Objective drives the consistency of representational similarity across datasets

TL;DR

This paper investigates how the choice of training objective influences the consistency of model representations across different datasets, revealing that self-supervised vision models generalize better in their similarity structures.

Contribution

It introduces a systematic framework to measure and analyze how representational similarity varies with datasets and links these similarities to model objectives and task behavior.

Findings

Self-supervised vision models show better generalization of representational similarity across datasets.

Representation similarity is most consistent for models trained on single-domain datasets.

The training objective significantly influences the dataset-dependent consistency of representations.

Abstract

The Platonic Representation Hypothesis claims that recent foundation models are converging to a shared representation space as a function of their downstream task performance, irrespective of the objectives and data modalities used to train these models (Huh et al., 2024). Representational similarity is generally measured for individual datasets and is not necessarily consistent across datasets. Thus, one may wonder whether this convergence of model representations is confounded by the datasets commonly used in machine learning. Here, we propose a systematic way to measure how representational similarity between models varies with the set of stimuli used to construct the representations. We find that the objective function is a crucial factor in determining the consistency of representational similarities across datasets. Specifically, self-supervised vision models learn representations whose relative pairwise similarities generalize better from one dataset to another compared to those of image classification or image-text models. Moreover, the correspondence between representational similarities and the models' task behavior is dataset-dependent, being most strongly pronounced for single-domain datasets. Our work provides a framework for analyzing similarities of model representations across datasets and linking those similarities to differences in task behavior.