AI summary

This paper introduces a new perspective on adversarial examples, arguing that their primary cause is the linear nature of neural networks, supported by quantitative results and a simple method for generating them, and demonstrates that adversarial training can reduce test set error on the MNIST dataset using a maxout network.

Abstract

Several machine learning models, including neural networks, consistently mis- classify adversarial examples-inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed in- put results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to ad- versarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Us- ing this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset.