Bayesian source separation of linear-quadratic and linear mixtures through a MCMC method

In this work, we deal with source separation of linear - quad-ratic (LQ) and linear mixtures. By relying on a Bayesian approach, the developed method allows one to take into account prior informations such as the non-negativity and the temporal structure of the sources. Concerning the inference scheme, the implementation of a Gibbs' sampler equipped with latent variables simplifies the sampling steps. The obtained results confirm the effectiveness of the new proposal and indicate that it may be particularly useful in situations where classical ICA-based solutions fail to separate the sources.