On Robustness and Transferability of Convolutional Neural Networks

TL;DR

This paper investigates how modern CNNs perform under distributional shifts and transfer learning, revealing that larger models and data, along with simple preprocessing changes, enhance robustness and transferability.

Contribution

It provides a systematic analysis of factors affecting CNN robustness, introduces a synthetic dataset for evaluation, and highlights the impact of data size, model scale, and preprocessing.

Findings

Increasing training data and model size improves robustness.

Simple preprocessing changes can significantly enhance transferability.

A new synthetic dataset enables systematic robustness evaluation.

Abstract

Modern deep convolutional networks (CNNs) are often criticized for not generalizing under distributional shifts. However, several recent breakthroughs in transfer learning suggest that these networks can cope with severe distribution shifts and successfully adapt to new tasks from a few training examples. In this work we study the interplay between out-of-distribution and transfer performance of modern image classification CNNs for the first time and investigate the impact of the pre-training data size, the model scale, and the data preprocessing pipeline. We find that increasing both the training set and model sizes significantly improve the distributional shift robustness. Furthermore, we show that, perhaps surprisingly, simple changes in the preprocessing such as modifying the image resolution can significantly mitigate robustness issues in some cases. Finally, we outline the shortcomings of existing robustness evaluation datasets and introduce a synthetic dataset SI-Score we use for a systematic analysis across factors of variation common in visual data such as object size and position.