Xuchen Feng

CDFlow introduces efficient invertible layers for generative models

Researchers have developed CDFlow, a novel approach to building invertible layers for deep generative models using circulant and diagonal matrices. This method reduces parameter complexity and computational cost for mat…