Evaluating Defensive Distillation For Defending Text Processing Neural Networks Against Adversarial Examples
This work addresses the problem of adversarial attacks in text processing for AI security, but it is incremental as it applies an existing method to a new domain with negative findings.
The study evaluated defensive distillation, a method used to protect neural networks from adversarial examples in images, for text classification tasks and its effect on adversarial example transferability. The results showed that defensive distillation had minimal impact, failing to improve robustness or prevent transferability in text classification networks.
Adversarial examples are artificially modified input samples which lead to misclassifications, while not being detectable by humans. These adversarial examples are a challenge for many tasks such as image and text classification, especially as research shows that many adversarial examples are transferable between different classifiers. In this work, we evaluate the performance of a popular defensive strategy for adversarial examples called defensive distillation, which can be successful in hardening neural networks against adversarial examples in the image domain. However, instead of applying defensive distillation to networks for image classification, we examine, for the first time, its performance on text classification tasks and also evaluate its effect on the transferability of adversarial text examples. Our results indicate that defensive distillation only has a minimal impact on text classifying neural networks and does neither help with increasing their robustness against adversarial examples nor prevent the transferability of adversarial examples between neural networks.