Backpropatation can also be likened to creating feature attributes in a vector which are scored and ranked making for rich continuously valued vectors which can then be compared (or treed) for similarity using experimental methods.