Sparse Exponential Family Latent Variable Models
By Shakir Mohamed, Department of Engineering, University of Cambridge.
Latent variable models are an essential tool for the analysis of data in
numerous application areas, whether it be source separation, image analysis,
matrix completion or preference elicitation; since they provide a means of
understanding the inherent structure in data. This talk will examine the
construction of latent variable, and more specifically, factor-analytic models
with sparse properties.
In a Bayesian setting, the discussion will examine the priors that are used to
learn latent representations of data, in the context of the factor-analytic
models generalized to the exponential family. The focus will be on sparse
Bayesian learning: weak sparsity achieved through priors based on the Gaussian
scale-mixture construction; and strong sparsity using discrete mixture priors. A
comparison of these two methods will be given and an efficient sampler for
learning with discrete mixture priors will be described, which has many
advantages over a corresponding optimization approach to learning.