Comparing the Decision-Making Mechanisms by Transformers and CNNs via Explanation Methods

TL;DR

This paper introduces methodologies to compare the decision-making processes of Transformers and CNNs in visual recognition, revealing differences in compositionality, the impact of normalization, and feature-sharing patterns.

Contribution

It proposes systematic explanation-based methods to analyze and compare the decision mechanisms of different visual backbones, highlighting the role of normalization and feature similarity.

Findings

Transformers and ConvNeXt are more compositional than CNNs.

Batch normalization reduces model compositionality.

Feature-sharing analysis reveals similarities among different backbones.

Abstract

In order to gain insights about the decision-making of different visual recognition backbones, we propose two methodologies, sub-explanation counting and cross-testing, that systematically applies deep explanation algorithms on a dataset-wide basis, and compares the statistics generated from the amount and nature of the explanations. These methodologies reveal the difference among networks in terms of two properties called compositionality and disjunctivism. Transformers and ConvNeXt are found to be more compositional, in the sense that they jointly consider multiple parts of the image in building their decisions, whereas traditional CNNs and distilled transformers are less compositional and more disjunctive, which means that they use multiple diverse but smaller set of parts to achieve a confident prediction. Through further experiments, we pinpointed the choice of normalization to be especially important in the compositionality of a model, in that batch normalization leads to less compositionality while group and layer normalization lead to more. Finally, we also analyze the features shared by different backbones and plot a landscape of different models based on their feature-use similarity.