Bayesian sparse factor analysis with kernelized observations

Multi-view problems can benefit from latent representations since they find low-dimensional projections that fairly capture the correlations among multiple views characterise data. On other hand, high-dimensionality and non-linear issues are traditionally handled by kernel methods, inducing a (non)-linear function between projection data itself. However, usually come with exposition to overfitting. Here, we combine Bayesian factor analysis what refer as kernelized observations, in which proposed model focuses on reconstructing not itself, but its relationship points measured function. In turn, extend previous FA formulations be able relationships means of and, at same time, include additional facilities obtain compact an automatic selection Relevance Vectors (RVs), feature relevance even, learning approach. Besides, this is flexibly included into modular framework where easily adapt capabilities needs them functionalities such heterogeneous or semi-supervised learning. Using several public databases, demonstrate potential approach (and extensions) w.r.t. common multi-view models canonical correlation analysis, incomplete – variational autoenconder manifold determination, shows ability outperform baselines while indistinctly combining extensions.