![[2602.14486] Revisiting the Platonic Representation Hypothesis: An Aristotelian View](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.14486] Revisiting the Platonic Representation Hypothesis: An Aristotelian View
Summary
This paper revisits the Platonic Representation Hypothesis in neural networks, introducing a new Aristotelian perspective that emphasizes local neighborhood relationships over global convergence.
Why It Matters
Understanding the representational dynamics of neural networks is crucial for advancing AI and machine learning. This study challenges existing metrics and proposes a refined framework, which could lead to more accurate interpretations of neural network behavior and performance.
Key Takeaways
- Existing metrics for representational similarity in neural networks may be confounded by model scale.
- A new permutation-based null-calibration framework is introduced to enhance representational similarity metrics.
- The study reveals that global convergence in representations largely diminishes after calibration.
- Local neighborhood relationships remain significant across different modalities.
- The Aristotelian Representation Hypothesis is proposed as a more accurate framework for understanding neural network representations.
Computer Science > Machine Learning arXiv:2602.14486 (cs) [Submitted on 16 Feb 2026] Title:Revisiting the Platonic Representation Hypothesis: An Aristotelian View Authors:Fabian Gröger, Shuo Wen, Maria Brbić View a PDF of the paper titled Revisiting the Platonic Representation Hypothesis: An Aristotelian View, by Fabian Gr\"oger and 2 other authors View PDF HTML (experimental) 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. Subjects: Machine Learning (cs.LG); Artificial Inte...