Dependent normalized random measures

Changyou Chen, Vinayak Rao, Wray Buntine, Yee Whye Teh

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review

Abstract

In this paper we propose two constructions of dependent normalized random measures, a class of nonparametric priors over dependent probability measures. Our constructions, which we call mixed normalized random measures (MNRM) and thinned normalized random measures (TNRM), involve (respectively) weighting and thinning parts of a shared underlying Poisson process before combining them together. We show that both MNRM and TNRM are marginally normalized random measures, resulting in well understood theoretical properties. We develop marginal and slice samplers for both models, the latter necessary for inference in TNRM. In time-varying topic modeling experiments, both models exhibit superior performance over related dependent models such as the hierarchical Dirichlet process and the spatial normalized Gamma process.
Original languageEnglish
Title of host publicationProceedings of the 30th International Conference on Machine Learning (ICML 2013)
Subtitle of host publicationJune 16 – June 21, 2013, Atlanta, Georgia, USA
EditorsSanjoy Dasgupta, David McAllester
PublisherInternational Machine Learning Society (IMLS)
Pages2006-2014
Number of pages9
Publication statusPublished - 2013
Externally publishedYes
EventInternational Conference on Machine Learning 2013 - Atlanta, United States of America
Duration: 16 Jun 201321 Jun 2013
Conference number: 30th
https://icml.cc/Conferences/2013/

Conference

ConferenceInternational Conference on Machine Learning 2013
Abbreviated titleICML 2013
CountryUnited States of America
CityAtlanta
Period16/06/1321/06/13
Internet address

Cite this

Chen, C., Rao, V., Buntine, W., & Teh, Y. W. (2013). Dependent normalized random measures. In S. Dasgupta, & D. McAllester (Eds.), Proceedings of the 30th International Conference on Machine Learning (ICML 2013): June 16 – June 21, 2013, Atlanta, Georgia, USA (pp. 2006-2014). International Machine Learning Society (IMLS).
Chen, Changyou ; Rao, Vinayak ; Buntine, Wray ; Teh, Yee Whye. / Dependent normalized random measures. Proceedings of the 30th International Conference on Machine Learning (ICML 2013): June 16 – June 21, 2013, Atlanta, Georgia, USA. editor / Sanjoy Dasgupta ; David McAllester. International Machine Learning Society (IMLS), 2013. pp. 2006-2014
@inproceedings{d8ab3257639b4c0886fc70401b502341,
title = "Dependent normalized random measures",
abstract = "In this paper we propose two constructions of dependent normalized random measures, a class of nonparametric priors over dependent probability measures. Our constructions, which we call mixed normalized random measures (MNRM) and thinned normalized random measures (TNRM), involve (respectively) weighting and thinning parts of a shared underlying Poisson process before combining them together. We show that both MNRM and TNRM are marginally normalized random measures, resulting in well understood theoretical properties. We develop marginal and slice samplers for both models, the latter necessary for inference in TNRM. In time-varying topic modeling experiments, both models exhibit superior performance over related dependent models such as the hierarchical Dirichlet process and the spatial normalized Gamma process.",
author = "Changyou Chen and Vinayak Rao and Wray Buntine and Teh, {Yee Whye}",
year = "2013",
language = "English",
pages = "2006--2014",
editor = "Sanjoy Dasgupta and David McAllester",
booktitle = "Proceedings of the 30th International Conference on Machine Learning (ICML 2013)",
publisher = "International Machine Learning Society (IMLS)",

}

Chen, C, Rao, V, Buntine, W & Teh, YW 2013, Dependent normalized random measures. in S Dasgupta & D McAllester (eds), Proceedings of the 30th International Conference on Machine Learning (ICML 2013): June 16 – June 21, 2013, Atlanta, Georgia, USA. International Machine Learning Society (IMLS), pp. 2006-2014, International Conference on Machine Learning 2013, Atlanta, United States of America, 16/06/13.

Dependent normalized random measures. / Chen, Changyou; Rao, Vinayak; Buntine, Wray; Teh, Yee Whye.

Proceedings of the 30th International Conference on Machine Learning (ICML 2013): June 16 – June 21, 2013, Atlanta, Georgia, USA. ed. / Sanjoy Dasgupta; David McAllester. International Machine Learning Society (IMLS), 2013. p. 2006-2014.

Research output: Chapter in Book/Report/Conference proceedingConference PaperResearchpeer-review

TY - GEN

T1 - Dependent normalized random measures

AU - Chen, Changyou

AU - Rao, Vinayak

AU - Buntine, Wray

AU - Teh, Yee Whye

PY - 2013

Y1 - 2013

N2 - In this paper we propose two constructions of dependent normalized random measures, a class of nonparametric priors over dependent probability measures. Our constructions, which we call mixed normalized random measures (MNRM) and thinned normalized random measures (TNRM), involve (respectively) weighting and thinning parts of a shared underlying Poisson process before combining them together. We show that both MNRM and TNRM are marginally normalized random measures, resulting in well understood theoretical properties. We develop marginal and slice samplers for both models, the latter necessary for inference in TNRM. In time-varying topic modeling experiments, both models exhibit superior performance over related dependent models such as the hierarchical Dirichlet process and the spatial normalized Gamma process.

AB - In this paper we propose two constructions of dependent normalized random measures, a class of nonparametric priors over dependent probability measures. Our constructions, which we call mixed normalized random measures (MNRM) and thinned normalized random measures (TNRM), involve (respectively) weighting and thinning parts of a shared underlying Poisson process before combining them together. We show that both MNRM and TNRM are marginally normalized random measures, resulting in well understood theoretical properties. We develop marginal and slice samplers for both models, the latter necessary for inference in TNRM. In time-varying topic modeling experiments, both models exhibit superior performance over related dependent models such as the hierarchical Dirichlet process and the spatial normalized Gamma process.

UR - http://www.scopus.com/inward/record.url?scp=84897505920&partnerID=8YFLogxK

M3 - Conference Paper

SP - 2006

EP - 2014

BT - Proceedings of the 30th International Conference on Machine Learning (ICML 2013)

A2 - Dasgupta, Sanjoy

A2 - McAllester, David

PB - International Machine Learning Society (IMLS)

ER -

Chen C, Rao V, Buntine W, Teh YW. Dependent normalized random measures. In Dasgupta S, McAllester D, editors, Proceedings of the 30th International Conference on Machine Learning (ICML 2013): June 16 – June 21, 2013, Atlanta, Georgia, USA. International Machine Learning Society (IMLS). 2013. p. 2006-2014