Model-based sparse coding beyond Gaussian independent model

Sparse coding aims to model data vectors as sparse linear combinations of basis elements, but a majority related studies are restricted continuous without spatial or temporal structure. A new model-based (MSC) method is proposed provide an effective and flexible framework for learning features from different types: continuous, discrete, categorical, modeling types correlations: temporal. The specification the sparsity level how adapt estimation large-scale also addressed. fast EM algorithm estimation, its superior performance demonstrated in simulation multiple real applications such image denoising, brain connectivity study, transcriptomic imaging.