Uncovering Physical Drivers of Dark Matter Halo Structures with Auxiliary-Variable-Guided Generative Models

2026-03-02 · Source: stat.ML updates on arXiv.org · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Data Science & Analytics · Depth: Expert, extended

Summary

Researchers from Argonne National Laboratory and National Renewable Energy Laboratory introduce Disentangled Latent-Conditional Flow Matching (DL-CFM), a novel deep generative model designed to disentangle physical factors in the latent space of thermal Sunyaev–Zel’dovich (tSZ) maps of dark matter halos. This framework extends latent conditional flow matching (LCFM) by incorporating halo mass and concentration as auxiliary variables, using a lightweight alignment penalty to ensure latent dimensions reflect these physical quantities. DL-CFM recovers the established mass-concentration scaling relation and identifies latent space outliers, potentially corresponding to unusual halo formation histories. The model demonstrates high fidelity in generating realistic tSZ halo maps while providing interpretable control over mass and concentration, transforming the latent space into a diagnostic tool for cosmological structure analysis.

Key takeaway

For AI Researchers developing generative models for scientific data, DL-CFM offers a robust approach to enhance interpretability without sacrificing generation quality. You should consider integrating auxiliary-variable guidance and lightweight alignment penalties into your flow-based models to achieve disentangled latent spaces, enabling more precise control over generated samples and facilitating anomaly detection in complex datasets like cosmological simulations.

Key insights

DL-CFM disentangles physical factors in generative models using auxiliary variables, enhancing interpretability and control for scientific data.

Principles

• Auxiliary guidance preserves generative flexibility.
• Disentangled embeddings yield physically meaningful representations.
• Latent space can serve as a diagnostic tool.

Method

DL-CFM integrates an auxiliary-guided VAE encoder with conditional flow matching, using alignment and decorrelation penalties to link latent coordinates to physical variables like halo mass and concentration, then generates high-resolution images.

In practice

• Generate tSZ halo maps with controlled mass and concentration.
• Identify unusual halo formation histories via latent outliers.
• Perform sensitivity analyses with interpretable representations.

Topics

• Generative Models
• Flow Matching
• Disentangled Representations
• Dark Matter Halos
• Cosmological Simulations

Best for: AI Researcher, AI Scientist, Research Scientist

Related on AIssential

• Disentangled autoencoding equivariant diffusion model for controlled generation of 3D molecules — Disentangled semantic embeddings enable fine-grained, multi-property control in 3D molecule generation via equivariant diffusion models.
• Conflict-Aware Additive Guidance for Flow Models under Compositional Rewards — Gradient misalignment causes off-manifold drift when composing multiple constraints in guided sampling for flow models.
• Week Ending 6.21.2026 — Diffusion models' transparency can be significantly improved by mapping latent space information through interpretable token bottlenecks.
• A Unified Latent Space Disentanglement VAE Framework with Robust Disentanglement Effectiveness Evaluation — Robustly evaluate VAE latent space disentanglement and interpretability without ground-truth generative factors using novel assessment tools.
• Physics-Inspired Generative Modeling: Diffusion, Flow Matching, and Energy-Based Models — This article explores physics-inspired generative modeling techniques, including Denoising Diffusion Probabilistic Models (DDPM), Score-Based Diffusion, Flow Matching, and Energy-Based Models (EBMs) w…
• DecQ: Detail-Condensing Queries for Enhanced Reconstruction and Generation in Representation Autoencoders — DecQ uses detail-condensing queries to balance reconstruction and generation in RAEs, improving both with minimal overhead.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by stat.ML updates on arXiv.org.