Measuring the tendency of CNNs to Learn Surface Statistical Regularities

TL;DR

This paper investigates whether deep CNNs primarily learn surface statistical regularities rather than high-level abstractions, revealing a tendency to rely on dataset-specific Fourier image statistics which affects their generalization capabilities.

Contribution

The study introduces Fourier filtering to create datasets with preserved high-level features but altered surface statistics, providing a method to measure CNNs' reliance on surface regularities.

Findings

CNNs tend to latch onto Fourier image statistics of training data.

Significant generalization gaps up to 28% are observed across test sets.

Increasing network depth has minimal impact on reducing the generalization gap.

Abstract

Deep CNNs are known to exhibit the following peculiarity: on the one hand they generalize extremely well to a test set, while on the other hand they are extremely sensitive to so-called adversarial perturbations. The extreme sensitivity of high performance CNNs to adversarial examples casts serious doubt that these networks are learning high level abstractions in the dataset. We are concerned with the following question: How can a deep CNN that does not learn any high level semantics of the dataset manage to generalize so well? The goal of this article is to measure the tendency of CNNs to learn surface statistical regularities of the dataset. To this end, we use Fourier filtering to construct datasets which share the exact same high level abstractions but exhibit qualitatively different surface statistical regularities. For the SVHN and CIFAR-10 datasets, we present two Fourier filtered variants: a low frequency variant and a randomly filtered variant. Each of the Fourier filtering schemes is tuned to preserve the recognizability of the objects. Our main finding is that CNNs exhibit a tendency to latch onto the Fourier image statistics of the training dataset, sometimes exhibiting up to a 28% generalization gap across the various test sets. Moreover, we observe that significantly increasing the depth of a network has a very marginal impact on closing the aforementioned generalization gap. Thus we provide quantitative evidence supporting the hypothesis that deep CNNs tend to learn surface statistical regularities in the dataset rather than higher-level abstract concepts.