Harmonic Decompositions of Convolutional Networks

The dataset used in this paper is a collection of images of faces, each with a different expression.

BibTex: