The Limitations of Deep Learning in Adversarial Settings

TL;DR

This paper formalizes the vulnerability of deep neural networks to adversarial samples, introduces algorithms to craft such samples with high success rates, and explores potential defenses, highlighting significant security concerns in deep learning applications.

Contribution

The paper introduces a novel formal framework for understanding adversaries against DNNs and develops algorithms to generate highly effective adversarial samples.

Findings

97% success rate in fooling DNNs with minimal input modifications

Average of 4.02% feature modification per adversarial sample

Vulnerability varies across different sample classes

Abstract

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.