クラスタ例からの学習(機械学習)(<特集>人工知能分野における博士論文)

概要

論文の詳細を見る
Learning from cluster examples (LCE) is a hybrid task combining features of two common classification tasks : clustering and learning from examples. In LCE, each example is an object set with the true partition for the set, where the true partition is the one that users consider as the most appropriate for their aim among the possible partitions. The task is then to acquire a rule for partitioning unseen object sets from this example set. A method for learning such partitioning rules is useful in any situation where explicit algorithms for deriving partitions are hard to formalize, but where individual examples of true partitions are easy to specify. Clustering techniques have been of necessity applied to such situations, despite being essentially unsuited to the problems. We point out faults in using clustering techniques under such a situation, and explain why the techniques for LCE task expected to be overcome these faults. We then present a solution technique for LCE task, and apply the method to the problems in two domains; one with dot patterns and the other with more realistic vector-data images.
社団法人人工知能学会の論文
2001-11-01