From this we see that an iterative approach for OO is evolving along the line of the iterative approach in traditional cardinal optimization. Instead of marching from one point to another in the optimization space, we march along one representation or search subspace to another. This way knowledge and learning can be gradually injected into the search. Much remains to be done. Only the surface has been scratched. This problem of representation is also known as Bias Selection or Encoding in AI. [See D. Gordon and M. Desjardins, "Evaluation and Selection of Bias in Machine Learning", Machine Learning, 20:5-22, 1995, and Wheeler Ruml, J. Thomas Ngo, Joel Marks, and Stuart Sheiber, "Easily Search Encoding for Number Partitioning", JOTA, to appear 1996].