Optimal Kotz Distribution Hyperparameters For Infomax Affinity

A recently proposed method for blind source separation called multidataset independent subspace analysis (MISA) leverages the general Kotz distribution to model latent source distributions. The parameters of that distribution allow it to flexibly model both super- and sub-gaussian sources. Here, we demonstrate a principled strategy to tune these parameters in order to optimally approximate a logistic distribution and, effectively, typical Infomax estimates. Such affinity is valuable given the success of Infomax in the field of functional magnetic resonance imaging (fMRI).