Pattern Repulsion Revisited

Abstract

Marques and Almeida recently proposed a nonlinear data seperation technique based on the maximum entropy principle of Bell and Sejnowsky. The idea behind is a pattern repulsion model alluding to repulsion of physical particles in a restricted system which leads to a uniform distribution, hence a maximum entropy, under certain conditions to the energy function. In this paper, we want to revisite their model and give a rigorous mathematical framework in which this algorithm indeed converges to a demixing function.