Combination of self organizing maps and feedforward nets inspired by population coding

Self-organizing feature maps and feedforward nets trained with the backpropagation algorithm are the most common used and the most analyzed artificial neural nets. Although a combination of them, called counterpropagation, already had been proposed, its full capabilities had not been recognized. In this paper an extension of this net structure is proposed, which takes into account the whole activity pattern of the self-organizing feature mappsilas competitive layer. It is inspired by both neurological studies about information coding in the human brain and the psychological thesis of H. Benesch, who states that each information processing system is based on the triple of medium, pattern and meaning. A practical example using situations in robot soccer games is given and the results are analyzed.