Description:

Ideal PCA replaces the square kernel matrix k(X,X) in kernel PCA with a non-square kernel matrix k(X,Z) - the cross-kernel matrix - where Z are points different from the data input.

(a) Cross-kernels scale more favourably in the number of data points, allowing to obtain PCA-like components more quickly (in linear time).

(b) The right singular values of a cross-kernel-like matrix can be used for manifold learning with kernels.

The IPCA package allows extraction of both kind of features (left/right) and also implements several derived ones which can then be used in further learning tasks. See

Franz J. Király, Martin Kreuzer, Louis Theran. Learning with Cross-Kernels and IPCA. http://arxiv.org/abs/1406.2646

Changes to previous version:

Initial Announcement on mloss.org.

BibTeX Entry: Download

Corresponding Paper BibTeX Entry: Download

Supported Operating Systems: Agnostic

Data Formats: Agnostic

Tags: Manifold Learning, Feature Extraction, Dimension Reduction, Principal Components

Archive: download here

Comments

No one has posted any comments yet. Perhaps you'd like to be the first?

Leave a comment

You must be logged in to post comments.