Alignment with human representations supports robust few-shot learning

TL;DR

This paper demonstrates that models with representations aligned to humans tend to perform better on few-shot learning, are more robust to attacks and domain shifts, highlighting the importance of human-like representations.

Contribution

It provides an information-theoretic analysis predicting a U-shaped relationship between human alignment and few-shot performance, confirmed through empirical analysis of 491 vision models.

Findings

Highly aligned models perform better on few-shot tasks.

Aligned models show increased robustness to adversarial attacks.

Human-alignment correlates with better generalization and robustness.

Abstract

Should we care whether AI systems have representations of the world that are similar to those of humans? We provide an information-theoretic analysis that suggests that there should be a U-shaped relationship between the degree of representational alignment with humans and performance on few-shot learning tasks. We confirm this prediction empirically, finding such a relationship in an analysis of the performance of 491 computer vision models. We also show that highly-aligned models are more robust to both natural adversarial attacks and domain shifts. Our results suggest that human-alignment is often a sufficient, but not necessary, condition for models to make effective use of limited data, be robust, and generalize well.