MATH · IN · MODELS

Euclid Q1 Galaxy Morphology Models (Zoobot ConvNeXt-Nano + self-supervised ViT-S MAE)

STScI / Johns Hopkins / University of Toronto (Wu & Walmsley)

Structures found in this family (1)

Linear Direction

By model (1)

Self-supervised ViT-S MAE (30.1M params, trained on 3M Euclid Q1 galaxy images, 90% masking) · 30.1M
Linear Direction

Observations (1)

SAEs trained on a real Euclid galaxy MAE surface monosemantic morphology directions outperforming raw PCA

Papers

Re-envisioning Euclid Galaxy Morphology: Identifying and Interpreting Features with Sparse Autoencoders (2025)