An Information-centric Algorithm for Feature Extraction in High-dimensional Data

Abstract

This thesis develops a novel technique for extracting features in high-dimensional data. The proposed method is based on the concept of maximal correlation and local information theory, which demonstrates the importance of the information vector space in feature extraction. More specifically, a hidden Markov model is used to consider the relation between high-dimensional data and their low-dimensional features. Feature extraction is regarded as an optimization problem to figure out the corresponding information vector space. Several approaches are proposed to solve this problem and mathematical proof is provided to validate the effectiveness of them.