Update README.md

README.md

@@ -21,7 +21,7 @@ This version is trained on Euclid DR1 (13.6M images, 12M for training). This upd
 
 The base model is a custom `timm` vision transformer; see config.yaml for the exact hyperparameters. These don't matter much; the main things to note are that we need finegrained patches (8x8 pixels here) but the model otherwise need not be large to work well. This version is under 100M parameters and comfortably runs predictions on CPU.
 
-This version is as presented in `Re-envisioning Euclid Galaxy Morphology: Identifying and Interpreting Features with Sparse Autoencoders, Wu & Walmsley, NeurIPS ML4Science workshop 2025` (on arxiv very shortly, bibtex to follow), except that I have removed the angular-scale-dependent positional encoding for simplicity (it turns out to not be necessary for good performance). Please cite the workshop paper if you find our work helpful.
+This version is as presented in the citation below, except that I have removed the angular-scale-dependent positional encoding for simplicity (it turns out to not be necessary for good performance), and we trained on Euclid DR1. Please cite the workshop paper if you find our work helpful.
 
 ```
 @misc{wu2025reenvisioningeuclidgalaxymorphology,