Dimensions underlying the representational alignment of deep neural networks with humans

TL;DR

This paper introduces a framework to compare human and AI representations by identifying underlying latent dimensions, revealing differences in visual and semantic processing strategies between deep neural networks and humans.

Contribution

The study presents a novel, generalizable method to analyze and compare the underlying representational dimensions of humans and AI, moving beyond scalar measures.

Findings

DNNs have a low-dimensional embedding of visual and semantic features

DNNs predominantly emphasize visual over semantic properties

Significant differences exist between human and DNN image processing strategies

Abstract

Determining the similarities and differences between humans and artificial intelligence (AI) is an important goal both in computational cognitive neuroscience and machine learning, promising a deeper understanding of human cognition and safer, more reliable AI systems. Much previous work comparing representations in humans and AI has relied on global, scalar measures to quantify their alignment. However, without explicit hypotheses, these measures only inform us about the degree of alignment, not the factors that determine it. To address this challenge, we propose a generic framework to compare human and AI representations, based on identifying latent representational dimensions underlying the same behavior in both domains. Applying this framework to humans and a deep neural network (DNN) model of natural images revealed a low-dimensional DNN embedding of both visual and semantic dimensions. In contrast to humans, DNNs exhibited a clear dominance of visual over semantic properties, indicating divergent strategies for representing images. While in-silico experiments showed seemingly consistent interpretability of DNN dimensions, a direct comparison between human and DNN representations revealed substantial differences in how they process images. By making representations directly comparable, our results reveal important challenges for representational alignment and offer a means for improving their comparability.