On the Optimal K-term Approximation of a Sparse Parameter Vector MMSE Estimate

• This paper considers approximations of marginalization sums thatarise in Bayesian inference problems. Optimal approximations ofsuch marginalization sums, using a fixed number of terms, are analyzedfor a simple model. The model under study is motivated byrecent studies of linear regression problems with sparse parametervectors, and of the problem of discriminating signal-plus-noise samplesfrom noise-only samples. It is shown that for the model understudy, if only one term is retained in the marginalization sum, thenthis term should be the one with the largest a posteriori probability.By contrast, if more than one (but not all) terms are to be retained,then these should generally not be the ones corresponding tothe components with largest a posteriori probabilities.

Keyword

MMSE estimation

Bayesian inference

marginalization

TECHNOLOGY

TEKNIKVETENSKAP

Publication and Content Type

ref (subject category)

kon (subject category)