Nonparametric learning of dictionaries for sparse representation of sensor signals

Nonparametric Bayesian techniques are considered for learning dictionaries for sparse data representations, with applications in sparse rendering of sensor data. The beta process is employed as a prior for learning the dictionary, and this nonparametric method naturally infers an appropriate dictionary size. The proposed method can learn a sparse dictionary, and may also be used to denoise a signal under test. The noise variance need not be known, and can be non-stationary. The dictionary coefficients for a given sensor signal may be employed within a classifier. Several example results are presented, using both Gibbs and variational Bayesian inference, with comparisons to other state-of-the-art approaches.