DIRECTIONS IN DATA SCIENCE: Michael Bronstein 19 November 2015 at 12:30 Room A14
Social Network Analysis Research Center
Data d'inizio: 19 Novembre 2015
Data di fine: 20 Novembre 2015
Deep learning in geometric data
Assistant Professor
Faculty of Informatics
USI Università della Svizzera italiana
- 19 November 2015
- 12:30 - 13:30
- Room A14
The past decade has witnessed the re-emergence of "deep learning." In computer vision research, convolutional neural network techniques have allowed us to learn task-specific features from examples and achieve a breakthrough in performance in a wide range of applications. However, in the geometry processing and computer graphics communities, these methods are practically unknown. One of the reasons stems from the fact that 3D shapes (typically modelled as Riemannian manifolds) are not shift-invariant spaces, hence the very notion of convolution is rather elusive.
In this talk, Michael Bronstein will show some recent works from his research group that attempt to bridge this gap. Specifically, he will show the construction of intrinsic convolutional neural networks on meshes and point clouds, with applications such as finding dense correspondence between deformable shapes and shape retrieval. More broadly, such methods could be applicable to learning generic geometric structures such as social graphs and networks.
Based on joint works with D. Boscaini, J. Masci, P. Vandergheynst, S. Melzi, U. Castellani, E. Rodola', D. Cremers
