The Limitations of Deep Learning in Adversarial Settings
This work addresses a critical security problem for deep learning systems, particularly in computer vision, by exposing and quantifying their susceptibility to adversarial attacks.
The authors tackled the vulnerability of deep neural networks to adversarial samples by formalizing the adversary space and introducing novel algorithms to craft such samples, achieving a 97% success rate in causing misclassification with only 4.02% average input modification.
Deep learning takes advantage of large datasets and computationally efficient training algorithms to outperform other approaches at various machine learning tasks. However, imperfections in the training phase of deep neural networks make them vulnerable to adversarial samples: inputs crafted by adversaries with the intent of causing deep neural networks to misclassify. In this work, we formalize the space of adversaries against deep neural networks (DNNs) and introduce a novel class of algorithms to craft adversarial samples based on a precise understanding of the mapping between inputs and outputs of DNNs. In an application to computer vision, we show that our algorithms can reliably produce samples correctly classified by human subjects but misclassified in specific targets by a DNN with a 97% adversarial success rate while only modifying on average 4.02% of the input features per sample. We then evaluate the vulnerability of different sample classes to adversarial perturbations by defining a hardness measure. Finally, we describe preliminary work outlining defenses against adversarial samples by defining a predictive measure of distance between a benign input and a target classification.