Revisiting the Platonic Representation Hypothesis: An Aristotelian View

2026-02-16

AI Summary

Existing representational similarity measures are susceptible to confounding effects from model scale, often leading to spurious conclusions about convergence in neural network representations. To address this, this work proposes a permutation-based null-calibration framework that statistically calibrates similarity metrics—including spectral methods and local neighborhood analyses—against appropriate null models. After calibration, the previously reported global representational convergence vanishes; however, strong consistency in local neighborhood structures across models persists. This finding supports the newly formulated “Aristotelian Representation Hypothesis,” which posits that neural network representations converge not in global geometry but in shared local relational structures. The proposed framework establishes a statistically rigorous paradigm for evaluating representational similarity, offering a more reliable foundation for comparative analysis of deep learning models.

Technology Category

• Machine Learning: Representation Learning
• Knowledge Representation and Reasoning: Argumentation
• Computer Vision: Representation Learning for Vision

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Web query analysis, representation and understanding
• User Modeling, Personalization and Recommendation: Metrics for user behavior and evaluating success

Abstract

The Platonic Representation Hypothesis suggests that representations from neural networks are converging to a common statistical model of reality. We show that the existing metrics used to measure representational similarity are confounded by network scale: increasing model depth or width can systematically inflate representational similarity scores. To correct these effects, we introduce a permutation-based null-calibration framework that transforms any representational similarity metric into a calibrated score with statistical guarantees. We revisit the Platonic Representation Hypothesis with our calibration framework, which reveals a nuanced picture: the apparent convergence reported by global spectral measures largely disappears after calibration, while local neighborhood similarity, but not local distances, retains significant agreement across different modalities. Based on these findings, we propose the Aristotelian Representation Hypothesis: representations in neural networks are converging to shared local neighborhood relationships.

Problem

• representational similarity
• Platonic Representation Hypothesis
• network scale
• calibration
• neural representations

Innovation

• representational similarity
• null calibration
• permutation test
• Platonic Representation Hypothesis
• Aristotelian Representation Hypothesis