Bayesian models of data streams with Hierarchical Power Priors

Andrés Masegosa, Thomas Dyhre Nielsen, Helge Langseth, Darío Ramos-López, Antonio Salmerón, Anders Læsø Madsen

Publikation: Bidrag til bog/antologi/rapport/konference proceedingKonferenceartikel i proceedingForskningpeer review

Resumé

Making inferences from data streams is a pervasive problem in many modern data analysis applications. But it requires to address the problem of continuous model updating, and adapt to changes or drifts in the underlying data generating distribution. In this paper, we approach these problems from a Bayesian perspective covering general conjugate exponential models. Our proposal makes use of non-conjugate hierarchical priors to explicitly model temporal changes of the model parameters. We also derive a novel variational inference scheme which overcomes the use of non-conjugate priors while maintaining the computational efficiency of variational methods over conjugate models. The approach is validated on three real data sets over three latent variable models.
OriginalsprogEngelsk
TitelProceedings of the 34th International Conference on Machine Learning
Vol/bind70
Publikationsdato2017
Sider2334-2343
StatusUdgivet - 2017
Begivenhed 34th International Conference on Machine Learning - Sydney, Australien
Varighed: 6 aug. 201711 aug. 2017

Konference

Konference 34th International Conference on Machine Learning
LandAustralien
BySydney
Periode06/08/201711/08/2017

Fingerprint

Computational efficiency

Citer dette

Masegosa, A., Nielsen, T. D., Langseth, H., Ramos-López, D., Salmerón, A., & Madsen, A. L. (2017). Bayesian models of data streams with Hierarchical Power Priors. I Proceedings of the 34th International Conference on Machine Learning (Bind 70, s. 2334-2343)
Masegosa, Andrés ; Nielsen, Thomas Dyhre ; Langseth, Helge ; Ramos-López, Darío ; Salmerón, Antonio ; Madsen, Anders Læsø. / Bayesian models of data streams with Hierarchical Power Priors. Proceedings of the 34th International Conference on Machine Learning. Bind 70 2017. s. 2334-2343
@inproceedings{4e6fc97950ac4e8b9ae9a1f0631641a3,
title = "Bayesian models of data streams with Hierarchical Power Priors",
abstract = "Making inferences from data streams is a pervasive problem in many modern data analysis applications. But it requires to address the problem of continuous model updating, and adapt to changes or drifts in the underlying data generating distribution. In this paper, we approach these problems from a Bayesian perspective covering general conjugate exponential models. Our proposal makes use of non-conjugate hierarchical priors to explicitly model temporal changes of the model parameters. We also derive a novel variational inference scheme which overcomes the use of non-conjugate priors while maintaining the computational efficiency of variational methods over conjugate models. The approach is validated on three real data sets over three latent variable models.",
author = "Andr{\'e}s Masegosa and Nielsen, {Thomas Dyhre} and Helge Langseth and Dar{\'i}o Ramos-L{\'o}pez and Antonio Salmer{\'o}n and Madsen, {Anders L{\ae}s{\o}}",
year = "2017",
language = "English",
volume = "70",
pages = "2334--2343",
booktitle = "Proceedings of the 34th International Conference on Machine Learning",

}

Masegosa, A, Nielsen, TD, Langseth, H, Ramos-López, D, Salmerón, A & Madsen, AL 2017, Bayesian models of data streams with Hierarchical Power Priors. i Proceedings of the 34th International Conference on Machine Learning. bind 70, s. 2334-2343, Sydney, Australien, 06/08/2017.

Bayesian models of data streams with Hierarchical Power Priors. / Masegosa, Andrés; Nielsen, Thomas Dyhre; Langseth, Helge; Ramos-López, Darío; Salmerón, Antonio; Madsen, Anders Læsø.

Proceedings of the 34th International Conference on Machine Learning. Bind 70 2017. s. 2334-2343.

Publikation: Bidrag til bog/antologi/rapport/konference proceedingKonferenceartikel i proceedingForskningpeer review

TY - GEN

T1 - Bayesian models of data streams with Hierarchical Power Priors

AU - Masegosa, Andrés

AU - Nielsen, Thomas Dyhre

AU - Langseth, Helge

AU - Ramos-López, Darío

AU - Salmerón, Antonio

AU - Madsen, Anders Læsø

PY - 2017

Y1 - 2017

N2 - Making inferences from data streams is a pervasive problem in many modern data analysis applications. But it requires to address the problem of continuous model updating, and adapt to changes or drifts in the underlying data generating distribution. In this paper, we approach these problems from a Bayesian perspective covering general conjugate exponential models. Our proposal makes use of non-conjugate hierarchical priors to explicitly model temporal changes of the model parameters. We also derive a novel variational inference scheme which overcomes the use of non-conjugate priors while maintaining the computational efficiency of variational methods over conjugate models. The approach is validated on three real data sets over three latent variable models.

AB - Making inferences from data streams is a pervasive problem in many modern data analysis applications. But it requires to address the problem of continuous model updating, and adapt to changes or drifts in the underlying data generating distribution. In this paper, we approach these problems from a Bayesian perspective covering general conjugate exponential models. Our proposal makes use of non-conjugate hierarchical priors to explicitly model temporal changes of the model parameters. We also derive a novel variational inference scheme which overcomes the use of non-conjugate priors while maintaining the computational efficiency of variational methods over conjugate models. The approach is validated on three real data sets over three latent variable models.

UR - http://proceedings.mlr.press/v70/masegosa17a/masegosa17a.pdf

M3 - Article in proceeding

VL - 70

SP - 2334

EP - 2343

BT - Proceedings of the 34th International Conference on Machine Learning

ER -

Masegosa A, Nielsen TD, Langseth H, Ramos-López D, Salmerón A, Madsen AL. Bayesian models of data streams with Hierarchical Power Priors. I Proceedings of the 34th International Conference on Machine Learning. Bind 70. 2017. s. 2334-2343