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

Pablo Marcos-Manchón, Rishi Jha, Lluís Fuentemilla

arXiv:2605.20496 · 2026-05-22 공개 · arXiv · PDF

self-supervised-learning latent-space fmri orthogonal-rotations neural-geometry natural-scenes-dataset cross-subject-retrieval isometric-transformations

Abstract

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.

한국어 요약

한 줄 요약

**[뇌-AI 표현 / Platonic Geometry]** Natural Scenes Dataset fMRI에서 self-supervised encoder가 subject별 embedding 학습, paired data·중간 모델 없이 unsupervised orthogonal rotation으로 cross-subject 변환 가능 — 인간 시각피질의 shared neural geometry 증거.

핵심 기여도

핵심 아이디어

인간 시각피질의 neural 표현은 subject별로 학습된 독립 공간임에도 approximately isometric하며, paired data·중간 모델 없이 purely geometric orthogonal transformation으로 변환 가능하다 — 뇌 간 universal Platonic geometry의 unsupervised 증거.

기술적 접근법

주요 결과

의의 및 한계

**의의**: ANN의 Platonic Hypothesis를 뇌 영역에 확장 — universal geometry 가설의 첫 unsupervised 뇌 증거, paired data·중간 모델 없는 cross-subject 변환 가능성으로 BCI·neuroscience 응용 길 제공, AI·뇌과학 융합. **한계**: 시각피질 중심으로 다른 인지 영역 일반화 추가 검증, NSD 단일 데이터셋 의존, fMRI의 시공간 해상도 한계로 미세 표현 차이 포착 한계, orthogonal rotation 가정의 표현력 제약.

실용적 활용