Browse Teach Curate

Ollivier-Ricci Curvature Analysis of Neural Representational Alignment

A geometry-aware, embedding-agnostic framework for comparing how neural networks and brains represent stimuli. It treats a representation as a discrete manifold and uses Ollivier-Ricci Curvature and discrete Ricci flow to characterize and align representational geometry, capturing local curvature structure that flat, correlation-based Representational Similarity Analysis misses.

Data Text

Geometry matters: insights from Ollivier Ricci Curvature and Ricci Flow into representational

This paper argues that comparing the internal representations of neural networks and biological brains should account for the geometry of the representational space, not just Euclidean or correlation…