Some applications of simulated annealing to pattern recognition

The simulated annealing technique for solving combinatorial problems is applied to cluster analysis, isomorphisms of attributed relational graphs, piecewise curve fitting, and feature selection. A novel class of clustering algorithms based on simulated annealing are presented. One such algorithm, ALKMEANS, is proposed as a contrast to the commonly used heuristic clustering algorithm KMEANS; test results demonstrate that ALKMEANS is superior to KMEANS. A simulated annealing algorithm, ALISON, is presented for the problem of isomorphisms of relational graphs.EI