A New Probabilistic Approach to Independent Component Analysis Suitable for On-Line Learning in Artificial Neural Networks

Abstract

Recently, elements of probabilistic model that are suitable for modeling of learning algorithms in biologically plausible artificial neural networks framework, have been introduced. Model was based on two of the main concepts in quantum physics – a density matrix and the Born rule. In this paper we will show that proposed probabilistic interpretation is suitable for modeling of on-line learning algorithms for Independent Component Analysis (ICA), which could be realized on parallel hardware based on very simple computational units. Proposed concept (model) can be used in the context of improving algorithm convergence speed, learning factor choice, input signal scale robustness, and can be easily deployed on parallel hardware.