Machine learning evaluation demonstrates high classification accuracy across radio galaxies, highlighting the utility of compact variational autoencoder latent representations.
Key Points
Latent space representations learned by a variational autoencoder successfully differentiate radio galaxy types and enable semantic similarity search across disparate datasets.
Simple classifiers achieve accuracy ≥ 76% and roc-auc ≥ 0.86 on MiraBest Confident, matching deep convolutional networks while using compressed latent codes from galaxy images.
Training a very deep variational autoencoder on RGZ DR1 unlabeled data pairs with a masked autoregressive flow density estimator, supporting future anomaly and novelty detection.