Manifold Relevance Determination

Andreas DamianouCarl Henrik EkMichalis K. TitsiasNeil D. Lawrence
,  29, 2012.

Abstract

In this paper we present a fully Bayesian latent variable model which exploits conditional nonlinear (in)-dependence structures to learn an efficient latent representation. The latent space is factorized to represent shared and private information from multiple views of the data. In contrast to previous approaches, we introduce a relaxation to the discrete segmentation and allow for a “softly” shared latent space. Further, Bayesian techniques allow us to automatically estimate the dimensionality of the latent spaces. The model is capable of capturing structure underlying extremely high dimensional spaces. This is illustrated by modelling unprocessed images with tenths of thousands of pixels. This also allows us to directly generate novel images from the trained model by sampling from the discovered latent spaces. We also demonstrate the model by prediction of human pose in an ambiguous setting. Our Bayesian framework allows us to perform disambiguation in a principled manner by including latent space priors which incorporate the dynamic nature of the data.

Cite this Paper


BibTeX
@InProceedings{pmlr-v-damianou-manifold12, title = {Manifold Relevance Determination}, author = {Andreas Damianou and Carl Henrik Ek and Michalis K. Titsias and Neil D. Lawrence}, year = {}, editor = {}, volume = {29}, address = {San Francisco, CA}, url = {http://inverseprobability.com/publications/damianou-manifold12.html}, abstract = {In this paper we present a fully Bayesian latent variable model which exploits conditional nonlinear (in)-dependence structures to learn an efficient latent representation. The latent space is factorized to represent shared and private information from multiple views of the data. In contrast to previous approaches, we introduce a relaxation to the discrete segmentation and allow for a “softly” shared latent space. Further, Bayesian techniques allow us to automatically estimate the dimensionality of the latent spaces. The model is capable of capturing structure underlying extremely high dimensional spaces. This is illustrated by modelling unprocessed images with tenths of thousands of pixels. This also allows us to directly generate novel images from the trained model by sampling from the discovered latent spaces. We also demonstrate the model by prediction of human pose in an ambiguous setting. Our Bayesian framework allows us to perform disambiguation in a principled manner by including latent space priors which incorporate the dynamic nature of the data.} }
Endnote
%0 Conference Paper %T Manifold Relevance Determination %A Andreas Damianou %A Carl Henrik Ek %A Michalis K. Titsias %A Neil D. Lawrence %B %C Proceedings of Machine Learning Research %D %E %F pmlr-v-damianou-manifold12 %I PMLR %J Proceedings of Machine Learning Research %P -- %U http://inverseprobability.com %V %W PMLR %X In this paper we present a fully Bayesian latent variable model which exploits conditional nonlinear (in)-dependence structures to learn an efficient latent representation. The latent space is factorized to represent shared and private information from multiple views of the data. In contrast to previous approaches, we introduce a relaxation to the discrete segmentation and allow for a “softly” shared latent space. Further, Bayesian techniques allow us to automatically estimate the dimensionality of the latent spaces. The model is capable of capturing structure underlying extremely high dimensional spaces. This is illustrated by modelling unprocessed images with tenths of thousands of pixels. This also allows us to directly generate novel images from the trained model by sampling from the discovered latent spaces. We also demonstrate the model by prediction of human pose in an ambiguous setting. Our Bayesian framework allows us to perform disambiguation in a principled manner by including latent space priors which incorporate the dynamic nature of the data.
RIS
TY - CPAPER TI - Manifold Relevance Determination AU - Andreas Damianou AU - Carl Henrik Ek AU - Michalis K. Titsias AU - Neil D. Lawrence BT - PY - DA - ED - ID - pmlr-v-damianou-manifold12 PB - PMLR SP - DP - PMLR EP - L1 - UR - http://inverseprobability.com/publications/damianou-manifold12.html AB - In this paper we present a fully Bayesian latent variable model which exploits conditional nonlinear (in)-dependence structures to learn an efficient latent representation. The latent space is factorized to represent shared and private information from multiple views of the data. In contrast to previous approaches, we introduce a relaxation to the discrete segmentation and allow for a “softly” shared latent space. Further, Bayesian techniques allow us to automatically estimate the dimensionality of the latent spaces. The model is capable of capturing structure underlying extremely high dimensional spaces. This is illustrated by modelling unprocessed images with tenths of thousands of pixels. This also allows us to directly generate novel images from the trained model by sampling from the discovered latent spaces. We also demonstrate the model by prediction of human pose in an ambiguous setting. Our Bayesian framework allows us to perform disambiguation in a principled manner by including latent space priors which incorporate the dynamic nature of the data. ER -
APA
Damianou, A., Ek, C.H., Titsias, M.K. & Lawrence, N.D.. (). Manifold Relevance Determination. , in PMLR :-

Related Material