RECURSIVE INTERFEROMETRIC REPRESENTATIONS


Signal and image analysis requires building invariant representations to deformations such as
non rigid translations, scalings and rotations for images, or transpositions for sounds. After reviewing state of the art learning and computer vision approaches, it will be shown that recursive interferometric transformations provide invariants needed for general pattern and texture estimation. It also provides a mathematical and algorithmic framework to understand and simplify deep learning neural network architectures, and applications will be given.


Professor Stéphane Mallat
École Polytechnique
Centre de Mathématiques Appliquées
91128 Palaiseau  France