Selection of important nodes and feature trees on random trees

FLESIA, ANA GEORGINA

Maresias, Sao Paulo. Brazil. 4 al 11 de agosto del 2007.

Conferencia; X Brazilian School of Probability; 2007

GOAL: Selection of important nodes and “features trees” in supervised and unsupervised classification. This is a correlate of the problem of selection of most informative variables that help to explain a classification method with high-dimensional data. METHODOLOGY:First Approach:  "blinding" uninformative branches of trees,  Second Approach: family of distances that take into account notions of probability of each node. MOTIVATION: A strong motivation for this work is to derive a way of distinguish  when two samples of sequences have been produced by different VLMC models with he same context tree.