On Generalization of Deep Neural Networks for Visual Recognition Tasks

Takayuki Okatani

doi:https://doi.org/10.36463/idw.2020.0961

4:40 PM - 5:05 PM

[AIS1/PRJ1-1(Invited)] On Generalization of Deep Neural Networks for Visual Recognition Tasks

*Takayuki Okatani^1,2 (1. Tohoku University (Japan), 2. RIKEN (Japan))

Deep learning, Computer vision, Image restoration, Out-of-distribution detection

https://doi.org/10.36463/idw.2020.0961

This article discusses generalization ability of deep neural networks (DNNs) for visual recognition tasks. It is known that DNNs easily fail for images to which noises are added, when they have not learned the noisy images. We discuss how to cope with such limitation of DNNs.

Proceedings of the International Display Workshops Volume 27 (IDW '20)

[AIS1/PRJ1-1(Invited)] On Generalization of Deep Neural Networks for Visual Recognition Tasks