The Platonic Universe: Do Foundation Models See the Same Sky?

2025/09/23 by UniverseTBD, :, Duraphe, Kshitij +3
#FOS: Computer and information sciences #FOS: Physical sciences #Instrumentation and Methods for Astrophysics (astro-ph.IM) #Machine Learning (cs.LG)

paper · doi:10.48550/arxiv.2509.19453

Abstract

We test the Platonic Representation Hypothesis (PRH) in astronomy by measuring representational convergence across a range of foundation models trained on different data types. Using spectroscopic and imaging observations from JWST, HSC, Legacy Survey, and DESI, we compare representations from vision transformers, self-supervised models, and astronomy-specific architectures via mutual k-nearest neighbour analysis. We observe consistent scaling: representational alignment generally increases with model capacity across our tested architectures, supporting convergence toward a shared representation of galaxy astrophysics. Our results suggest that astronomical foundation models can use pre-trained general-purpose architectures, allowing us to capitalise on the broader machine learning community's already-spent computational investment.

Citations

• SpecCLIP v1.0.0: Aligning and Translating Spectroscopic Measurements for Stars
• Large Language Models -- the Future of Fundamental Physics?
• FALCO: a Foundation model of Astronomical Light Curves for time dOmain astronomy
• AstroLLaVA: towards the unification of astronomical data and natural language
• Euclid Quick Data Release (Q1) Exploring galaxy properties with a multi-modal foundation model
• The Multimodal Universe: Enabling Large-Scale Machine Learning with 100TB of Astronomical Scientific Data
• SpectraFM: Tuning into Stellar Foundation Models
• The Scaling Law in Stellar Light Curves
• AstroPT: Scaling Large Observation Models for Astronomy
• PAPERCLIP: Associating Astronomical Observations and Natural Language with Multi-Modal Models
• AstroLLaMA-Chat: Scaling AstroLLaMA with Conversational and Diverse Datasets
• AstroLLaMA: Towards Specialized Foundation Models in Astronomy
• Radio Galaxy Zoo: Towards building the first multi-purpose foundation model for radio astronomy with self-supervised learning
• Visual Instruction Tuning
• DINOv2: Learning Robust Visual Features without Supervision
• Self-Supervised Learning from Images with a Joint-Embedding Predictive Architecture
• ConvNeXt V2: Co-designing and Scaling ConvNets with Masked Autoencoders
• Astronomia ex machina: a history, primer, and outlook on neural networks in astronomy
• The Astropy Project: Sustaining and Growing a Community-oriented Open-source Project and the Latest Major Release (v5.0) of the Core Package
• Revisiting Model Stitching to Compare Neural Representations
• Emerging Properties in Self-Supervised Vision Transformers
• Emerging Properties in Self-Supervised Vision Transformers
• An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
• Predicting galaxy spectra from images with hybrid convolutional neural networks
• Similarity of Neural Network Representations Revisited
• Overview of the DESI Legacy Imaging Surveys
• The Astropy Project: Building an Open-science Project and Status of the v2.0 Core Package^*
• Convergent Learning: Do different neural networks learn the same representations?
• Understanding image representations by measuring their equivariance and\n equivalence
• Hyper Suprime-Cam: System design and verification of image quality

Related

• Deep Learning in Astrophysics
• Astronomia ex machina: a history, primer, and outlook on neural networks in astronomy
• The Platonic Representation Hypothesis
• Multi-modal Foundation Model for Cosmological Simulation Data
• AstroPT: Scaling Large Observation Models for Astronomy