Vera's suggestion for extending the MI-DN algorithm for finding higher order sm primatives

The figure below shows the archive of s-m-p triplets that we've evolved and stored during phase 1 of NAOs motor babbling. This archive can then be used to construct higher order s-m-p-groups. This implies developmental phases with the first phase being used to detect the lower order s-m-p triplets,


and the next phase being to combine these. It may also be the case that in the second phase, further mutation is permitted, allow new motor and sensory inputs NOT already in the low level pairs to interject and influence the higher order units directly. This is rather similar to what Goren was doing with his pruning based FFNNs but I've decided to use mutual information and a GA, rather than exhaustive search.