大阪大学　数理・データ科学教育研究センター
Center for Mathematical Modeling and Data Science,Osaka University

COMPOSITIONAL MODELS FOR DATA MINING: AN ILLISTRATIVE EXAMPLE

Radim Jiroušek (Professor Emeritus), Václav Kratochvíl (Senior Researcher), The Czech Academy of Sciences, Institute of Information Theory and Automation

1. ホーム
2. セミナー・集中講義
3. セミナー
4. COMPOSITIONAL MODELS FOR DATA MINING: AN ILLISTRATIVE EXAMPLE

大阪大学 数理・データ科学セミナー　数理モデルセミナーシリーズ　第17回

COMPOSITIONAL MODELS FOR DATA MINING: AN ILLISTRATIVE EXAMPLE

Radim Jiroušek (Professor Emeritus), Václav Kratochvíl (Senior Researcher), The Czech Academy of Sciences, Institute of Information Theory and Automation

　Like Bayesian networks, also compositional models can be used to acquire knowledge from data. The knowledge is acquired during a supervised process of model learning. The process, which will be explained by a simple illustrative example, take full advantage of the information-theoretic characteristics (mutual information, informational content and Kullback-Leibler divergence), the properties of which are for compositional models known. The supervised process is used because of the following main reasons.
　First, no generally accepted method for optimum model construction is known.
　Second, the user usually has some prior knowledge about the area of application, and this knowledge should be utilized during the process. Further, the user can have some knowledge about data, based on which the model is constructed. They may know that the data are not well stratified and some properties should be suppressed some others highlighted. All this prior knowledge is thus fully employed in the process of new knowledge mining.