Section: New Software and Platforms

Convolutional Kernel Networks

Participants : Julien Mairal, Piotr Koniusz, Zaid Harchaoui, Cordelia Schmid.

This is an open-source software package corresponding to a paper published at NIPS in 2014, available at http://ckn.gforge.inria.fr/ , and which is continuously updated. In this software package, convolutional neural networks are learned in an unsupervised manner. We control what the non-linearities of the network are really doing: the network tries to approximate the kernel map of a reproducing kernel.