<ul><li>Machine learning models, including convolutional neural networks, have been effective in classifying variable stars over the last two decades.</li><li>These models require high-quality data and a large number of labelled samples for each star type to generalize well, which can be challenging in time-domain surveys.</li><li>Biases in variable star data can lead to reinforcement of training data biases in models, posing a challenge for validation.</li><li>A new self-regulated training approach utilizing a physics-enhanced variational autoencoder and synthetic samples has shown significant improvements in classifying variable stars on biased datasets.</li></ul>

A self-regulated convolutional neural network for classifying variable stars

Discover more