New Dimensions of Information Warfare by Roberto Di Pietro & Simone Raponi & Maurantonio Caprolu & Stefano Cresci

Author:Roberto Di Pietro & Simone Raponi & Maurantonio Caprolu & Stefano Cresci
Language: eng
Format: epub
ISBN: 9783030606183
Publisher: Springer International Publishing

Gradient Hiding

A large family of adversarial attacks exploits the known characteristics of the attacked model. Among the most important of such characteristics is the gradient of deep neural network models, which is used during the training phase of the model to tune the model’s parameters. However, the same information can also be used at test time to tune adversarial inputs so as to trigger wrong classifications by the model. The gradient hiding technique represents a natural defense against gradient-based attacks and attacks using adversarial crafting methods (e.g., FGSM [248]) and simply consists of hiding the information about the model’s gradient from the adversary. For instance, if the model is non-differentiable (e.g., decision trees, nearest neighbor classifiers, random forests), gradient-based attacks are rendered ineffective.

Download

Copyright Disclaimer:
This site does not store any files on its server. We only index and link to content provided by other sites. Please contact the content providers to delete copyright contents if any and email us, we'll remove relevant links or contents immediately.