Neural coding by redundancy reduction and correlation

Redundancy reduction as a form of neural coding has been a topic of large research interest. A number of strategies has been proposed, but the one which is attracting the most attention assumes that this coding is carried out so that the output signals are mutually independent. In this work we go one step further and suggest an algorithm that separates also non-orthogonal signals (i.e. dependent signals). The resulting algorithm is very simple, as it is computationally economical and based on second order statistics that, as it is well know, is more robust to errors than higher order statistics. Moreover, the permutation/scaling problem is also avoided. The framework is given with a biological background, and we point out that the algorithm can also be used in other applications such as biomedical engineering and telecommunications.