Jonathan Kobold a soutenu sa thèse à l’Université d’Évry Paris-Saclay en 2019, sous la direction de Vincent Vigneron et de Hichem Maaref, de l’équipe SIAM de IBISC.
Titre: The Receptive Field in CNNs
Convolutional neural networks (CNNs) are currently the number one tool for pattern recognition, object detection and segmentation. At the core of these network architectures is the concept of the receptive field. If it is too small a CNN architecture will not be able learn its task. Developers of new CNN architectures need a good understanding of the receptive field. It is also useful to determine whether a CNN architecture is able to learn a given task or not. This talk will give an introduction to the receptive field and why it is so important. The different methods for increasing the receptive field are discussed and a method for investigating the receptive field, the effective receptive field, will be presented.
- Date: 27/02/2020, 14h
- Lieu: IBISC, site Pelvoux, amphithéâtre Bx30
- Invitante: Naïma AIT OUFROUKH-MAMMAR (MCF Univ. Evry, équipe SIAM)