# Mass Estimation of Galaxy Clusters with Deep Learning II: CMB Cluster Lensing

arXiv: https://arxiv.org/abs/2005.13985 · HTML: https://mlsanity.com/paper/2005.13985/

## Key facts

- Cited by 4 papers in the map (data through 14 Oct 2026).
- First posted on arXiv 28 May 2020.

## Built on by (most cited first, 4 of 4)

- [Machine Learning and Cosmology](https://mlsanity.com/paper/2203.08056/) (2022)
- [Reconstructing Patchy Reionization with Deep Learning](https://mlsanity.com/paper/2101.01214/) (2021)
- [DeepSZ: Identification of Sunyaev-Zel'dovich Galaxy Clusters using Deep Learning](https://mlsanity.com/paper/2102.13123/) (2021)
- [Set-based Implicit Likelihood Inference of Galaxy Cluster Mass](https://mlsanity.com/paper/2507.20378/) (2025)

## Abstract (arXiv)

We present a new application of deep learning to reconstruct the cosmic microwave background (CMB) temperature maps from the images of microwave sky, and to use these reconstructed maps to estimate the masses of galaxy clusters. We use a feed-forward deep learning network, mResUNet, for both steps of the analysis. The first deep learning model, mResUNet-I, is trained to reconstruct foreground and noise suppressed CMB maps from a set of simulated images of the microwave sky that include signals from the cosmic microwave background, astrophysical foregrounds like dusty and radio galaxies, instrumental noise as well as the cluster's own thermal Sunyaev Zel'dovich signal. The second deep learning model, mResUNet-II, is trained to estimate cluster masses from the gravitational lensing signature in the reconstructed foreground and noise suppressed CMB maps. For SPTpol-like noise levels, the trained mResUNet-II model recovers the mass for 104 galaxy cluster samples with a 1-σ uncertainty Delta Mrm 200crm est/Mrm 200crm est = 0.108 and 0.016 for input cluster mass Mrm 200crm true=1014~rm Modot and 8× 1014~rm Modot, respectively. We also test for potential bias on recovered masses, finding that for a set of 105 clusters the estimator recovers Mrm 200crm est = 2.02 × 1014~rm Modot, consistent with the input at 1% level. The 2 σ upper limit on potential bias is at 3.5% level.