Open AccessEstimation of Intrinsic Dimension using Supervised Parameter Selection Method
Author(s) -
Haiquan Qiu,
Shulun Yang
Publication year - 2019
Publication title -
journal of physics. conference series
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.21
H-Index - 85
eISSN - 1742-6596
pISSN - 1742-6588
DOI - 10.1088/1742-6596/1302/2/022019
Subject(s) - intrinsic dimension , dimension (graph theory) , scaling , multidimensional scaling , projection (relational algebra) , adjacency list , adjacency matrix , selection (genetic algorithm) , computer science , dimensionality reduction , matrix (chemical analysis) , pattern recognition (psychology) , mathematics , artificial intelligence , algorithm , sample (material) , statistics , theoretical computer science , combinatorics , physics , geometry , graph , materials science , composite material , thermodynamics , curse of dimensionality
In this paper, we propose a new method for estimating the intrinsic dimension of datasets. The new method uses the local information of different scales of the sample points (by adjacency matrix) to estimate the intrinsic dimension. The only parameter used in the new method is the scaling ratio k , which determines the adjacency matrix of different scales. We propose a parameter selection method based on the difference of estimated dimension and the classification accuracy of projection data. Experiments on real datasets demonstrate the effectiveness of the proposed method.