# Removing Dust from CMB Observations with Diffusion Models

**Authors:** David Heurtel-Depeiges, Blakesley Burkhart, Ruben Ohana, Bruno Régaldo-Saint Blancard  
**arXiv:** [2310.16285](https://arxiv.org/abs/2310.16285)  
**Date:** 2023-10-25  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2310.16285v1/

## Key points

- Researchers are searching for primordial $B$-modes in cosmic microwave background (CMB) observations.
- Galactic dust foreground poses a challenge in accurately detecting these $B$-modes.
- The authors propose a diffusion-based modeling approach to separate the dust foreground from CMB signals.
- Diffusion models can be trained using examples of dust emission maps to capture the characteristics of the dust foreground.
- Simulations show that common summary statistics accurately recover properties of both components through diffusion-based modeling.
- A model conditioned by specific features and parameters outperforms models trained using a single cosmology for component separation.
- This study has implications for future cosmological inference based on diffusion models.
- Effective component separation techniques can enhance the ability to detect primordial $B$-modes in CMB observations.

## Summary

In the field of cosmology, researchers are actively searching for primordial $B$-modes in cosmic microwave background (CMB) observations. However, the presence of Galactic dust foreground poses a significant challenge in accurately detecting these $B$-modes. To address this issue, the authors of this study propose a diffusion-based modeling approach to better understand and separate the dust foreground from CMB signals. The authors begin by assuming a Gaussian CMB with known cosmology or covariance matrix. They demonstrate that diffusion models can be trained using examples of dust emission maps, allowing their sampling process to align directly with posterior sampling in the context of component separation. This means that the diffusion models can effectively capture and represent the characteristics of the dust foreground. To evaluate the effectiveness of their approach, the authors conduct simulations involving mixtures of dust emission and CMB signals. They find that common summary statistics such as power spectrum and Minkowski functionals accurately recover the properties of both components through this diffusion-based modeling process. This suggests that their method has potential for successfully separating and characterizing the dust foreground in real CMB observations. Furthermore, they introduce a model conditioned by specific features and parameters associated with different cosmological scenarios which outperforms models trained using a single cosmology when it comes to component separation. The findings presented in this study have important implications for future research in cosmological inference based on diffusion models. By refining our understanding of Galactic dust foreground through effective component separation techniques, scientists will be able to enhance their ability to detect primordial $B$-modes in CMB observations. Ultimately, this could provide valuable insights into fundamental questions about the origins and evolution of our universe.