Bayesian exponential family projections for coupled data sources
Arto Klami, Seppo Virtanen, Samuel Kaski
Exponential family extensions of principal component analysis (EPCA) have received a considerable amount of attention in recent years, demonstrating the growing need for basic modeling tools that do not assume the squared loss or Gaussian distribution. We extend the EPCA model toolbox by present- ing the first exponential family multi-view learning methods of the partial least squares and canonical correlation analysis, based on a unified representation of EPCA as matrix factorization of the natural parameters of ex- ponential family. The models are based on a new family of priors that are generally us- able for all such factorizations. We also in- troduce new inference strategies, and demon- strate how the methods outperform earlier ones when the Gaussianity assumption does not hold.
PDF Link: /papers/10/p286-klami.pdf
AUTHOR = "Arto Klami
and Seppo Virtanen and Samuel Kaski",
TITLE = "Bayesian exponential family projections for coupled data sources",
BOOKTITLE = "Proceedings of the Twenty-Sixth Conference Annual Conference on Uncertainty in Artificial Intelligence (UAI-10)",
PUBLISHER = "AUAI Press",
ADDRESS = "Corvallis, Oregon",
YEAR = "2010",
PAGES = "286--293"