February 26, 2010 : Zaid Harchaoui

INRIA, LEAR Project-team

Temporal Segmentation with Kernel Change-point Detection

We introduce a kernel-based approach for temporal segmentation based on a, kernel-based change-point detector. We propose a test statistic based upon the, maximum kernel Fisher discriminant ratio as a measure of homogeneity between, segments. We establish both consistency in level and consistency in power for our, test statistics in a large sample setting. Applying a sequence of this test statistic, through a sliding-window along the signal yields an efficient kernel-based temporal, segmentation method. Promising experimental results in temporal segmentation, of mental tasks from BCI data and human action segmentation are presented.

Note : this meeting is scheduled for Friday 02/26 at 2:30pm

seminars/seminaritems/2010-02-26.txt · Last modified: 2010/02/25 12:09 by koray