Photometric Redshifts from SDSS Images with an Interpretable Deep Capsule Network

Published in arXiv, 2021

Dey, B., Andrews, B. H., et al. 2021, arXiv:2112.03939.

We use a deep capsule network to predict photo-z’s for SDSS galaxies from images. Our network produces state-of-the-art results for the fraction of outliers and near SOTA accuracy, while being much more interpretable than traditional convolutional neural networks.

Code | Visualizations | Data