Martha White, Yaoliang Yu, Xinhua Zhang and Dale Schuurmans. Convex Multi-view Subspace Learning. In Advances in Neural Information Processing Systems (NIPS), 2012.

Download

[PDF] 

Abstract

Subspace learning seeks a low dimensional representation of datathat enables accurate reconstruction.However, in many applications, data is obtainedfrom multiple sources rather than a single source(e.g. an object might be viewed by cameras at different angles,or a document might consist of text and images).The conditional independence of separate sources imposes constraints ontheir shared latent representation, which, if respected, canimprove the quality of the learned low dimensionalrepresentation.In this paper, we present a convex formulation of multi-view subspacelearning that enforces conditional independence while reducing dimensionality.For this formulation, we develop an efficient algorithm thatrecovers an optimal data reconstruction by exploiting an implicitconvex regularizer, then recovers the corresponding latent representationand reconstruction model, jointly and optimally.Experiments illustrate that the proposed method produces high quality results.

@InProceedings(white2012convex,
  Title = "Convex Multi-view Subspace Learning",
  Author = "Martha White, Yaoliang Yu, Xinhua Zhang, Dale Schuurmans",
  Year = "2012",
  Booktitle = "Advances in Neural Information Processing Systems (NIPS)",
  AcceptRate = "25.2\%",
  AcceptNumbers = "370 of 1467"
  bib2html_dl_pdf="../publications/12nips-multiview.pdf"
)