An Empirical Study of Aegis
This is an incremental study assessing the robustness of an existing defense framework for neural networks against adversarial attacks.
The paper empirically evaluated the Aegis framework for defending against bit-flipping attacks on neural networks, finding that both its dynamic-exit strategy and robustness training showed drawbacks, including accuracy drops on perturbed data and adversarial examples compared to baselines.
Bit flipping attacks are one class of attacks on neural networks with numerous defense mechanisms invented to mitigate its potency. Due to the importance of ensuring the robustness of these defense mechanisms, we perform an empirical study on the Aegis framework. We evaluate the baseline mechanisms of Aegis on low-entropy data (MNIST), and we evaluate a pre-trained model with the mechanisms fine-tuned on MNIST. We also compare the use of data augmentation to the robustness training of Aegis, and how Aegis performs under other adversarial attacks, such as the generation of adversarial examples. We find that both the dynamic-exit strategy and robustness training of Aegis has some drawbacks. In particular, we see drops in accuracy when testing on perturbed data, and on adversarial examples, as compared to baselines. Moreover, we found that the dynamic exit-strategy loses its uniformity when tested on simpler datasets. The code for this project is available on GitHub.