Sparse data imputation with Bayesian non-linear factor analysis

TMLR Paper5052 Authors

Abstract: We propose a new method for non-linear modeling of latent variables and factors via random Fourier features for high-dimensional data. Essentially, we apply a basis function expansion of a factor analysis model to approximate a Gaussian process mapping of the latent variable and the latent factors to the observed data space. This paper demonstrates the effectiveness of our proposed model with experiments on real datasets in comparison with competing latent variable models. In particular, we show that our proposed model is effective for missing data imputation, especially when the percentage of missing data is high.

Submission Length: Regular submission (no more than 12 pages of main content)

Assigned Action Editor: ~Jes_Frellsen1

Submission Number: 5052