qGaussian Distribution Proposed for Robust Sparse Learning in Correlated Data

A recent paper on arXiv, identified as 2608.17244, questions the dependence on Gaussian distribution within statistical sparse learning when dealing with correlated and heterogeneous datasets. The authors contend that Gaussian models are often inadequate in the presence of outliers and overly sensitive to distributional assumptions, which hinders their application in areas such as biostatistics. They introduce the qGaussian distribution, obtained through Tsallis entropy maximization, as a more resilient alternative. Additionally, the paper re-derives the multivariate probability density function for correlated datasets and presents a computational framework that employs numerical methods for flow systems, enhancing sparse learning with the qGaussian model. This methodology is significant for statistical modeling and machine learning, particularly when data independence and outlier resilience are essential.