Conference Paper

Hierarchical Feature Extraction for Compact Representation and Classification of Datasets

DOI: 10.1007/978-3-540-69158-7_58 Conference: Neural Information Processing, 14th International Conference, ICONIP 2007, Kitakyushu, Japan, November 13-16, 2007, Revised Selected Papers, Part I
Source: DBLP


Feature extraction methods do generally not account for hierarchical structure in the data. For example, PCA and ICA provide
transformations that solely depend on global properties of the overall dataset. We here present a general approach for the
extraction of feature hierarchies from datasets and their use for classification or clustering. A hierarchy of features extracted
from a dataset thereby constitutes a compact representation of the set that on the one hand can be used to characterize and
understand the data and on the other hand serves as a basis to classify or cluster a collection of datasets. As a proof of
concept, we demonstrate the feasibility of this approach with an application to mixtures of Gaussians with varying degree
of structuredness and to a clinical EEG recording.

Full-text preview

Available from:
  • Source
    • "Inspiration from human cognition, fast or real-time decision-making with environment is more useful instead of exhaustive search for getting global optimization. From [17], we know that the local information is important, and from analysis of human cognition, hierarchical features extraction and classification [18] [19] also investigated in our study. "
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents a hierarchical feature extraction and classification method for electroencephalogram (EEG) hidden information mining. It consists of supervised learning for fewer features, hierarchical knowledge base (HKB) construction and classification test. First, the discriminative rules and the corresponding background conditions are extracted by using autoregressive method in combination with the nonparametric weighted feature extraction (NWFE) and k-nearest neighbor. Second, through ranking the discriminative rules according to validation test correct rate, a hierarchical knowledge base HKB is constructed. Third, given an EEG sequence, it chooses one or several discriminative rules from the HKB using the up-bottom search strategy and calculates classification accuracy. The experiments are carried out upon real electroencephalogram (EEG) recordings from five subjects and the results show the better performance of our method.
    Full-text · Conference Paper · Jul 2010