Learning with Data Adaptive Features
Abstract
It is frequently observed that the dimension of inputs is much larger than the sample size. Examples are image construction, microarray data, data mining etc. In such cases, standard learning methods either are not applicable or perform badly. Also, identifying a small subset of important features, which discriminate outputs, becomes an important subject. Hence, good learning algorithms with high dimensional inputs should provides a classification rule which not only yields high accuracy but also has an ability of identifying few important features. Apparently, there are two popularly used such algorithms. One is decision trees and the other is LASSO. The objective of this research is to develop new algorithms for decision tree and LASSO for improving computational power, prediction accuracy, and ability of detecting significant features.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 27, 2006
- Accession Number
- ADA451670
Entities
People
- Yongdai Kim
Organizations
- Seoul National University