SCSSE Seminar

Title: Manifold learning and its application in posture detection
Speaker: Mr. Peng Cheng

Day: Wednesday 9nd September, 2009
Location: Building 3. Tearoom
Time: 11.30-12.30

Abstract: Manifold learning refers to the problem of detecting lower-dimensional substructure embedded in noisy and high-dimensional data. It is a critical problem in machine learning. Kernel Principle Component Analysis (KPCA) is a typical manifold learning technique that offers an implicit (hyperplane) representation of the manifold rather than parameterizing the manifold. However, KPCA has high computational cost since all training samples have to be kept for the representation. In this talk, we present our recently proposed algorithm that approximates the manifold with a compact representation by discarding redundant samples. This algorithm is employed and evaluated in detecting specific types of human postures from single images. Experimental results have verified its efficiency and effectiveness.