Adaptive learning machines for nonlinear classification and Bayesian information criteria. (English) Zbl 1270.68235

Ando, Tomohiro; Imoto, Seiya; Konishi, Sadanori

Bull. Inf. Cybern. 36, 147-162 (2004).

Summary: Regularization is a well-known method for the treatment of mathematically ill-posed problems. By using the method of regularization, we propose a new machine learning algorithm, adaptive learning machine, to classify the high-dimensional data with complex structure. A crucial issue in the model constructing process is the choice of a suitable model among candidates. We present a Bayesian information criterion to evaluate models estimated by regularization. Real data analysis and Monte Carlo experiments show that our proposed method performs well in various situations.

MSC:

68T05	Learning and adaptive systems in artificial intelligence
62H30	Classification and discrimination; cluster analysis (statistical aspects)

Keywords:

Bayes approach; classification; genetic algorithm; regularization theory