タイトル | A Note on Concept Learning with Membership Queries |
---|---|
著者 | 浮田善文 、松嶋敏泰 、平澤茂一 |
年度 | 1996 |
形式 | 国際学会 |
分野 | 知識情報処理 |
掲載雑誌名 | Proceedings of International Symposium on Information Theory and Its Applications (Canada) |
掲載号・ページ | pp.308-311, Victoria, Canada |
掲載年 | 1996 |
掲載月 | 9 |
アブスト (日本語) |
学会名:1996 International Symposium on Information Theory and Its Applications (ISITA1996) 日程:1996年9月17日~1996年9月20日 場所:Victoria, Canada |
アブスト (英語) |
In this paper, a new model of learning from queries is proposed. First, the two criteria to evaluate learning algorithms under a limited number of queries are newly proposed. Next, a tight bound of these criteria is derived when the branch-and-bound algorithm is applied to construct an optimal learning strategy, i.e., a presentation order for the instances. Moreover, it is shown that a learning strategy can be selected more efficiently in the case that the instance space of the concept class satisfies symmetric property such as monotone monomials. Consequently, the algorithm using the derived bound and the symmetric property of the instance space can extremely reduce the search complexity to select the optimal learning strategy for each criterion. |
備考 (日本語) |
1 |
備考 (英語) |
1 |
論文原稿 | |
発表資料 |