A Penalized Autoencoder Approach for Nonlinear Independent Component Analysis

Abstract

We propose Independent Component Autoencoder (ICAE), a deep neural network-based framework for nonlinear Independent Component Analysis (ICA). The proposed method consists of a penalized autoencoder and a training objective that is to minimize a combination of the reconstruction loss and an ICA contrast. Unlike many previous ICA methods that are usually tailored to separate specific mixture, our method can recover sources from various mixtures, without prior knowledge on the nature of that mixture.