Papers
arxiv:2605.20496

Platonic Representations in the Human Brain: Unsupervised Recovery of Universal Geometry

Published on May 19
· Submitted by
Pablo Marcos
on May 22
Authors:
,

Abstract

Self-supervised encoding of brain data reveals shared neural geometry across individuals through geometric transformations without paired data.

AI-generated summary

The Strong Platonic Representation Hypothesis suggests that representational convergence in artificial neural networks can be harnessed constructively: embeddings can be translated across models through a universal latent space without paired data. We ask whether an analogous geometry can be recovered across human brains. Using fMRI data from the Natural Scenes Dataset, we propose a self-supervised encoder that learns subject-specific embeddings from brain data alone by exploiting repeated stimulus presentations. We show that these independently learned spaces can be translated across subjects using unsupervised orthogonal rotations, without paired cross-subject samples or intermediate model representations. Synchronizing pairwise rotations into a single shared latent space further improves cross-subject retrieval, indicating that subject-specific spaces are mutually compatible with a common coordinate system. These results provide evidence for a shared neural geometry in the human visual cortex: subject-specific fMRI representations are approximately isometric across individuals and can be translated through purely geometric transformations.

Community

Paper author Paper submitter

Can the Strong Platonic Representation Hypothesis extend beyond artificial models to human brains?

In this paper, we ask whether independently learned fMRI representations from the visual cortex of different subjects can be translated using geometry alone, without paired cross-subject samples or paired stimulus identities.

Using the Natural Scenes Dataset, we learn subject-specific embeddings from repeated fMRI responses with a self-supervised encoder. We then adapt unsupervised orthogonal translation methods to map embeddings across subjects, and synchronize pairwise rotations into a shared latent space.

The main result is that simple rotations recover accurate instance-level correspondences across subjects, suggesting that visual-cortex representations are approximately isometric across individuals.

Feedback is more than welcome, especially on:

  • how to develop fully unsupervised seed/model-selection criteria for rotations;
  • whether similar geometry should hold for other modalities or tasks;
  • applications to improve cross-subject brain downstream tasks.

Sign up or log in to comment

Get this paper in your agent:

hf papers read 2605.20496
Don't have the latest CLI?
curl -LsSf https://hf.co/cli/install.sh | bash

Models citing this paper 0

No model linking this paper

Cite arxiv.org/abs/2605.20496 in a model README.md to link it from this page.

Datasets citing this paper 0

No dataset linking this paper

Cite arxiv.org/abs/2605.20496 in a dataset README.md to link it from this page.

Spaces citing this paper 0

No Space linking this paper

Cite arxiv.org/abs/2605.20496 in a Space README.md to link it from this page.

Collections including this paper 1