Re-envisioning Euclid Galaxy Morphology: Identifying and Interpreting Features with Sparse Autoencoders

TL;DR

This paper demonstrates that sparse autoencoders can identify interpretable and astrophysically meaningful features in galaxy images, surpassing PCA in alignment with human classifications and revealing new phenomena.

Contribution

It introduces the use of sparse autoencoders for galaxy morphology analysis, highlighting their ability to discover features beyond traditional classification frameworks.

Findings

SAEs outperform PCA in feature interpretability and alignment with Galaxy Zoo labels.

SAEs identify novel features outside existing classification schemes.

The publicly released MAE achieves superhuman image reconstruction performance.

Abstract

Sparse Autoencoders (SAEs) can efficiently identify candidate monosemantic features from pretrained neural networks for galaxy morphology. We demonstrate this on Euclid Q1 images using both supervised (Zoobot) and new self-supervised (MAE) models. Our publicly released MAE achieves superhuman image reconstruction performance. While a Principal Component Analysis (PCA) on the supervised model primarily identifies features already aligned with the Galaxy Zoo decision tree, SAEs can identify interpretable features outside of this framework. SAE features also show stronger alignment than PCA with Galaxy Zoo labels. Although challenges in interpretability remain, SAEs provide a powerful engine for discovering astrophysical phenomena beyond the confines of human-defined classification.