Using optimal transport (Wasserstein distance) to measure non-Gaussianity provides a theoretically grounded and practically effective alternative to classical ICA methods, with better performance across different data distributions.
This paper proposes OT-ICA, a new algorithm for Independent Component Analysis that uses Wasserstein distance instead of traditional proxy measures like cumulants. The key insight is that maximizing the Wasserstein distance between a projection and a standard Gaussian recovers independent components.