Timezone: »

LaVAN: Localized and Visible Adversarial Noise
Danny Karmon · Daniel Zoran · Yoav Goldberg

Thu Jul 12 07:50 AM -- 08:00 AM (PDT) @ A7

Most works on adversarial examples for deep-learning based image classifiers usenoise that, while small, covers the entire image. We explore the case where thenoise is allowed to be visible but confined to a small, localized patch of theimage, without covering any of the main object(s) in the image. We show that it is possible to generate localized adversarial noises that cover only 2% of the pixels in the image, none of them over the main object, and that are transferable acrossimages and locations, and successfully fool a state-of-the-art Inception v3model with very high success rates.

Author Information

Danny Karmon (Bar Ilan University)
Daniel Zoran (DeepMind)
Yoav Goldberg (Bar Ilan University)

Related Events (a corresponding poster, oral, or spotlight)

More from the Same Authors