Impact of Aliasing on Generalization in Deep Convolutional Networks

TL;DR

This paper examines how aliasing affects the generalization ability of deep convolutional networks and proposes architectural modifications, like low-pass filters, to improve robustness without increasing model complexity.

Contribution

It introduces a simple method of inserting non-trainable low-pass filters to mitigate aliasing, enhancing generalization in CNNs under various conditions.

Findings

Improved generalization on ImageNet-C and Meta-Dataset

Architectural changes do not increase trainable parameters

State-of-the-art results achieved with default hyper-parameters

Abstract

We investigate the impact of aliasing on generalization in Deep Convolutional Networks and show that data augmentation schemes alone are unable to prevent it due to structural limitations in widely used architectures. Drawing insights from frequency analysis theory, we take a closer look at ResNet and EfficientNet architectures and review the trade-off between aliasing and information loss in each of their major components. We show how to mitigate aliasing by inserting non-trainable low-pass filters at key locations, particularly where networks lack the capacity to learn them. These simple architectural changes lead to substantial improvements in generalization on i.i.d. and even more on out-of-distribution conditions, such as image classification under natural corruptions on ImageNet-C [11] and few-shot learning on Meta-Dataset [26]. State-of-the art results are achieved on both datasets without introducing additional trainable parameters and using the default hyper-parameters of open source codebases.