Foveation-based Mechanisms Alleviate Adversarial Examples

Foveation-based Mechanisms Alleviate Adversarial Examples

We show that adversarial examples, i.e., the visually imperceptible perturbations that result in Convolutional Neural Networks (CNNs) fail, can be alleviated with a mechanism based on foveations---applying the CNN in different image regions. To see this, first, we report results in ImageNet that lea...

Full description

Bibliographic Details
Main Authors: Lou, Yan, Boix, Xavier, Roig, Gemma, Poggio, Tomaso, Zhao, Qi
Format: Technical Report
Language:en_US
Published: Center for Brains, Minds and Machines (CBMM), arXiv 2016
Subjects:
Convolutional Neural Networks (CNN)
Foveation
Online Access:http://hdl.handle.net/1721.1/100981

Internet

http://hdl.handle.net/1721.1/100981

Similar Items