What do neural networks learn in image classification? A frequency shortcut perspective

TL;DR

This paper investigates how neural networks learn frequency-based shortcuts in image classification, revealing that they often rely on texture or shape cues, which can transfer across datasets and are resistant to mitigation strategies.

Contribution

It introduces a metric and method to identify frequency shortcuts in neural networks, expanding understanding of frequency bias in classification tasks.

Findings

Neural networks tend to learn simple solutions based on dominant frequency features.

Frequency shortcuts can be texture-based or shape-based, depending on the task.

Frequency shortcuts transfer across datasets and are not fully mitigated by larger models or data augmentation.

Abstract

Frequency analysis is useful for understanding the mechanisms of representation learning in neural networks (NNs). Most research in this area focuses on the learning dynamics of NNs for regression tasks, while little for classification. This study empirically investigates the latter and expands the understanding of frequency shortcuts. First, we perform experiments on synthetic datasets, designed to have a bias in different frequency bands. Our results demonstrate that NNs tend to find simple solutions for classification, and what they learn first during training depends on the most distinctive frequency characteristics, which can be either low- or high-frequencies. Second, we confirm this phenomenon on natural images. We propose a metric to measure class-wise frequency characteristics and a method to identify frequency shortcuts. The results show that frequency shortcuts can be texture-based or shape-based, depending on what best simplifies the objective. Third, we validate the transferability of frequency shortcuts on out-of-distribution (OOD) test sets. Our results suggest that frequency shortcuts can be transferred across datasets and cannot be fully avoided by larger model capacity and data augmentation. We recommend that future research should focus on effective training schemes mitigating frequency shortcut learning.