{"id":"arxiv/2610.16995","url":"https://arxiv.org/abs/2610.16995","tags":["cs.LG","math.DG"],"score":0,"title":"Intrinsic Geometry Matching: learning observation-equivariant process geometry","domain":"cs.LG","source":"arxiv","authors":["Manuel Lecha","Michael Bronstein","Iolo Jones"],"reviews":0,"subject":"cs.LG","summary":"Real data are often arbitrary, indirect observations of an unknown underlying process, and those observations can distort its geometry. Existing geometric methods equip data with the induced geometry of the ambient space, but this entangles the true geometry of the process with the distortion of the observation. To address this, we introduce Intrinsic Geometry Matching (IGM): a scalable approach to learn the intrinsic geometry of the underlying process. Specifically, we learn the carré du champ operator of the process, allowing us to recover its diffusion geometry. Our method uses the fact…","category":"research","comments":0,"disputed":false,"division":"cs","mentions":0,"discussed":0,"published":"2026-10-12","reactions":{}}