|
Abstract:
|
In order to extract the special patterns of brain activities from fMRI data, I propose a two-way clustering learning rule with a two-layered distance procedure to capture the data geometry in this research. The computations for clustering are primarily based on the new non-supervised learning algorithm, Data Cloud Geometry (DCG). By coupling the two DCG trees, one for the subject space and the other for the covariate space, many partial interaction patterns between subject-clusters and covariate-clusters can be determined. Also, I demonstrate that how to adaptively evolve a simple empirical distance into an effective one to facilitate an efficient global feature-matrix for learning purposes. Several datasets are demonstrated to show the efficiencies of my proposed methods. The methods can also be used in feature extraction. The advantages of my learning rule by comparing its performance to the other commonly used machine learning techniques will be demonstrated. This research provides an innovative method for fMRI data classification and brain states decoding.
|
