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

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Wu, John F., Walmsley, Michael
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915612621537280
author Wu, John F.
Walmsley, Michael
author_facet Wu, John F.
Walmsley, Michael
contents 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.
format Preprint
id arxiv_https___arxiv_org_abs_2510_23749
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Re-envisioning Euclid Galaxy Morphology: Identifying and Interpreting Features with Sparse Autoencoders
Wu, John F.
Walmsley, Michael
Instrumentation and Methods for Astrophysics
Machine Learning
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.
title Re-envisioning Euclid Galaxy Morphology: Identifying and Interpreting Features with Sparse Autoencoders
topic Instrumentation and Methods for Astrophysics
Machine Learning
url https://arxiv.org/abs/2510.23749