Summary
International Symposium on Nonlinear Theory and its Applications
2017
Session Number:A1L-A
Session:
Number:A1L-A-5
Dynamic Learning of Embeddings for Cognitive Classification from High-Dimensional Data
Jr-Shin Li, Liang Wang, Wei Zhang,
pp.11-11
Publication Date:2017/12/4
Online ISSN:2188-5079
DOI:10.34385/proc.29.A1L-A-5
PDF download (26.2KB)
Summary:
Supervised machine learning provides powerful tools for data classification and pattern recognition. However, for classification tasks involving time-series data presenting active dynamic features, many of the state-of-the-art classifiers may not perform well because they overlook the underlying dynamic temporal properties of the data. In this work, we develop a dynamic learning framework integrating the Koopman operator theory and support vector machines, which enables the embeddings of high-dimensional data to a low-dimensional space through extracting dynamics of the underlying dynamical system. We apply and validate the developed methodology by showing high accuracy in cognitive classifications using fMRI visual cognition datasets.