Strong Gravitational Lensing Parameter Estimation with Vision Transformer

TL;DR

This paper investigates the use of Vision Transformer models for parameter estimation in strong gravitational lensing, demonstrating competitive accuracy and efficiency compared to traditional CNN approaches, with implications for future large-scale surveys.

Contribution

First application of Vision Transformer to simulated strong gravitational lensing data, showing promising results and highlighting its potential for future lensing studies.

Findings

ViT achieves competitive results with CNNs in lens parameter estimation.

ViT performs well on key mass-related parameters such as lens center, ellipticities, and slope.

The approach offers a faster alternative to MCMC methods for large datasets.

Abstract

Quantifying the parameters and corresponding uncertainties of hundreds of strongly lensed quasar systems holds the key to resolving one of the most important scientific questions: the Hubble constant ($H_{0}$) tension. The commonly used Markov chain Monte Carlo (MCMC) method has been too time-consuming to achieve this goal, yet recent work has shown that convolution neural networks (CNNs) can be an alternative with seven orders of magnitude improvement in speed. With 31,200 simulated strongly lensed quasar images, we explore the usage of Vision Transformer (ViT) for simulated strong gravitational lensing for the first time. We show that ViT could reach competitive results compared with CNNs, and is specifically good at some lensing parameters, including the most important mass-related parameters such as the center of lens $\theta_{1}$ and $\theta_{2}$, the ellipticities $e_1$ and $e_2$, and the radial power-law slope $\gamma'$. With this promising preliminary result, we believe the ViT (or attention-based) network architecture can be an important tool for strong lensing science for the next generation of surveys. The open source of our code and data is in \url{https://github.com/kuanweih/strong_lensing_vit_resnet}.