Deep Neural Networks Can Learn Generalizable Same-Different Visual Relations

TL;DR

This study demonstrates that with appropriate training and architecture choices, deep neural networks, especially pretrained transformers, can learn and generalize the same-different visual relation beyond the training distribution.

Contribution

It shows that certain pretrained transformer models can learn and generalize the same-different relation both within and out-of-distribution, challenging prior assumptions.

Findings

Pretrained transformers achieve near-perfect out-of-distribution generalization.

Fine-tuning on abstract, textureless shapes enhances generalization.

Deep neural networks can learn abstract relations with the right approach.

Abstract

Although deep neural networks can achieve human-level performance on many object recognition benchmarks, prior work suggests that these same models fail to learn simple abstract relations, such as determining whether two objects are the same or different. Much of this prior work focuses on training convolutional neural networks to classify images of two same or two different abstract shapes, testing generalization on within-distribution stimuli. In this article, we comprehensively study whether deep neural networks can acquire and generalize same-different relations both within and out-of-distribution using a variety of architectures, forms of pretraining, and fine-tuning datasets. We find that certain pretrained transformers can learn a same-different relation that generalizes with near perfect accuracy to out-of-distribution stimuli. Furthermore, we find that fine-tuning on abstract shapes that lack texture or color provides the strongest out-of-distribution generalization. Our results suggest that, with the right approach, deep neural networks can learn generalizable same-different visual relations.