CosmoFlow: Scale-Aware Representation Learning for Cosmology with Flow Matching

2025-07-15

AI Summary

This work addresses the high dimensionality and multiscale nature of cold dark matter (CDM) simulation field data by proposing the first unsupervised flow-matching generative model tailored for cosmological field data, enabling scale-aware latent representation learning. Methodologically, it employs the flow-matching framework to directly model the field distribution without supervision, learning a compact (32× compression), semantically interpretable low-dimensional latent space—where distinct latent channels explicitly encode features at different cosmological scales. Contributions include: (i) the first application of flow matching to cosmological field modeling; (ii) end-to-end disentanglement of multiscale features with physically grounded interpretability; and (iii) support for high-fidelity field reconstruction, physically consistent synthetic data generation, and cosmological parameter inference with sub-percent accuracy.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Computer Vision: Diffusion Models for Vision
• Natural Language Processing: Generation

Application Category

• User Modeling, Personalization and Recommendation: Fairness-aware retrieval and ranking
• Web Mining and Content Analysis: Web data generation and simulation
• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs

Abstract

Generative machine learning models have been demonstrated to be able to learn low dimensional representations of data that preserve information required for downstream tasks. In this work, we demonstrate that flow matching based generative models can learn compact, semantically rich latent representations of field level cold dark matter (CDM) simulation data without supervision. Our model, CosmoFlow, learns representations 32x smaller than the raw field data, usable for field level reconstruction, synthetic data generation, and parameter inference. Our model also learns interpretable representations, in which different latent channels correspond to features at different cosmological scales.

Problem

• Learn compact latent representations of CDM simulation data
• Enable field reconstruction and synthetic data generation
• Identify interpretable features across cosmological scales

Innovation

• Flow matching for unsupervised latent learning
• Compact 32x smaller CDM representations
• Interpretable multi-scale cosmological features