, thereby allowing for a more robust approximation of the underlying data distribution. We used Fashion-MNIST data in two experiments to compare the proposed VAEs with the standard Gaussian priors. Both experiments showed a better reconstruction of the images with VAEs using Student’s t-prior distribution.

AB - We propose a new structure for the variational auto-encoders (VAEs) prior, with the weakly informative multivariate Student’s t-distribution. In the proposed model all distribution parameters are trained, thereby allowing for a more robust approximation of the underlying data distribution. We used Fashion-MNIST data in two experiments to compare the proposed VAEs with the standard Gaussian priors. Both experiments showed a better reconstruction of the images with VAEs using Student’s t-prior distribution.

M3 - Paper in conference proceeding

BT - ESANN 2019 - Proceedings

T2 - 27th European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning

Y2 - 24 April 2019 through 26 April 2019

ER -

Variational auto-encoders with Student’s t-prior

Sammanfattning

Konferens

Ämnesklassifikation (UKÄ)

Tillgång till dokument

Fingeravtryck

Projekt

AI Lund: AI-forskning vid Lunds universitet

Citera det här