analiza aiqbench
This research applies machine learning techniques to test the "Platonic Representation Hypothesis" within astronomical datasets, determining whether geometric alignment in neural network layers implies scientific interchangeability across different surveys. The study reveals that while models can recover similar spatial geometries for objects, they fail to encode invariant physical properties like redshift, indicating a fundamental distinction between learned geometry and physical reality.
Shared Geometry Is Not Shared Physics: A Layerwise Test of the Platonic Representation Hypothesis in Astronomy
arXiv:2610.04130v1 Announce Type: cross Abstract: In this paper we investigate whether geometrically aligned astronomical representations are also scientifically interchangeable. We analyze every layer's representation from 36 pretrained models using cross-matched optical images, infrared images, and spectra. All 139 analyzed model-survey comparisons show significant local-neighborhood alignment somewhere in the network after accounting for the search over depth. However, alignment does not generally increase with network depth and can be stronger for the changes between consecutive computational blocks than for the block outputs themselves. We find that geometry-selected stages under-perform label-selected stages in every image-transfer comparison; moreover, every final HSC-COSMOS-Web model pair is aligned while every redshift transfer has negative R^2. Models can therefore recover a similar cross-survey geometry among astronomical objects without recovering a survey-invariant linear encoding of the physical properties tested here.
Abstract:In this paper we investigate whether geometrically aligned astronomical representations are also scientifically interchangeable. We analyze every layer's representation from 36 pretrained models using cross-matched optical images, infrared images, and spectra. All 139 analyzed model-survey comparisons show significant local-neighborhood alignment somewhere in the network after accounting for the search over depth. However, alignment does not generally increase with network depth and can be stronger for the changes between consecutive computational blocks than for the block outputs themselves. We find that geometry-selected stages under-perform label-selected stages in every image-transfer comparison; moreover, every final HSC-COSMOS-Web model pair is aligned while every redshift transfer has negative R^2. Models can therefore recover a similar cross-survey geometry among astronomical objects without recovering a survey-invariant linear encoding of the physical properties tested here.
Instrumentation and Methods for Astrophysics (astro-ph.IM); Machine Learning (cs.LG)
arXiv:2610.04130 [astro-ph.IM]
(or arXiv:2610.04130v1 [astro-ph.IM] for this version)
https://doi.org/10.48550/arXiv.2610.04130 Focus to learn more arXiv-issued DOI via DataCite (pending registration)
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Text integral sub licența deschisă a publicației · 227 cuvinte
