Improving Robustness Against Adversarial Example Attacks Using Non-Parametric Models on MNIST

초록

Deep learning research has been actively conducted, and neural networks including CNN have outstanding performance in computer vision. However, parametric models such as neural networks are known to be vulnerable to adversarial example attacks, making them inappropriate to employ when security becomes significant. Hence, non-parametric models are considered but there is a problem of having low accuracy. To solve the issue, we proposed a scheme where images are segmented into patch units for non-parametric models. Experimental results display that the proposed scheme improves both accuracy as well as robustness against adversarial example attacks. © 2020 IEEE.

키워드