Intriguing properties of neural networks

TL;DR

This paper explores two surprising properties of neural networks: the semantic information resides in the space of high-level features rather than individual units, and small perturbations can cause misclassification, revealing their discontinuous nature.

Contribution

It uncovers that semantic information is distributed across high-level feature spaces and demonstrates the vulnerability of neural networks to imperceptible adversarial perturbations.

Findings

Semantic information resides in the space of high-level units.

Neural networks can be fooled by imperceptible input perturbations.

The same perturbation can mislead different networks trained on different data subsets.

Abstract

Deep neural networks are highly expressive models that have recently achieved state of the art performance on speech and visual recognition tasks. While their expressiveness is the reason they succeed, it also causes them to learn uninterpretable solutions that could have counter-intuitive properties. In this paper we report two such properties. First, we find that there is no distinction between individual high level units and random linear combinations of high level units, according to various methods of unit analysis. It suggests that it is the space, rather than the individual units, that contains of the semantic information in the high layers of neural networks. Second, we find that deep neural networks learn input-output mappings that are fairly discontinuous to a significant extend. We can cause the network to misclassify an image by applying a certain imperceptible perturbation, which is found by maximizing the network's prediction error. In addition, the specific nature of these perturbations is not a random artifact of learning: the same perturbation can cause a different network, that was trained on a different subset of the dataset, to misclassify the same input.