Unveiling the Backbone-Optimizer Coupling Bias in Visual Representation Learning

TL;DR

This paper investigates the bias in the interaction between vision backbones and optimizers, revealing how different architectures are coupled with specific optimizer types, which impacts training and fine-tuning of vision models.

Contribution

It uncovers the backbone-optimizer coupling bias (BOCB), demonstrating its presence across various architectures and optimizers, and provides empirical insights for improving vision model robustness.

Findings

CNNs like VGG and ResNet are coupled with SGD.

Transformers like ViTs are coupled with adaptive optimizers.

BOCB affects pre-training and fine-tuning outcomes.

Abstract

This paper delves into the interplay between vision backbones and optimizers, unvealing an inter-dependent phenomenon termed \textit{\textbf{b}ackbone-\textbf{o}ptimizer \textbf{c}oupling \textbf{b}ias} (BOCB). We observe that canonical CNNs, such as VGG and ResNet, exhibit a marked co-dependency with SGD families, while recent architectures like ViTs and ConvNeXt share a tight coupling with the adaptive learning rate ones. We further show that BOCB can be introduced by both optimizers and certain backbone designs and may significantly impact the pre-training and downstream fine-tuning of vision models. Through in-depth empirical analysis, we summarize takeaways on recommended optimizers and insights into robust vision backbone architectures. We hope this work can inspire the community to question long-held assumptions on backbones and optimizers, stimulate further explorations, and thereby contribute to more robust vision systems. The source code and models are publicly available at https://bocb-ai.github.io/.