Hierarchical Gaussian Naive Bayes Classifier for Multiple-Subject fMRI Data

Posted in Science on August 22, 2008


Hierarchical Gaussian Naive Bayes Classifier for Multiple-Subject fMRI Data

The Gaussian Na¨?ve Bayes (GNB) [2] classifier has been successfully applied to fMRI data. However, it is not specifically designed to account for data from multiple subjects and is usually applied to data from a single subject (referred to as GNB-indiv). An extension to the GNB classifier has been proposed ([4], referred to as GNB-pooled), in which the data from all the subjects are combined together na¨?vely by assuming that they all come from the same subjects. However, this extension ignores subject-specific variations that might exist. Here I describe another extension of the GNB classifier—the hierarchical GNB classifier [3]—that can account for subject-specific variations, and in addition, has the flexibility to increase or reduce the weight of the contribution of the data from the other subjects based on the number of examples available from the test subject.

Author: Indrayana Rustandi, Computer Science Department, Carnegie Mellon University

Watch Video

Tags: Science, Lectures, Computer Science, Machine Learning, VideoLectures.Net, Gaussian Processes, Image Analysis, fMRI - Analysis