Main result: simple orthogonal rotations recover accurate instance-level correspondences across subjects.
This suggests that independently learned subject spaces are approximately isometric: different brains encode visual information in geometrically compatible spaces.
Using the Natural Scenes Dataset, we learn subject-specific fMRI embeddings from brain data alone, using repetitions as self-supervision.
Without paired samples or model representations, we learn rotations that translate one subject’s embeddings into another's space.
Feedback, comments, and discussions are very welcome :)
📜 Paper: arxiv.org/abs/2605.204...
⌨️ Code: github.com/memory-forma...
The Strong Platonic Representation Hypothesis suggests independently learned embeddings may be translated using geometry alone, without shared data.
We ask whether this also holds across human brains: do visual-cortex representations form compatible versions of a shared neural geometry?
The Strong Platonic Representation Hypothesis suggests that representational convergence in artificial neural networks can be harnessed constructively: embeddings can be translated across models throu...
Very cool: it's amazing that this doesn't rely on paired data!
@mickbonner.bsky.social and I recently found that the shared geometry is characterized by a ~universal power-law: doi.org/10.1371/jour..., Fig 2
Now I'm curious if we could have done it without the paired stimuli...
Author summary The human cerebral cortex is thought to encode sensory information in population activity patterns, but the statistical structure of these population codes has yet to be characterized. ...
Synchronizing all pairwise rotations into a single shared latent space improves cross-subject translation.
Pairwise translations are not isolated solutions: they are mutually compatible with a common coordinate system.
Can neural representations from different people be translated into a shared space without paired data?
In our preprint, "Platonic Representations in the Human Brain: Unsupervised Recovery of Universal Geometry", with Rishi Jha and @fuentemilla.bsky.social, we test this using fMRI 🧵👇
Pablo Marcos-Manchón
Pablo Marcos-Manchón
Are gist-like shifts in memory over time evidence of engram transformation? Really? Or partly a general response bias?
At #APS2026 in Barcelona this week, Mattia Delmarco will present our work on object typicality and visual memory.
psychologicalscience.confex.com/psychologica...
Episodic memory is reconstructive: recollection combines stored perceptual deta...
