Self-sufficient Independent Component Analysis for Demixing Flows
The paper introduces Self-sufficient Independent Component Analysis (SICA) to learn disentangled features from data using non-linear Independent Component Analysis (ICA). It formulates ICA as minimizing a conditional KL divergence based on a self-sufficiency assumption, leading to an iterative de-mixing flow learning process. This method avoids unstable adversarial training and achieves superior performance in signal separation tasks.