# Deep Learning 21cm Lightcones in 3D

**Authors:** Caroline Heneka  
**arXiv:** [2311.17553](https://arxiv.org/abs/2311.17553)  
**Date:** 2023-11-29  
**Summary page:** https://www.summarizepaper.com/en/arxiv-id/2311.17553v1/

## Key points

- Interferometric measurements of the 21cm signal are important in astrophysics
- Deep networks designed for 3D tomographic 21cm light-cones are being used
- The primary objective is to detect and characterize HI sources
- Various architectures, particularly 3D CNN models, have been compared and proven to be accurate and efficient
- Deep networks can effectively analyze and interpret interferometric measurements of the 21cm signal
- Tailored 3D CNN architectures can extract valuable insights regarding global astrophysical and cosmological model parameters
- This research contributes to our understanding of astrophysics by using deep learning techniques
- It opens up new avenues for exploring the universe through data-driven approaches
- Findings serve as a stepping stone for advancements in astrophysics research using machine learning methodologies

## Summary

In the field of astrophysics, interferometric measurements of the 21cm signal have become a significant aspect of the data-driven era. With current and upcoming experiments, researchers are delving into the use of deep networks specifically designed for the structure of 3D tomographic 21cm light-cones. The primary objective is to detect and characterize HI (neutral hydrogen) sources, followed by inferring global astrophysical and cosmological model parameters directly. To achieve this, various architectures have been compared, with a particular focus on 3D CNN (Convolutional Neural Network) models that closely mirror the data structure. These architectures have proven to be the best-performing models in terms of accuracy and efficiency. The paper showcases how deep networks can effectively analyze and interpret interferometric measurements of the 21cm signal. By leveraging tailored 3D CNN architectures, researchers can not only detect and characterize HI sources but also extract valuable insights regarding global astrophysical and cosmological model parameters. This research contributes to our understanding of astrophysics by harnessing the power of deep learning techniques in analyzing complex datasets. It opens up new avenues for exploring the vast universe through advanced data-driven approaches. The findings presented in this paper serve as a stepping stone towards further advancements in astrophysics research using machine learning methodologies. Overall, this study demonstrates how deep learning algorithms applied to interferometric measurements can significantly enhance our ability to unravel the mysteries of the cosmos. By leveraging sophisticated neural network architectures tailored for specific data structures, researchers can unlock crucial information about HI sources and gain deeper insights into fundamental aspects of astrophysics and cosmology.