Spectral Leakage and Rethinking the Kernel Size in CNNs

TL;DR

This paper investigates spectral leakage in CNNs caused by small kernels and proposes using larger kernels with Hamming window functions to improve accuracy and robustness.

Contribution

It introduces a novel approach of applying window functions to CNN kernels, specifically larger kernels with Hamming windows, to reduce spectral leakage and enhance performance.

Findings

Improved classification accuracy on multiple datasets

Increased robustness against adversarial attacks

Spectral leakage mitigation through larger kernels with window functions

Abstract

Convolutional layers in CNNs implement linear filters which decompose the input into different frequency bands. However, most modern architectures neglect standard principles of filter design when optimizing their model choices regarding the size and shape of the convolutional kernel. In this work, we consider the well-known problem of spectral leakage caused by windowing artifacts in filtering operations in the context of CNNs. We show that the small size of CNN kernels make them susceptible to spectral leakage, which may induce performance-degrading artifacts. To address this issue, we propose the use of larger kernel sizes along with the Hamming window function to alleviate leakage in CNN architectures. We demonstrate improved classification accuracy on multiple benchmark datasets including Fashion-MNIST, CIFAR-10, CIFAR-100 and ImageNet with the simple use of a standard window function in convolutional layers. Finally, we show that CNNs employing the Hamming window display increased robustness against various adversarial attacks.